Executive takeaways

Most supply chain resilience work in high-tech manufacturing happens at the Tier 1 level. That’s understandable—Tier 1 relationships are the ones with names attached, contracts negotiated, and quarterly business reviews on the calendar. If you’re a CSCO or head of procurement, it’s also where you have the most leverage and the clearest visibility, so it’s natural that’s where the resilience budget goes.

It’s also, in my experience, not where most of the damage actually happens.

The disruptions that blow up a production schedule tend to originate two or three layers deeper, in the sub-tier suppliers providing memory, storage, and semiconductor allocation that a lot of Tier 1 partners are themselves dependent on and don’t fully control. DRAM and NAND lead times can move from weeks to months with little warning. Controller silicon allocation gets reprioritized industry-wide when a large customer places an outsized order. None of that shows up in a Tier 1 scorecard until it’s already a problem, because by the time it reaches your Tier 1 partner, it’s already someone else’s crisis that’s about to become yours.

I’ve sat with procurement teams who had genuinely sophisticated Tier 1 risk management—dual-sourced assembly, geographic diversification, solid contractual protections—and watched all of it get overrun by a NAND allocation cycle nobody on the team saw coming because nobody was looking that far down the BOM. A fully optimized Tier 1 strategy doesn’t protect you from a problem that starts three tiers below where your visibility ends.

Start by mapping where your BOM is actually vulnerable

The first step isn’t buying more inventory. It’s figuring out where you need to. Most complex bills of materials have a small number of components driving most of the risk—usually memory, storage controllers, or specialty semiconductors with concentrated supplier bases and long, allocation-driven lead times. Everything else on the BOM is comparatively replaceable or multi-sourced already.

The audit that matters here isn’t a generic supplier risk review. It’s component-specific: for each critical part, who actually manufactures it, how concentrated is that manufacturing base, what’s the current allocation environment look like, and what’s your actual exposure if lead times double. I’ve seen procurement teams get this wrong in both directions—either treating the whole BOM as equally risky, which makes the problem unmanageable, or assuming risk is evenly distributed across their Tier 1 base, which misses where it’s actually concentrated. The teams that handle this well narrow down to a manageable list of 5, 10, 15 components that deserve real attention, and treat everything else as lower priority.

Don’t choose between allocation agreements and buffer stock—use both, deliberately

Once you know what’s actually vulnerable, the next question is how to secure it. The instinct is often to pick one lane: either lock in direct allocation agreements with manufacturers, or build a buffer of inventory to ride out disruption. In practice, the more resilient approach usually blends both, and the mix should differ by component.

For components with a small number of qualified suppliers and long design-in cycles, direct allocation agreements are worth the negotiating effort—they buy you priority when a shortage hits, even if they don’t fully protect you from price movement. For components where you have more sourcing flexibility or where allocation agreements aren’t practically available at your volume, a buffer strategy does more work. The mistake is applying one approach uniformly across the BOM instead of matching the strategy to how each component’s supply base actually behaves.

Build the buffer without letting it wreck your balance sheet

This is where a lot of otherwise sound strategies stall, and it’s the piece I see under-discussed relative to how much it actually determines whether a resilience plan gets executed. Building a meaningful buffer of memory, storage, or allocation-constrained silicon means tying up real capital in components sitting in inventory, and depending on how a company is structured, that inventory shows up on the balance sheet in a way finance teams notice immediately—working capital, inventory turns, the ratios a board or lender is watching. A procurement team can have the right component-level risk assessment and still get the buffer plan scaled back or rejected because nobody worked out how to carry it financially.

This is a tension I deal with constantly at Wintec, where we manufacture key hardware components and help clients structure the inventory and financing side of decisions like this. The practical fix isn’t to avoid holding buffer inventory—for allocation-constrained components, that’s often not optional. It’s to structure how that inventory is held so it doesn’t force a tradeoff between component resilience and financial discipline. Companies that get this right treat the financing conversation as part of the resilience plan from the outset, not something they figure out after procurement has already committed to a position.

Design for flexibility, not just supply

The longest-term fix is upstream of all of this: getting sourcing and engineering to qualify alternative component equivalents early in the product lifecycle, before a shortage forces the decision under time pressure. This sounds obvious yet gets skipped constantly, usually because qualifying alternates takes engineering time that feels hard to prioritize when the current component supply looks fine. The teams that avoid the worst of an allocation crisis are usually the ones who did this qualification work 18 months before they needed it, not during the crisis itself.

None of this eliminates sub-tier volatility. Memory, storage, and semiconductor allocation cycles are going to keep moving in ways that are hard to predict, and no amount of Tier 1 optimization changes that. What separates the companies that absorb these disruptions from the ones that get knocked off schedule is whether they built visibility, sourcing flexibility, and financing structure at the sub-tier level before they needed it, rather than scrambling to build all three at once when the next allocation cycle turns.

About the author

David Jeng is CEO of Wintec Industries, a company that provides key hardware components, supply chain logistics, and flexible inventory financing solutions to global technology companies.

Today is Thursday, Sept. 24.

1. U.S. financing plan targets Argentine minerals and energy capacity

The U.S. Export-Import Bank plans to finance as much as $7 billion of Argentine critical-minerals and energy projects, potentially supporting equipment purchases from American manufacturers over the next two years.

Reuters  |  Read the original article

2. Mercuria commits $500 million to U.S. strategic mineral inventories

Mercuria pledged $500 million to Project Vault, an initiative supported by the U.S. Export-Import Bank to build critical-mineral inventories that industrial users could access during supply disruptions or market dislocations.

Reuters  |  Read the original article

3. Major shippers pool demand for 2,500 electric heavy trucks

Microsoft, PepsiCo, Ikea and Red Bull joined a coalition ordering 2,500 battery-electric Class 8 trucks, using aggregated demand and leasing to lower deployment costs across major U.S. freight hubs.

Supply Chain Dive  |  Read the original article

4. Tariff period reshapes U.S. import volumes and sourcing patterns

A Descartes Datamyne analysis found U.S. imports fell 4.5% in the year after the April 2025 tariff announcement as sourcing shifted away from China toward Mexico, Vietnam, Taiwan and Southeast Asia.

FreightWaves  |   Read the original article

5. Amazon links Los Angeles imports directly to eastern fulfillment centers

Amazon launched Standard Ocean Express, routing eligible seller inventory through Los Angeles and direct rail to East Coast fulfillment centers; pricing and comparative transit details were not disclosed in the announcement.

Supply Chain Dive  |   Read the original article

6. Old Dominion schedules 4.9% increase on selected LTL tariffs

Old Dominion Freight Line will apply a 4.9% general rate increase to selected standard LTL, cubic-meter and fuel-related tariffs beginning Oct. 5, with actual changes varying by lane and customer.

Supply Chain Dive  |   Read the original article

7. Uniper locks in long-term synthetic jet-fuel supply

Uniper agreed to buy 40,000 metric tons of synthetic aviation fuel annually for more than a decade from Arcadia eFuels’ planned Danish project, with initial deliveries expected in the early 2030s.

Uniper newsroom  |  Read the original article

8. Qualcomm adds MoveIt steward to robotics software portfolio

Qualcomm agreed to acquire PickNik Robotics, steward of the open-source MoveIt framework, seeking closer integration with its Dragonwing platforms while promising continued community governance and support for third-party hardware.

Robotics 24/7  |   Read the original article

Executive takeaways

Editor's Note: The SCM thesis Predicting Truck Parking Availability to Improve Driver Safety and Delivery Reliability was authored by Bhavya Budhia and Samantha Clarke and supervised by Dr. Angela Acocella (acocella@mit.edu) and Tim Russell (trussell@mit.edu). For more information on this research, please contact the thesis supervisor.

The clock is running out, but there’s nowhere to stop

Imagine you are a truck driver. You have been on the road for several hours and only have an hour left on your hours-of-service (HOS) clock, meaning you need to stop soon and rest. You are still hundreds of miles from your destination and have no reliable way of knowing whether the next truck stop you come across will have enough parking for you. Stopping early wastes valuable driving time. Cutting it close risks an HOS violation or unsafe parking decision. This is the difficult reality truck drivers face every day.

In the U.S., truck drivers are essential for keeping supply chains moving. In 2024, nearly 75% of U.S. domestic freight by weight moved by truck.

Despite their critical role, drivers face a pressing daily challenge: finding safe and adequate parking. Recent estimates find there is just 1 truck parking space available for every 11 drivers on the road, and 90% of drivers report struggling to find parking at night.

When safe parking is unavailable, especially at the end of a long workday, drivers face a difficult choice: park in an unsafe area, park illegally, or risk violating HOS regulations. All three outcomes have tangible safety and financial consequences for drivers and the supply chains they serve.

Making every mile count and every stop matter

Why not just build more truck stops? The median cost to construct a single truck parking spot is nearly $94K, meaning this option is highly expensive and time-consuming.

To address this issue, we develop a driver tool that takes in a driver’s location, destination, remaining HOS drive time, and personal preferences to reliably rank truck stops along the route. The tool recommends when and where the driver should park to best utilize their driving time.

We consider several distinct factors to rank truck stops and answer these questions for drivers:

1. Will there be parking when I arrive?
2. Can I legally reach this stop within my remaining HOS?
3. Does this truck stop meet my preferences?

One of the core components is parking availability, predicted using historical data. We also consider on-site amenities at each truck stop, as well as a driver’s personal preference towards those amenities.

To keep drivers on a logical route, the model favors stops that bring them closer to their destination without overshooting.  Historical traffic congestion around each stop is used to create realistic travel time estimates. And, we prioritize truck stops that help drivers to maximize use of their remaining HOS.

We allow drivers to make tradeoffs amongst these priorities, to adjust while on the road, and to make informed decisions.

Rest easy: Putting drivers back in control

The final model offers a user-friendly interface that drivers can consult in real time to create a personalized ranking of truck stops, tailored to their route and their preferences. We deliver these rankings in feasibility frontier maps, which are color-coded, lane-specific views of preferred truck stops that match a driver’s remaining HOS.

Our analysis uncovered several key findings. Corridor structure (i.e., the physical highway network) emerges as a dominant driver of truck stop accessibility. When there is only one main route from origin to destination, we find drivers face higher competition for premium parking locations as well as a greater sensitivity to parking shortages. On the other hand, when they have greater routing flexibility (i.e., multiple route options), drivers have a lower dependence on any single stop.

Additionally, the highest-ranking truck stops remain relatively stable throughout the day, despite changes in traffic congestion. The most preferable truck stops in the network have structural advantages that create ranking resilience, not just temporary swings.

Beyond these insights, our capstone provides drivers with something they do not have today: confidence in where they are going, so they can park safely, rest, and get back on the road.

Today is Wednesday, Sept. 23.

1. FAA outage exposes fragile links in U.S. aviation infrastructure

A severed fiber line and failed switch delayed or canceled about 9,500 flights, prompting federal officials to emphasize redundant telecom routes as they seek another $30 billion for aviation-system modernization.

Reuters  |   Read the original article

2. Glencore secures $1 billion recycled-minerals supply agreement

Nth Cycle signed a $1 billion offtake agreement to supply Glencore with lithium and other materials recovered from batteries, expanding a circular source of critical minerals ahead of the startup's planned listing.

Reuters  |  Read the original article

3. Indian recycler pursues overseas nickel and lithium assets

Lohum is evaluating nickel mines in Indonesia and the Philippines while developing lithium assets in Zimbabwe, aiming to scale battery-material production and reduce Indian manufacturers’ dependence on concentrated foreign supply.

Reuters  |   Read the original article

4. Germany’s uneven factory cycle raises supplier-capacity risk

German machinery production is forecast to decline for a fourth year even as orders improve, increasing the risk that supplier consolidation and capacity cuts could constrain manufacturers when demand eventually recovers.

Logistics Viewpoints  |   Read the original article

5. Procurement leaders confront sticky costs and weaker demand

An ISM expert roundtable warns that persistent input inflation, softer orders and tariff uncertainty are complicating sourcing, inventory and capital decisions, while AI programs require measurable workload reduction and stronger governance.

Institute for Supply Management  |  Read the original article

6. Unified inventory data can bridge wholesale and direct channels

An ISM analysis argues that manufacturers serving B2B, retail and direct customers need real-time inventory records, dynamic allocation and integrated returns to reduce phantom stock, double commitments and trapped working capital.

Institute for Supply Management  |   Read the original article

7. Cognex moves into robotic perception with $500 million deal

Cognex agreed to acquire RealSense for about $500 million, adding 3D depth-sensing technology used in robotic arms, mobile robots and humanoids to its established industrial machine-vision portfolio.

Robotics 24/7  |   Read the original article

8. NVIDIA adds agent-based development tools to robotics platform

NVIDIA released Isaac ROS 5.0 with agent-ready workflows, updated platform support and GPU-accelerated libraries intended to help developers build and deploy robotic perception and manipulation applications more quickly.

Robotics 24/7  |   Read the original article

Executive takeaways

There is no shortage of supply chain conferences where leaders can hear about the technologies expected to shape the future. The more important takeaway from any conference, though, is what to do with that information when they return to work.

Which innovations are mature enough to deploy? What skills will employees need as artificial intelligence and automation take on more operational work? How can organizations move from pilots and isolated successes to meaningful transformation across the supply chain?

With the 2026 NextGen Supply Chain Conference now less than one month away, those are the questions supply chain leaders should consider bringing with them to Nashville.

Taking place Oct. 21-23, 2026, at the W Nashville hotel, NextGen is designed to create candid conversations among executives, practitioners, academics and technology leaders confronting many of the same decisions.

Attendees planning to stay at the W Nashville should also act soon. The deadline to reserve a room at the discounted NextGen conference rate is Wednesday, Sept. 30, at 5 p.m. Eastern. Rooms are subject to availability.

[Click to reserve a room at the W Nashville]

The objective of NextGen is not simply to showcase what might be possible, but to examine what organizations are implementing today, what they have learned on those implementation journeys, and what other leaders can apply within their own operations.

This year’s theme—Innovate. Upskill. Transform.—offers attendees a way to organize those conversations and begin developing their own roadmap for action.

Innovate: Which capabilities can solve a real operational problem?

Innovation is easy to admire from a distance. It becomes much harder when an organization must decide where to invest, how to integrate a new capability and whether it will produce a measurable return.

The NextGen agenda will explore artificial intelligence, robotics, warehouse automation, computer vision, intelligent planning and other emerging capabilities through the experiences of organizations putting them to work.

Dr. Mar Gimeno, associate vice president of U.S. supply chain at Eli Lilly, will examine agentic AI and its potential to change supply chain decision-making. Fanatics will discuss how agentic AI can support demand forecasting, while GE HealthCare will explore how AI investments can improve inventory performance, cash flow and financial outcomes.

Other sessions will look at machine-learning-based carrier risk management, autonomous inventory intelligence, computer vision, warehouse automation and AI-driven execution.

Together, these discussions can help attendees consider a more useful innovation question. Instead of asking, “Where can we use AI or automation?” leaders can ask, “Which recurring decision, bottleneck or operational failure are we trying to improve?”

The experiences of others can help inform your own initiatives and help them become a useful operational capability rather than another disconnected technology experiment.

[Click to view the NextGen agenda]

Upskill: How must the organization change with the technology?

Technology does not transform a supply chain on its own. New tools alter jobs, decisions, workflows and expectations. They can require planners to become orchestrators, supervisors to manage increasingly automated operations and front-line employees to respond to exceptions rather than complete repetitive tasks.

NextGen will devote significant attention to the human side of transformation.

A series of workforce discussions will examine what happens when AI begins running more of the workflow, which capabilities belong in the emerging supply chain skill stack, and whether automation can meaningfully address persistent talent shortages.

Piu Ghosh of Apple will explore how planning, procurement and operations roles are evolving in the age of AI. Mukul Parkhe of DHL Supply Chain will examine the human advantage in automated warehouses and how organizations can provide employees with the workflows, visibility and decision support needed to manage exceptions effectively.

Friday’s panel, “The Future of Supply Chain Talent: Rethinking Education, Training, and Career Pathways,” will broaden that conversation by examining how non-degree credentials, internships, industry-university partnerships and accelerated upskilling programs can help organizations build a workforce capable of keeping pace with technological transformation. Ron Volans of Janssen Pharmaceuticals and Elena Manta of Schneider Electric will join moderator Dan Pellathy of the University of Tennessee for the discussion.

The question of upskilling extends far beyond training people to operate a new system. Leaders must decide how authority changes when machines recommend or initiate actions, which decisions still require human judgment and how employees can develop confidence in tools they may not fully understand.

Those are organizational questions as much as technology questions—and they are becoming increasingly difficult to separate.

[Click to view the NextGen agenda]

Transform: How does an idea become an operational result?

Most companies do not struggle to generate transformation ideas. Where they do struggle is moving those ideas through implementation and scaling them across the business.

That execution challenge runs throughout the NextGen program.

Nitin Kapoor of Wayfair will join SCMR Editor-in-Chief Brian Straight for a keynote fireside chat examining the technology and operating decisions behind Wayfair’s integrated home-delivery network.

Tractor Supply’s Visionary Award keynote will explore how network expansion, fulfillment capabilities, operational scalability and last-mile delivery can help turn supply chain performance into an engine for business growth.

Leaders from DP World, Penske Logistics, Amazon, Target, GXO Logistics, Southern Glazer’s Wine & Spirits, Ryder, BJC HealthCare and other organizations will add perspectives from logistics, retail, healthcare, food and beverage and additional sectors.

The diversity of industries is intentional. Although operating environments differ, many of the obstacles to transformation are shared: fragmented data, competing priorities, integration challenges, workforce resistance, unclear ownership and the difficulty of expanding a successful pilot.

Cross-industry discussions can help attendees see how other organizations have approached those barriers and which lessons may translate into their own operations.

[Click to view the NextGen agenda]

Sponsors help turn the agenda into conversation

The NextGen program is also supported by supply chain technology and service organizations contributing their expertise, customer experiences and resources to the conference.

Zion Solutions Group is the 2026 Diamond Sponsor, while Gather AI is the Platinum Sponsor.

Gold Sponsors include Cycle Labs, Dematic, Geek+, Dexory, The Modern Data Company and Zimark. Verity is a Bronze Sponsor, and AutoScheduler and Argano are Associate Sponsors.

Several sponsors will participate in the Small Group Sessions alongside their customers, giving attendees an opportunity to examine how technologies are being applied in working supply chain environments.

Those sessions are structured around implementation rather than product demonstrations. Presenters will discuss the business challenge, deployment experience, measurable results and lessons learned from the project.

Cycle Labs and Polaris, for example, will examine where AI can improve supply chain testing and where deterministic automation remains necessary. Southern Glazer’s Wine & Spirits and Dematic will discuss the development of a scalable beverage fulfillment network.

Geek+ and Neovia Logistics will share results from a shelf-to-person fulfillment implementation, while Dexory and ODW Logistics will explore the move from manual warehouse audits to continuous autonomous inventory intelligence. Vitti Logistics and Zimark will explain how computer vision can help keep warehouse systems aligned with physical inventory and shipping activity.

The sponsors also help support the meals, networking opportunities, awards program and other elements that allow attendees to continue their conversations outside the formal sessions.

[Click to learn more about the 2026 NextGen sponsors]

Build a personal NextGen agenda

The breadth of the conference allows attendees to build an experience around their own priorities.

Main-stage keynotes and executive panels will provide strategic perspectives, while Small Group Sessions will create opportunities for more detailed conversations. The smaller sessions will examine real implementations involving warehouse automation, healthcare control towers, AI testing, computer vision, inventory intelligence, operational data and workforce development.

Sessions will repeat during morning and afternoon blocks on Thursday, Oct. 22, giving attendees more flexibility to participate in the discussions most closely aligned with their responsibilities.

The smaller format also gives participants an opportunity to move beyond the prepared presentation. They can question the people involved in an implementation, compare experiences with other attendees and discuss the complications that rarely appear in a final case study.

The countdown to Nashville begins

In less than one month, the supply chain community will come together in Nashville to discuss where innovation is delivering value, how work is changing and what it takes to move transformation from ambition to execution.

The answers will not be the same for every company. That is precisely why bringing practitioners from different industries and functions into the same conversation matters.

The goal is not for attendees to leave with someone else’s transformation strategy. It is to leave with better questions, useful relationships and a clearer sense of what their own organizations should do next.

The 2026 NextGen Supply Chain Conference will take place Oct. 21-23, 2026, at the W Nashville. Attendees who require hotel accommodations should reserve their rooms by Wednesday, Sept. 30, at 5 p.m. Eastern to receive the discounted NextGen conference rate, subject to availability.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

Today is Tuesday, Sept. 22.

1. Ineos idles three UK chemical plants as energy costs squeeze production

Ineos will mothball three facilities in Hull that supply chemicals used across medicines, detergents, clothing and construction, citing European gas prices and carbon costs that have weakened regional manufacturing competitiveness.

Reuters  |  Read the original article

2. EnerVenue chooses China’s manufacturing ecosystem over Kentucky

Battery startup EnerVenue opened its first factory in Changzhou after abandoning a proposed Kentucky plant, saying China’s engineering depth, supplier network and costs outweighed available U.S. manufacturing incentives.

Reuters  |  Read the original article

3. Tanker scarcity drives oil-shipping costs above $1 million per day

Tight vessel availability and disrupted Middle East export routes have pushed some very large crude carrier rates above $1 million a day, adding substantial logistics expense for refiners and energy buyers.

The Wall Street Journal  |  Read the original article

4. EU and Philippines outline a deal eliminating most bilateral tariffs

The European Union and the Philippines reached a political agreement on a trade pact expected to remove more than 94% of tariffs, expanding market access for machinery, transport equipment and agricultural products.

Reuters  |   Read the original article

5. UK seeks protection for integrated supply chains under Europe’s sourcing plan

Britain is pressing for participation in the EU’s Made in Europe initiative, warning that excluding UK producers could disrupt tightly connected industrial supply chains while the bloc reduces dependence on China.

Financial Times  |  Read the original article

6. Delayed public purchasing triggers disruption in India’s rice belt

Farmers blocked a major highway in Haryana after government rice procurement failed to begin as expected, highlighting how purchasing calendars and administrative delays can disrupt agricultural flows and pricing.

The Indian Express  |  Read the original article

7. FCC grants three ANSCER mobile robots conditional U.S. approval

The FCC conditionally authorized three ANSCER autonomous mobile robots for continued U.S. import and sale while the manufacturer advances domestic-production and supply chain transparency commitments under new security rules.

Robotics 24/7  |  Read the original article

8. Simbe reports 3,000 retail inventory robots under contract

Simbe says retailers have contracted for more than 3,000 autonomous shelf-intelligence units, signaling broader adoption of robotic inventory sensing while raising the need for independent deployment and performance verification.

Robotics 24/7  |  Read the original article

Executive takeaways

A July 2025 BBC article examined a fight over the proposed Prince William Digital Gateway, a massive data center campus planned near Manassas National Battlefield Park, the site of two major Civil War battles. The development was never built. After courts invalidated the rezonings needed for the project, its developers ultimately withdrew, ending the proposal in July 2026.

Although the Manassas project did not move forward, the controversy surrounding it raised questions that have only become more pressing: How should communities balance growing demand for artificial intelligence and cloud infrastructure against the protection of historic land, natural resources and residents’ quality of life? The episode also demonstrated that decisions about where data centers are placed can have consequences extending far beyond their property lines.

Northern Virginia remains at the center of that discussion. It is the world’s largest concentration of hyperscale data center capacity, according to Synergy Research Group, and continued demand for AI infrastructure is putting additional pressure on available land, electricity and other resources.

Preservationists are concerned about the encroachment of data centers upon historical sites, whether it is overtaking meaningful land, debasing views with ominous buildings, or corrupting the environment due to the extensive resource needs that data centers require, typically water to cool the hot-running servers.

The cloud is not something ethereal, but rather, it is something very tangible—very big, and very much grounded. Data centers—where the “cloud” actually lives—are big buildings that rely on lots of water. There are over 10,000 data centers worldwide, with the most located in the U.S. It is estimated that by 2027, AI-driven data centers could require 1.7 trillion gallons of water globally, which is approximately how much water flows over Niagara Falls in four days. 

Data center development provides construction jobs, and completed data centers provide technology jobs. Let’s not minimize the beneficial economic impacts that data centers bring.  But with the internet available everywhere, do data centers really have to be built near sensitive historical sites? Do they have to be built where competing and limited water resources could be compromised? Are some states just better at bargaining than other states with less concern for the lasting effects and consequences?    

(Oregon as a data center hub? Read why and how they are affecting the small towns they are being built in and near in this interesting article from MSN).

It seems obvious that data centers need to be built near plentiful water resources, but water is a resource that we are also running lean on. There is no doubt that our technology needs are increasing, but this shouldn’t require us to compromise on our commitment to preserving our precious historical sites and some fundamental quality-of-life basics. Data centers can use reclaimed or recycled water, and can utilize recirculating water systems to reduce environmental impact.

Data centers will soon require more power than the electric grid can handle. Nuclear energy has been suggested as the only viable solution, with mini reactors inside each data center or hub of data center buildings. Nuclear reactors that rely on water for cooling. Nuclear reactors that could become targets for terrorist strikes. But these small modular reactors (SMR) have high costs, safety concerns, regulatory hurdles, and are challenged by the fact that the technology has not been proven for deployment at this smaller scale. The practicality, or lack thereof, of SMRs means that they are not a likely solution in the short term compared to the seemingly exponential growth of data centers primarily focused on supporting AI.       

The growth of data centers is not going to slow down, but where they are placed really should be very thoughtfully considered for the short-term and long run.  Because once built, they are here to stay.       

Today is Monday, Sept. 21.

1. Rare-earth magnet flows to the U.S. retreat before leaders meet

Chinese rare-earth magnet shipments to the United States fell to 512 tonnes in August, down 20% from July, renewing supply concerns before this week’s U.S.-China summit.

Financial Times  |   Read the original article

2. Taiwan starts construction on a TSMC-anchored packaging cluster

Taiwan broke ground on an advanced-packaging park in Kaohsiung where TSMC plans validation and training facilities, creating a new hub for equipment and materials suppliers supporting AI-chip production.

Reuters  |  Read the original article

3. India-New Zealand trade agreement sets an October launch

The India-New Zealand free trade agreement takes effect Oct. 20, reducing or eliminating Indian tariffs on about 95% of New Zealand exports while granting Indian goods duty-free access to New Zealand.

Reuters  |   Read the original article

4. Pepco pairs a Polish logistics hub with longer freight contracts

Discount retailer Pepco is building a deconsolidation center near Gdansk and extending ocean-freight commitments, seeking greater routing flexibility and cost control without relying broadly on higher inventory buffers.

Reuters  |   Read the original article

5. Kuehne+Nagel expands its Amazon role into data-center logistics

Kuehne+Nagel signed a long-term Amazon partnership covering data-center equipment deployment, maintenance and upgrades, with Amazon receiving a share-purchase option tied to service delivery and commercial milestones.

The Wall Street Journal  |   Read the original article

6. Loblaw extends automated yard-gate controls across more facilities

Loblaw is expanding EAIGLE’s computer-vision gate automation across additional distribution yards, using automated vehicle recognition and appointment data to manage arrivals and strengthen visibility into transportation operations.

Robotics 24/7   |   Read the original article

7. MISUMI creates a $50 million fund for industrial technology startups

MISUMI Americas launched a corporate venture fund targeting 20 to 30 early-stage hardware, robotics, automation and industrial-AI companies, with initial investments expected to range from $500,000 to $1.5 million.

Robotics 24/7  |  Read the original article

8. ALVEST combines airport autonomy businesses under TLD Robotics

ALVEST has consolidated EasyMile, TractEasy and TLD’s driverless operations into TLD Robotics, creating a single provider of autonomous ground-support vehicles, fleet software and deployment services for airports and industrial sites.

Robotics 24/7  |   Read the original article

Executive takeaways

Artificial intelligence has quickly become central to conversations about the future of supply chain management. Across demand planning, transportation, customer service, and exception management, AI can evaluate more information, speed decision-making, and help operations run more efficiently.

Alfonso Quijano

Still, many organizations are finding it difficult to translate AI investment into meaningful business results. One recent estimate found that 95% of internal AI projects have delivered no measurable return (MIT Media Lab, “The GenAI Divide: State of AI in Business 2025,” July 2025). The obstacle is not simply the technology itself. More often, it is the failure to align technology with people and day-to-day operations.

The companies making the strongest progress are learning that AI creates the most value when it augments human expertise rather than attempts to replace it. A Human + AI model combines intelligent automation with experienced professionals who provide judgment, context, and accountability—turning technology into both a force multiplier and a more dependable operational capability.

Why AI investments fall short

Supply chain leaders are under constant pressure to improve visibility, react faster to disruption, control costs, and protect service levels while confronting labor shortages, volatile demand, and increasingly complex global networks.

Tasks that once consumed hours of manual effort can now be completed in minutes. Today’s systems can analyze enormous datasets, spot patterns, automate repetitive workflows, and surface recommendations in real time. But many organizations have discovered that deploying AI is far easier than embedding it in the business in a way that produces durable value.

In many cases, companies introduce AI before clearly defining the operational problem it is expected to solve. Others add new technology without determining how it will fit into existing workflows or how employees will use its output to make better decisions.

That creates a gap between what the technology can do and what the operation actually needs. AI can produce insights, but insight by itself does not change an outcome. People are still required to understand the business, identify exceptions, and convert recommendations into action.

The continuing value of human judgment

Supply chains do not function in predictable, controlled settings. They operate in an environment where disruption is routine. Weather delays a shipment. A supplier misses a production deadline. A customer changes priorities without warning. Congestion affects a transportation network. AI can analyze these developments quickly, but people must apply logic, experience, and judgment to determine the best response.

AI is highly effective at detecting anomalies and proposing options, yet it cannot fully comprehend every nuance surrounding a decision. It cannot establish trust with customers, negotiate with suppliers, or balance competing priorities across an enterprise. Nor can it replicate the contextual judgment that supply chain professionals develop through years of managing complex operations. This is particularly important in logistics, where relationships and exception management remain major sources of differentiation.

When customers encounter uncertainty, they want more than a system-generated answer. They want confidence from someone who understands their business and can guide them through the problem. That kind of engagement remains inherently human. AI is not eliminating the need for people; it is shifting human effort toward the work where it creates the greatest value.

Where human + AI delivers an advantage

The most effective deployments begin with a clear understanding that people and technology contribute different strengths.

AI is built to process large volumes of data, recognize patterns, monitor workflows, and execute repetitive tasks with speed and consistency. It can operate continuously and provide visibility across networks that are too complex for any person to track manually. People, by contrast, are strongest at judgment, relationship management, creative problem-solving, and navigating ambiguity.

Bringing those capabilities together creates an operating model that is more adaptable and resilient than either humans or AI could be on their own.

Transportation management offers a useful example. AI can track shipments, flag potential delays, and recommend alternative routes. Human operators can then assess those options in light of customer priorities, operational constraints, company policies, and broader business objectives. The combination enables decisions that are both faster and better informed.

The same approach applies to procurement, inventory management, planning, distribution, and customer service. AI takes on the data-intensive work, while people devote more attention to strategic decisions and relationship-driven activities that strengthen competitive advantage.

Designing for human oversight

One of the clearest lessons from early AI adoption is that human experts must remain actively involved. Many organizations initially approached AI as a way to remove people from workflows. Leading companies are increasingly designing a different model—one in which experienced professionals remain in the loop.

That involvement improves the system. Each correction, escalation, and exception provides feedback that helps the technology learn and adapt. Human expertise is not a drag on performance; it makes the system more reliable and provides an essential backstop for leaders who cannot afford to compromise customer relationships or operational outcomes.

Make no mistake about it, AI can make mistakes, too. It may generate an inaccurate recommendation, misread context, or produce an answer that sounds credible but is wrong. Human review helps ensure that consequential decisions are examined before they affect customers, operations, or financial performance. In high-stakes supply chain environments, that safeguard is indispensable.

Turning AI from experiment into capability

The industry’s AI conversation is beginning to move past experimentation. The more useful question now is not whether a company is using AI, but whether that use is creating measurable business value.

Organizations seeing results tend to start with a defined operational challenge, identify where automation can improve productivity, and integrate AI into workflows that preserve a central role for human expertise. They treat AI as an operating capability that must be developed, managed, and improved—not as a standalone technology product that can simply be purchased and switched on.

A practical foundation for the supply chain of the future

For supply chain leaders, this approach creates a significant opportunity. Organizations that combine AI-driven intelligence with experienced talent can improve visibility, respond faster, reduce operational friction, and build more resilient networks. They can also raise employee productivity by reducing time spent on routine work and giving teams more capacity for critical thinking, customer engagement, and higher-value problem-solving.

The objective should not be an autonomous supply chain that removes people from the process. It should be a smarter supply chain in which people and technology work together to achieve results neither could deliver alone.

As adoption accelerates, the strongest performers will be the organizations that use AI to amplify human expertise. The future of supply chain management is not a choice between people and technology. It is Human + AI.

About the Author

Alfonso Quijano is CTO of Lean Solutions Group and co-founder of Lean Tech, where he leads the development of enterprise technology and applied AI solutions for Lean Solutions Group, an operations heavyweight partner in the transportation and logistics industries. He is an experienced technology executive, entrepreneur, and investor focused on innovation, digital transformation, and AI-driven productivity. At Lean Solutions Group, Quijano oversees initiatives spanning AI voice agents, workflow automation, operational intelligence, and logistics technology platforms designed to improve efficiency and scalability. He is also a frequent speaker on artificial intelligence, leadership, and the future of work, with a strong emphasis on combining AI with human expertise to drive measurable business impact.

Executive takeaways

For years, supply chain teams have kept products moving by doing whatever each new challenge demanded. They’ve paid for premium freight, found emergency sources of supply, reallocated inventory at the last minute, and mobilized teams around the next urgent drill.

Each intervention made sense in the moment. Together, however, they have created a much thornier problem: operating model debt.

For supply chain leaders, this debt is accumulating through the short-term workarounds, ad hoc decision-making, and disconnected processes and technologies that define many operations today. Individually, these make-it-work trade-offs feel like appropriate, essential triage. But as they compound, they consume margin and staff bandwidth while making the operating model increasingly difficult to evolve.

The ambition to change this dynamic is clear: Three-quarters of supply chain leaders in a new KPMG survey say they want to execute a comprehensive transformation of their operating model within the next three years. Yet many of the same teams expected to redesign these operations are still consumed by the daily work of holding the current model together.

That tension helps explain why sustained transformation remains so difficult. The survey points to three places where operating model debt is accumulating fastest—and where leaders have the clearest opportunity to begin paying it down: stalled technology execution, a persistent talent gap, and an expanding risk landscape. Together, these chokepoints show how years of triage are constraining modernization—and how leaders can begin breaking the cycle.

Challenge #1: The digital disconnect

At first glance, modern supply chains appear technologically sound. An overwhelming 84% of leaders categorize their supply chain operations as fully digitalized. 

But what does “fully digital” actually mean in practice? For many organizations, running a global network on a fragmented web of disconnected tools and siloed software passes for digital maturity. But the result is operating model debt from investment in tools that have accumulated without the processes and governance needed to make them work together.

That fragmentation complicates execution at scale. Successful pilots stall during enterprise-wide deployment as organizations struggle with AI skills, organizational buy-in, data governance and security, and measuring the return on investment of their automation initiatives. Without an updated operating model, dropping advanced AI tools into disconnected processes just turns bad planning assumptions into bad operational decisions, faster.

Challenge #2: The persistent talent gap

You can’t automate your way out of an outdated operating model if you lack the talent required to integrate, govern, and orchestrate new AI-enabled workflows. Yet 77% of organizations report a pervasive talent gap—a disconnect that weakens supply chain visibility, demand planning, and customer service.

Operating model debt compounds that shortage by keeping the people best positioned to redesign the supply chain occupied with manual reconciliation, exception management, and recurring operational drills. And while AI and automation can help relieve that pressure, making the advanced tech work still depends on the training and expertise needed to integrate, govern, and apply it.

Without that foundation, systems remain disconnected, data lacks governance, and teams default to the workarounds they know. Planners second-guess the tools, operators second-guess the plans, and the business settles into familiar cycles of reactive drills and expedites. It might get the job done for “another crazy day”—but every such day deepens the operating debt.

Challenge #3: The expanding risk landscape

Cost efficiency has long been the supply chain’s mandate. But while cost remains a big focus, risk management has vaulted to the top of leadership’s agenda, cited by 51% of respondents.

The nature of this risk has fundamentally shifted. Leaders must manage more kinds of risk, and at greater depth across the network. Cybersecurity is now their No. 1 concern, followed by multi-tier supplier risks and regulatory concerns. And potential threats extend well past the company’s four walls. Leaders must protect against vulnerabilities that can emerge anywhere in their network—from Tier 1 suppliers, to their suppliers’ suppliers, and beyond.

As supply chains become more connected, operating model debt can amplify the exposure created when digital and supplier risks overlap. A corrupted supplier commit, a manipulated inventory position, or a false logistics signal from an upstream vendor can trigger automated workflows based on bad data, creating inventory misallocation, service failures, and compliance exposure.

Paying down operating model debt

Supply chain leaders can’t pause operations to redesign the operating model. They must create transformation capacity while serving customers and managing daily volatility. Four operational moves can help:

• Fix the workflow before adding the tech: Identify the workarounds and fragmented handoffs that consume the most capacity. Then invest in training, redefine roles, clarify decision rights, and embed physical constraints into the system so new AI tools can enhance execution.
• Connect signals to automated workflows: Stop admiring static dashboards. Establish targeted data pipelines and decision rules that trigger automated action when a physical constraint changes, reducing manual intervention.
• Scale AI strategically: Target one high-friction use case—a volatile logistics lane, constrained supplier network, or high-variance category—as a baseline model. Prove it works in one focused area, then scale the approach across the broader network.
• Expand the defensive perimeter: Bring supply chain, procurement, cybersecurity, and compliance teams into the same operating rhythm, and map deep-tier supplier dependencies before they cascade into the enterprise.

Leaders who recognize their supply chain operating model debt can begin paying it down and create room for transformation without compromising ongoing performance. Organizations that pull ahead will build adaptability into how decisions get made, with faster signals, clearer ownership, and workflows that absorb volatility without adding another layer of debt.

About the author

Chris McCarney is the U.S. Consulting Leader for Supply Chain & Procurement at KPMG. In his role, McCarney works closely with procurement, supply chain, and operations leaders to turn strategic vision into reality. He is also responsible for identifying supply chain and procurement priorities, growth areas, and capabilities to best serve KPMG clients.

Today is Thursday, Sept. 17.

1. Amazon secures long-term backup power for expanding data centers

Amazon signed a long-term agreement for about $2.4 billion of Generac backup generators expected in 2027 and 2028, linking supplier equity incentives to purchases that could eventually total $8 billion.

Reuters  |   Read the original article

2. Reckitt adds $400 million to regionalize U.S. production

Reckitt plans to expand U.S. manufacturing and research over four years, including a larger North Carolina plant intended to make more than 80% of Mucinex products closer to domestic demand.

The Wall Street Journal  |   Read the original article

3. China puts tighter industrial-chain control at the center of manufacturing policy

President Xi Jinping called for a larger advanced-manufacturing sector, greater control over key industrial chains and faster adoption of intelligent production as China pursues technological self-reliance and export competitiveness.

 Reuters  |   Read the original article

4. Macy’s moves AI replenishment forecasting beyond the pilot stage

Macy’s is broadening an artificial-intelligence forecasting overlay for replenishment, aiming to improve product availability and inventory efficiency as the retailer continues a wider supply chain transformation and facility consolidation program.

Supply Chain Dive  |   Read the original article

5. Marine fuel remains costly despite improving availability

Ship operators continue to face historically high bunker prices even as supplies recover at major hubs, sustaining pressure on vessel costs and carrier surcharges after disruptions around the Strait of Hormuz.

FreightWaves  |  Read the original article

6. Packaging machinery investment shifts toward flexibility and regional capacity

PMMI estimates the U.S. packaging machinery market reached $11.7 billion in 2025, with manufacturers emphasizing flexible automation, faster changeovers and North American capacity as labor and demand patterns evolve.

Modern Materials Handling  |  Read the original article

7. GXO and Exotec automate high-volume fashion fulfillment for Guess

GXO deployed 127 Exotec robots and 60,000 storage locations at its Netherlands facility for Guess, with the partners reporting capacity for 40,000 to 70,000 pieces per day.

Robotics 24/7 |  Read the original article

Executive takeaways

Knowing that AI can transform supply chains is one thing. Where to start and how to sequence the journey is most definitely another. For consumer companies where demand is volatile, fulfillment is complex, and margin pressure is relentless, the stakes of getting this wrong (or right) are especially high. Here we look at how the organizations pulling ahead have rewired their supply chains domain by domain.

The “what” and “how”

In McKinsey’s experience working across consumer supply chain transformations, sustainable AI impact requires organizations to address two things simultaneously. On the one hand, the what comprises a clear value roadmap, prioritized by domain, anchored in business outcomes. On the other, the how includes the operating model, data foundation, technology architecture, talent, and adoption practices that make transformation stick. Nailing the what without the how produces pilots that never scale. The how without the what produces capable platforms with no clear application.

Lasting impact requires both. In practice, this means choosing a supply chain domain—not an isolated use case, and not the entire enterprise—as the unit of transformation. A domain-led approach balances end-to-end impact, leadership excitement, and achievable results within a 6-to-12-month window. Domains create data and change management synergies, so that once the data is ready for one solution, subsequent applications become easier. In the same way, once a set of stakeholders has firmly adopted a new way of working, overlapping teams move faster.

For consumer organizations, six domains define the AI transformation agenda, each with a distinct set of use cases.

Six domains of consumer supply chain AI

1. Demand sensing and integrated planning

Demand planning is the highest-maturity, highest-impact starting point for most consumer organizations. AI and machine learning models incorporating hundreds of signals—store-level sales patterns, local trends, promotional calendars, weather, and digital behavior—can materially improve forecast accuracy, particularly for hard-to-predict items such as fresh and perishable products. Leading consumer companies are using AI models that clean and structure error-prone data for fresh items, link consumption forecasts across central commissaries and distribution centers, and integrate directly into store-level replenishment tasking.

Beyond forecasting, this domain includes promotion coordination with supply alignment, new item and seasonal planning (the “cold start” problem), and digital twin models for capacity and inventory policy scenario planning. Autonomous planning, where a closed-loop system continuously adjusts plans without human intervention, represents the frontier of this domain. Among senior consumer supply chain leaders surveyed by McKinsey in February 2026, demand sensing was the most widely scaled AI capability, with 38% of participants reporting deployment at scale (McKinsey RILA Supply Chain Forum, 2026).

2. Inventory availability and replenishment

Effective replenishment has always been the operational heartbeat of consumer supply chains. AI is now transforming it from a rule-based, periodic activity to a dynamic, continuous process. Agentic replenishment systems autonomously adjust store and distribution center replenishment based on real-time inventory signals and dynamic safety stock policies, triggering re-routing of excess supply to stores that need it most. Some consumer companies call this “self-healing inventory.”

Perpetual inventory accuracy, underpinned by AI-powered cycle counting, addresses a persistent root cause of replenishment failures. Agentic inventory allocation and omnichannel order promising has enabled some consumer companies to expand expected delivery date coverage from 69% to 99% of customers (McKinsey internal analysis). For some consumer organizations with fresh or perishable assortments, AI-driven ordering optimization has delivered documented waste reductions of approximately 25% and sales uplifts of approximately 3% in some deployments.

3. Omnichannel fulfillment and reverse logistics

Omnichannel fulfillment is where consumer supply chain complexity is most acute, and where AI delivers some of its most visible customer-facing impact. Dynamic order routing and end-to-end cost-to-serve optimization repositions inventory across fulfillment centers in real time, analyzing products frequently ordered together and adjusting shipment consolidation or splitting, based on requested delivery dates. In some deployments, this has reduced lead times by up to 15% (McKinsey internal analysis).

Delivery-promise accuracy, which provides customers with reliable estimated delivery dates, is a use case that directly drives conversion and reduces post-purchase service contacts. AI-powered last-mile optimization, including dynamic batching and routing for same-day and gig-delivery networks, is maturing rapidly. On the reverse logistics side, AI-vision-based inspection and grading of returns are emerging as meaningful margin recovery opportunities, optimizing decisions regarding restocking, refurbishing, liquidating, or recycling.

4. Transportation execution and network flow

McKinsey’s 2026 Digital Logistics Survey found that average AI transportation use case adoption has increased 2x year-over-year, with 88% of consumer sector adopters reporting that outcomes met or exceeded expectations. Forty-three percent of consumer leaders surveyed in February 2026 reported scaling transportation execution capabilities (McKinsey RILA Supply Chain Forum, 2026).

Key use cases include route planning optimization across linehaul and last mile, dynamic tendering and capacity forecasting, appointment and yard and dock optimization, and carrier performance management. Network design and flow optimization—using digital twin simulation to stress-test network configurations—is increasingly a board-level conversation for consumer companies navigating ongoing trade and geopolitical volatility.

5. DC execution and automation

Distribution center transformation is the domain with the most visible capital investment momentum. McKinsey’s 2026 Digital Logistics Survey found consumer sector projecting automation use case adoption at 58% on average, with shippers expecting an 18-plus percent boost in labor productivity alongside 15-plus percent improvements in picking accuracy and cost per order.

Agentic labor scheduling and end-to-end productivity orchestration, where AI optimizes staffing dynamically against real-time throughput and queue conditions, addresses the talent scarcity and turnover challenge that is among the most pressing operational issues for consumer DCs. Computer vision for receiving, damage detection, and mis-pick identification improves accuracy at the point of inbound and outbound processing. Warehouse control systems with robotics orchestration and dynamic tasking represent the integration layer that makes automation portfolios—conveyors, sortation systems, autonomous mobile robots, and picking arms—work as a coherent system rather than a collection of point solutions.

6. End-to-end visibility, resilience, and control tower

No domain operates in isolation. End-to-end supply chain visibility, with predictive alerts for upstream supply disruptions, is the connective tissue that makes the other five domains work together. Cross-functional orchestration agents that recommend and trigger re-routing, reallocation, expediting, or replanning actions across the supply chain represent the most advanced expression of autonomous supply chain operations.

GenAI-based insights for exception management see natural language interfaces enabling operators to query the supply chain, surface root causes, and receive recommended actions instantly. In McKinsey’s experience, organizations that have deployed cross-functional orchestration capabilities have achieved up to 10 times faster planning cycles and significantly improved responsiveness to disruptions.

Sequencing the journey

Not all six domains are equal starting points. Our experience across consumer supply chain transformations points to a consistent sequencing principle: start where business and leadership engagement are strongest. Leadership buy-in and support are the single most reliable predictors of program success (McKinsey RILA Supply Chain Forum, 2026). Choosing the right entry domain may also be influenced by where the most acute pain is felt, and where data and process foundations are most mature.

From the entry domain, it becomes about building momentum with measurable wins, using the data products and change management muscle developed in domain one to accelerate domain two, and scaling deliberately. Consumer supply chain leaders who rewire both the what and the how, domain by domain, are the ones turning AI ambitions into autonomous operations.

About the authors

Akash Srivastava is an Associate Partner at McKinsey & Company with expertise in AI-enabled supply chain and logistics.

Aniket Joglekar is a Partner at McKinsey & Company focusing on supply chain transformation and digital operations.

The views expressed in this article draw on McKinsey proprietary research, including the McKinsey State of AI 2025 Report, the McKinsey 2026 Digital Logistics Survey, and the McKinsey-facilitated RILA Supply Chain Forum (February 2026).

Today is Wednesday, Sept. 16.

1. SK Hynix weighs first U.S. memory-chip production with Intel

SK Hynix is discussing a possible U.S. manufacturing arrangement with Intel that could localize high-demand memory capacity, although cost, technology-transfer and South Korean policy questions leave the structure and timing unresolved.

Reuters  |    Read the original article

2. South Korea builds a Central Asian critical-minerals network

South Korea and five Central Asian countries signed more than 70 agreements spanning mining, energy, infrastructure and technology, with Seoul seeking partnerships across the critical-minerals chain rather than relying only on raw-material purchases.

Reuters  |   Read the original article

3. HPE locks in memory capacity as AI demand strains supply

HPE is using multiyear supplier agreements, alternative configurations and better availability forecasts to manage persistent memory constraints, but executives say shortages continue to delay converting AI and networking demand into revenue.

Supply Chain Dive  |  Read the original article

4. U.S. container imports reach third-highest monthly level

August container imports rose 3.8% from July to 2.6 million TEUs, even as delays increased across major gateways, signaling resilient demand alongside growing tariff, routing and port-capacity risks.

FreightWaves  |   Read the original article

5. Nissan ties Sunderland hybrid investment to policy changes

Nissan plans to build its Kicks hybrid SUV in Sunderland through a £170 million investment, but says the commitment depends on changes to Britain’s zero-emission vehicle mandate and European sourcing rules.

The Guardian  |   Read the original article

6. McLaren brings more powertrain work and assembly capacity in-house

McLaren’s £500 million U.K. expansion includes another assembly facility and in-house engine and transmission development, signaling deeper vertical integration as the automaker broadens its lineup and targets 1,000 jobs by 2032.

The Guardian  |    Read the original article

7. OnTrac tests capacity-based discounts for incremental parcel volume

OnTrac is piloting a pricing tool that accepts target rates when network conditions allow, aiming to monetize unused capacity while giving parcel shippers another option as carrier fuel surcharges increase.

Supply Chain Dive  |    Read the original article

8. Agility redesigns Digit humanoid for closer work with people

Agility Robotics introduced Digit 5 with a new safety architecture, higher payload and faster charging, positioning the humanoid for broader warehouse and manufacturing workflows when commercial availability begins in 2027.

Robotics 24/7  |  Read the original article

9. Open-source fleet software targets robot interoperability

InOrbit.AI released OpenRobOps under an Apache license, giving robot developers a shared fleet-management foundation with support for an upcoming ISO interoperability standard and multiple edge, cloud and open-source integrations.

Robotics 24/7 |   Read the original article

Executive takeaways

Supply chain leaders do not need another presentation telling them that artificial intelligence, automation and better data will change their operations. They need opportunities to ask the people doing the work what succeeded, what proved difficult and what they would do differently the next time.

That is the idea behind the Small Group Breakout Sessions at the 2026 NextGen Supply Chain Conference, taking place Oct. 21-23 at the W Nashville in Nashville, Tennessee. The sessions bring practitioners, technology providers and attendees together in smaller rooms for focused conversations about real implementations, measurable results and the lessons that do not always make it onto a main-stage slide.

The 2026 conference theme, Innovate. Upskill. Transform., will run through a program that combines executive presentations with practical education and peer discussion. The small-group sessions will be held in 90-minute morning and afternoon blocks on Thursday, Oct. 22. Sessions will repeat in the afternoon, allowing attendees to build a personalized schedule and participate in more of the conversations most relevant to their operations.

Rather than product demonstrations, the sessions are designed around real-world use cases, implementation challenges, business outcomes and candid discussion. This year’s lineup reaches across warehouse execution, inventory accuracy, healthcare supply chains, workforce development, system testing and the expanding role of AI in operational decision-making. Each small-group room will feature three sessions: two customer case studies led jointly by a solution provider and its customer, and one featured speaker session. The sponsored sessions will focus on real implementations, results and lessons learned, while the featured sessions will bring additional practitioner perspectives to the discussion.

Small group sessions

AI Is Not a Test Strategy: Where Supply Chain Leaders Still Need Automation Not Predictions

Speakers: Josh Owen, Cycle Labs, and Casey Brasford, Polaris

The session will examine where AI can improve supply chain testing and where deterministic test automation remains essential to protecting WMS, EDI and fulfillment operations from costly failures.

From Fragmentation to Orchestration: Advancing Healthcare Supply Chain Performance Through Control Tower Integration and Intelligent Planning

Featured Speaker: Omar Devlin, Stanford Medicine

Attendees will hear how Stanford Health Care integrated a digital control tower, intelligent planning and AI-driven automation to improve visibility, procurement, financial performance and alignment with clinical operations.

Modernizing Beverage Distribution: How Southern Glazer’s and Dematic Built a Scalable Fulfillment Network

Speakers: Karli Sage, Southern Glazer’s Wine & Spirits, and Paul Havens, Dematic

The case study will explore the consolidation of two legacy facilities into an automated fulfillment center that doubled throughput, increased picking capacity by 25% and improved delivery efficiency.

Closing the System Reality Gap: How Vitti Logistics Uses Computer Vision to Maintain Inventory and Shipping Accuracy with High-Value Goods

Speakers: Antonio Luna, Vitti Logistics, and Hilla Herzog Manor, Zimark

The discussion will show how computer vision, cameras mounted on material-handling equipment and dock doors, and edge AI are helping Vitti Logistics keep its WMS aligned with physical reality while reducing quality-assurance work.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

From Planner to Orchestrator: The Future of Supply Chain Talent in the Age of AI

Featured Speaker: Piu Ghosh, Apple

The session will explore how planning, procurement and operations roles are changing as intelligent decision-support tools alter the work, skills and responsibilities of supply chain professionals.

From Weeks to Real Time: Inside ODW Logistics’ Journey to Warehouse Intelligence with Dexory

Speakers: Kayla Watson, ODW Logistics, and Todd Boone, Dexory

ODW Logistics will share how it moved from manual warehouse audits to continuous autonomous inventory intelligence, auditing 1 million square feet in less than 24 hours and identifying 90% of discrepancies before they reached customers.

The Human Advantage in Automated Warehouses

Featured Speaker: Mukul Parkhe, DHL Supply Chain

As robotics, WMS, WES and WCS platforms take on more work, this session will look at how organizations can give human operators the workflows, visibility and decision support needed to manage exceptions effectively.

Geekplus x Neovia Logistics: PopPick Lite Shelf to Person Fulfillment

Speakers: Mike Ray, Geekplus America, and Dan Bombrys, Neovia Logistics

The speakers will discuss the implementation of Geek+ PopPick Lite and results that include a 67% improvement in workstation efficiency and a 10-second improvement in speed per tote.

The Two-Year-Old Pallet That Never Existed: What We Found When We Interrogated Our Aged Inventory Number

Featured Speaker: Prasad Sundaramoorthy, Nordstrom

Nordstrom will explore how data and AI helped uncover problems behind a trusted aged-inventory metric and why operational data must be validated before it drives decisions.

In addition, The Modern Data Company along with a customer will present. Session details are still being finalized.

One sponsored small group opportunity remains

One sponsored Small Group Breakout Session remains available for a solution provider to present a 30-minute customer case study with an end-user customer. The format is intended to give attendees a practical account of a real implementation, including the business challenge, deployment experience, measurable results and lessons learned. Companies interested in the remaining opportunity can review the conference sponsorship options at the link below.

View NextGen sponsorship opportunities

Keynotes and executive speakers

The small-group program complements a main-stage agenda featuring supply chain leaders from Wayfair, Tractor Supply, Eli Lilly, Fanatics, Penske Logistics, DP World, Target, GXO Logistics, Amazon, GE HealthCare, RealTruck, Janssen Pharmaceuticals and other organizations.

Nitin Kapoor, vice president of technology at Wayfair, will join SCMR Editor-in-Chief Brian Straight for the keynote fireside chat Building the Future of Home Delivery: Wayfair’s Logistics Evolution. The discussion will explore the technology and operating decisions behind a complex home-delivery network built for speed, reliability and scale.

Craig Ledbetter, senior vice president and chief supply chain officer at Tractor Supply, will deliver the Visionary Award keynote, Supply Chain as a Growth Engine. He will discuss how network expansion, fulfillment capabilities, operational scalability and last-mile delivery can help supply chains shape business growth and customer experience.

Dr. Mar Gimeno, associate vice president of U.S. supply chain at Eli Lilly, will present Agentic AI and the Future of Supply Chain Decision-Making: Lessons from Eli Lilly.

Additional featured sessions include Fanatics on agentic AI and demand forecasting; Penske Logistics on the judgment, execution and collaboration that endure as supply chains change; Amazon on machine-learning-based carrier risk management; GE HealthCare on scaling AI to improve inventory and cash flow; DP World on aligning operating models, process discipline and talent; and a retail panel featuring executives from GXO Logistics, Berry Direct and Target.

Conference sponsors

The 2026 NextGen Supply Chain Conference is supported by organizations helping supply chain leaders turn emerging capabilities into measurable operational results.

• Diamond sponsor: Zion Solutions Group
• Platinum sponsor: Gather AI
• Gold sponsors: Cycle Labs, Dematic, Geek+, Dexory, The Modern Data Company, and Zimark
• Bronze sponsor: Verity
• Associate sponsors: AutoScheduler and Argano

Zion Solutions Group is also sponsoring the 2026 NextGen Supply Chain Awards. Verity will sponsor Thursday breakfast, and Gather AI will sponsor Thursday lunch and networking.

Registration and event information

The 2026 NextGen Supply Chain Conference will be held Oct. 21-23 at the W Nashville. Registration includes access to keynotes, awards presentations, featured sessions, 30 interactive Small Group Sessions across the morning and afternoon blocks, networking receptions, meals and breaks, and Friday programming.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

Executive takeaways

• A single shipment produces two independent records, and they answer different questions. The carrier invoice states what you were billed for. The receiving record states what physically arrived. Each is checked rigorously against something. Neither is routinely checked against the other.
• Professionalizing both functions separately can make the gap more durable, not less. When freight audit and receiving each run a competent, auditable process, both can demonstrate they are doing their job. The comparison between them sits outside both scorecards, so improving either one does not close it.
• Federal claim rules are documentary evidence that the handoff fails. Regulation states explicitly that a shortage noted on a delivery receipt does not, by itself, constitute a claim. Rules are not written to correct mistakes nobody makes.
• The discipline already exists under a different name. Accounts payable will not pay a supplier invoice without matching it to a purchase order and a goods receipt. The freight bill is routinely exempt from the same test, despite describing an event the receiving dock witnessed directly.

Most cost-control conversations in transportation focus on the rate. That is reasonable, because the rate is negotiable, measurable, and owned by someone with a name. Freight audit as a discipline has matured accordingly. A shipper of any size can have carrier invoices checked against contracted rates, accessorial schedules, and fuel tables, either in-house or by a specialist provider, and many do.

That process answers one question well: does this invoice agree with the agreement? It cannot answer a second question: does this invoice agree with what actually arrived?

By design it never attempts to.

Answering that requires a document the freight audit process usually never sees, produced by a function that does not report to transportation, held in a system finance does not query. It is the receiving record, and it is the only independent account of the same event the invoice describes.

2 complete processes with a gap between them

Look at how the work is actually organized in most companies.

Transportation or finance owns invoice review. Its inputs are the invoice and the rate agreement. Its measure of success is pricing accuracy, and against that measure it can perform very well.

Operations or the warehouse owns receiving. Its inputs are the inbound shipment and the packing list or purchase order. Its measure of success is receiving accuracy against what was ordered, and against that measure it too can perform very well.

Both processes are complete on their own terms. Both can pass an audit. And the comparison that would catch a shipment billed at a weight the dock never saw, or a reclassification applied to freight the receiver could have described precisely, or an accessorial charged for a service the dock can confirm did not occur, belongs to neither.

This is the part worth sitting with, because it runs against intuition. Strengthening either function does not close the gap. A better freight audit compares the invoice to the contract more rigorously. A better receiving process compares the delivery to the purchase order more rigorously. Neither improvement brings the two records into contact. In organizations where both functions are mature, each can produce evidence of diligence, which makes the absent third check harder to see rather than easier.

What the regulation tells us

There is a piece of federal regulation that reads, on its face, like an obscure procedural detail, and is in fact a useful piece of evidence about how often this handoff fails.

Under 49 CFR 370.3(c), notations of shortage or damage on freight bills or delivery receipts, along with bad order reports and appraisal reports, are not on their own sufficient to satisfy the minimum requirements for filing a claim. A claim requires a written communication that identifies the shipment, asserts carrier liability, and states a specific or determinable amount of money.

Consider what that rule implies about ordinary practice. The person best positioned to observe a loss, at the moment it is most observable, produces a document that does not by itself preserve the right to recover. Somebody downstream has to convert that observation into a properly formed claim. The regulation exists because that conversion is commonly assumed to have happened when it has not.

The timelines that follow reinforce the point. Under 49 CFR 370.5(a), a carrier must acknowledge a written claim within 30 days unless it has already paid or declined it. Under 49 CFR 370.9(a), it must pay, decline, or make a firm compromise settlement offer within 120 days of receiving the claim, and if the claim is still open at that point, it must explain the delay and update the claimant every 60 days thereafter.

Those are obligations that favor a shipper who is tracking them, and they are only available to a shipper whose exception became a claim in the first place.

The filing deadline itself sits in the bill of lading and runs from delivery. It does not pause while a discrepancy waits to be noticed.

The test finance already applies to everything else

The cleanest way to describe this gap to a finance leader is to stop calling it a freight problem. In accounts payable, the three-way match is unremarkable control: an invoice is not paid until it agrees with the purchase order that authorized the spend and the goods receipt that confirms delivery. No one considers this sophisticated. It is the baseline expectation for supplier invoices.

The freight bill is the invoice that routinely skips it. It is matched to a rate agreement, which is the equivalent of checking a supplier invoice against the price list while never asking whether the goods arrived. The receipt exists. The dock produced it. It simply never reaches the person reviewing the freight charge, because the two documents live on opposite sides of an organizational boundary.

Framed that way, the conversation tends to move faster, because it stops being a transportation topic and becomes a control gap that a CFO recognizes on sight.

Where the ownership question actually lands

The first decision is not which system to buy. It is which function owns the comparison, because in most organizations today the honest answer is that no one does, and every subsequent step fails without it.

That decision is harder than it sounds, and it is worth naming why. Transportation has the expertise to interpret the invoice but not the physical evidence. The warehouse has the physical evidence but neither the rate context nor the claim process. Finance has the control discipline and the least visibility into either. There is a reasonable case for each of them owning it, which is precisely why it tends to be assigned to none of them.

What follows from the decision is comparatively mundane. The receiving record has to reach whoever reviews the freight invoice, in a usable form, which in practice means the piece count, the weight where it was captured, and any exception noted at delivery. A threshold has to be set for what counts as a discrepancy worth pursuing, so the process is not swamped by rounding. A noted exception has to trigger a properly formed claim rather than resting on a signed receipt. And both clocks have to be tracked, the filing deadline and the carrier’s response obligations, because a deadline nobody is watching is a deadline that passes.

None of that requires a large system. It requires somebody’s name against the check.

Why this has stayed invisible

There is no reliable public data on how often the freight invoice and the receiving record disagree, and that is itself part of the explanation.

Freight audit performance is measured and published largely by transportation vendors. Receiving and inventory record accuracy is studied largely by operations and retail researchers. The two bodies of work rarely reference one another, and the reconciliation between them falls outside both. A gap that no one measures is a gap that produces no report, appears on no dashboard, and reaches no quarterly review.

The organizations that close it are not doing anything clever. They have decided that a shipment produces two records, that the two are supposed to agree, and that somebody is accountable when they do not.

About the author

Aaron Brown is the founder of LanePilot Technology, a freight and warehouse platform for small and midsized LTL shippers.

Today is Tuesday, Sept. 15.

1. Panama Canal plans deeper traffic cuts as drought tightens capacity

The Panama Canal will reduce daily vessel transits to 29.5 in October as El Niño-driven drought lowers Gatún Lake, extending waits and adding another capacity constraint for Asia–U.S. and interregional shipping.

The Guardian  |   Read the original article

2. Manufacturers confront a renewed surge in supply chain costs

Rising energy, tariff, materials and freight expenses are pressuring U.S. manufacturers, while AI-driven electronics demand is lengthening component lead times and policy uncertainty is discouraging investment in additional capacity.

Financial Times  |   Read the original article

3. Einride and Lidl put a cabless truck into regular German service

Einride and Lidl began operating a Level 4 autonomous electric truck without a cab on a public-road route between a distribution center and store after receiving approval from Germany's transport authority.

Reuters  |  Read the original article

4. United Kingdom moves to preserve domestic specialty-steel capacity

The British government plans to acquire Speciality Steel UK after a private sale failed, seeking to protect more than 1,300 jobs and retain production capability across South Yorkshire and the West Midlands.

The Guardian  |    Read the original article

5. Asian LNG buyers prepare for demand recovery after supply disruption

Industry executives expect Chinese and Indian LNG demand to rebound when Middle East hostilities ease, after high prices and disrupted Hormuz flows forced buyers to conserve gas and seek alternative supplies.

Reuters  |   Read the original article

6. Universal Robots introduces an AI-ready industrial automation platform

Universal Robots unveiled its Gen 7 platform at IMTS, combining new collaborative arms, force-torque sensing, a redesigned controller and updated software intended to support industrial automation and physical-AI applications.

Robotics 24/7 |  Read the original article

7. Safety expo highlights connected tools for frontline risk reduction

The opening day of the National Safety Council expo showcased connected wearables, personal protective equipment, gas monitoring and safety software among offerings from more than 1,000 workplace-safety providers.

Work Safety 24/7  |   Read the original article

Executive takeaways

AI infrastructure is often described as a GPU shortage. That description is incomplete.

A deployable AI cluster requires far more than accelerators. It depends on advanced packaging, high-bandwidth memory, servers, racks, networking, storage, liquid-cooling equipment, power capacity, grid interconnection, land, permits, skilled labor, and commissioning. A GPU is not usable capacity until it is integrated into that larger physical system.

This combination of constraints has helped create a fast-growing category of specialized providers often called neoclouds. Companies such as CoreWeave, Nebius, Crusoe, and Lambda offer AI-focused compute capacity to customers that need rapid access to large GPU clusters but may not want to build, finance, and operate all of the underlying infrastructure themselves.

The most useful way to understand this model is not as a new type of cloud provider. It is as an old supply-chain model reappearing in a new industry: outsourced capacity.

The outsourced-capacity model

For decades, electronics companies have used contract manufacturers to scale production without building and operating every factory themselves. Apple may design products and define supplier requirements, but companies such as Foxconn assemble devices. Telecom and enterprise-hardware companies similarly rely on electronics manufacturing services providers such as Flex, Jabil, and Celestica.

The customer keeps control of product direction, demand planning, and commercial strategy. The specialized provider takes responsibility for operating capital-intensive capacity, managing labor, coordinating suppliers, and executing production. Long-term supply commitments give the provider confidence to invest in factories, equipment, and inventory.

Neoclouds serve a related function in AI infrastructure. They are not identical to contract manufacturers: they must operate a continuous cloud service, manage software and networking, maintain high availability, and address workload-utilization risk after infrastructure is deployed. Yet the underlying economics are similar.

A specialized provider builds and operates capital-intensive capacity against contracted customer demand.

CoreWeave’s public disclosures illustrate the model. The company reported that the vast majority of its revenue came from multi-year committed contracts structured on a take-or-pay basis, while it used asset-backed debt to finance infrastructure expansion. Its filings also show the scale of the physical system involved: as of the end of 2024, it operated 32 data centers, more than 250,000 GPUs, and more than 360 MW of active power.

For customers, the attraction is not only access to capital. Large technology companies can finance infrastructure themselves. The value proposition is often speed and specialization: a provider focused on AI infrastructure can concentrate on securing hardware, deploying purpose-built data centers, arranging power, integrating racks and networks, and making capacity available quickly.

In this sense, neoclouds offer an outsourced path to capacity.

AI has several shortages

The phrase “chip shortage” hides important differences among AI-infrastructure constraints. Supply chain leaders should distinguish between at least three categories.

First, some constraints are capacity-lag problems. Advanced packaging, including the processes used to combine high-performance GPUs with high-bandwidth memory, has become a major gating factor for AI systems. These shortages can persist for years because expanding qualified capacity requires specialized tools, technical know-how, capital investment, customer qualification, and time. But they are ultimately responsive to new capacity.

Second, high-bandwidth memory is both a capacity and an allocation challenge. HBM is essential to many advanced AI accelerators, and suppliers have strong incentives to prioritize it over lower-margin memory products. Micron has publicly said its HBM capacity for 2026 is sold out, reflecting the intensity of demand and the strategic importance of memory supply in the AI buildout.

This distinction matters. A capacity-lag shortage can be addressed over time through qualified second sources, supplier investment, multi-year agreements, and patience. A margin-driven allocation problem is different. Suppliers will direct constrained output toward the products, customers, and programs that offer the best economics and strongest long-term strategic value.

Third, AI infrastructure faces local execution constraints. A company can secure GPUs and servers yet still be unable to deploy them because it lacks power, interconnection, cooling infrastructure, site readiness, or qualified installation capacity. For many large AI clusters, the bottleneck is increasingly not the processor itself but the ability to put an entire system into operation.

That is why the relevant supply chain is not a semiconductor supply chain alone. It is a coordinated infrastructure supply chain.

What EMS history suggests

The early stages of outsourced-capacity markets tend to favor the provider. When customers need scarce capacity immediately, they accept long commitments, rigid volume terms, and limited flexibility. The provider can use those commitments to finance additional capacity, creating a cycle in which contracted demand supports the next round of investment.

That dynamic is visible in AI infrastructure today. Customers seeking dedicated GPU capacity may be willing to sign multi-year commitments because the alternative is delay: delayed model training, delayed product launches, or delayed access to compute needed for inference.

Electronics manufacturing history suggests that this leverage may not last indefinitely.

As more capacity enters a market and more providers compete, customers typically gain options. Contract terms shorten. Pricing becomes more competitive. Differentiation shifts away from simply having available capacity and toward execution quality, reliability, supply chain resilience, software performance, customer service, and financing discipline.

The same transition may occur in neocloud infrastructure, but not all constraints will normalize at the same time. GPU and advanced-packaging availability may improve as new capacity ramps. HBM supply may remain strategically constrained longer because it reflects not just physical production limits but also supplier choices about product mix and profitability. Power availability and data-center readiness may be even more geographically uneven.

The implication is simple: AI-infrastructure markets will not loosen all at once.

Three lessons for supply-chain leaders

The neocloud story offers useful lessons outside technology.

1. Classify the shortage before choosing a response. A sudden disruption, a capacity-lag bottleneck, a supplier-allocation decision, and a local deployment constraint require different actions. Safety stock and dual sourcing may help with a disruption. They are less effective when a supplier is rationally prioritizing a higher-margin product category.
2. Treat long-term commitments as strategic instruments. Purchase commitments can secure supply and help a supplier finance new capacity. But buyers should understand what they are trading away. Where possible, agreements should include volume-flexibility bands, price-review mechanisms, performance obligations, and renegotiation triggers tied to changing market conditions.
3. Manage the full deployable-capacity bill of materials. Tracking GPUs alone is not enough. Leaders need visibility into memory, packaging, servers, networking, racks, cooling, power, permitting, logistics, and commissioning. The key metric is not hardware received. It is usable capacity placed into service.

Neoclouds are not simply an AI trend. They are an outsourced-capacity model shaped by a new set of constraints. Their rise shows how scarcity can move capital risk, execution responsibility, and bargaining power across a supply chain.

The AI industry is moving quickly, but the underlying pattern is familiar: when capacity is scarce, specialized providers gain leverage; when capacity expands, customers regain options. The supply-chain leaders who understand which constraints are temporary, which are structural, and which are commercial choices will be better positioned for the next phase of AI infrastructure.

About the author

Rochisshil Varma is a supply chain program manager at Microsoft, where he works on cloud and data-center supply chain buildout. His interests include AI infrastructure, supply chain transformation, analytics, and the operational systems required to deploy cloud infrastructure at scale.

The views expressed are the author’s own and do not represent the views of Microsoft.

Today is Monday, Sept. 14.

1. Saudi pipeline damage threatens a major share of global oil supply

Drone damage shut Saudi Arabia’s East-West pipeline, a critical bypass around the Strait of Hormuz, raising the risk that limited Red Sea export inventories could tighten fuel supply and transportation markets.

The Guardian  |  Read the original article

2. Gemini carriers move four more Asia-Europe services back through Suez

Maersk and Hapag-Lloyd will route four additional Gemini services through the Suez Canal instead of around southern Africa, shortening scheduled transits while leaving future operating decisions dependent on regional security.

Maersk  |  Read the original article

3. United States commits roughly $2 billion to rebuild tungsten reserves

Elmet Group says a subsidiary secured a Defense Logistics Agency contract worth about $2 billion to replenish the National Defense Stockpile, underscoring federal efforts to secure domestic access to critical tungsten materials.

Reuters  |   Read the original article

4. Some manufacturers reconsider China exits as alternative hubs fall short

Companies that shifted sourcing from China are reassessing those moves after encountering infrastructure, labor, quality and cost limitations elsewhere, showing why supplier ecosystems can outweigh nominal tariff advantages in location decisions.

Reuters  |   Read the original article

5. Customs data errors could cost shippers their United States import privileges

Beginning Sept. 18, U.S. Customs and Border Protection can revoke importer-of-record privileges when contact or registration information is inaccurate, leaving shippers responsible for data submitted even when brokers handle their filings.

Supply Chain Dive  |  Read the original article

6. AI growth drives new PFAS capacity and fresh supplier liability questions

Major chemical producers are expanding PFAS capacity for semiconductor fabrication and advanced data-center cooling, prompting environmental groups to warn that AI infrastructure growth could deepen regulatory, remediation and supplier-screening risks.

The Guardian  |   Read the original article

7. Lands End works through inventory backlog after warehouse system rollout

Lands End says it is still clearing inventory and shipment disruption linked to a warehouse-management-system implementation, although executives expect the technology to improve efficiency after the operational recovery is complete.

Supply Chain Dive  |   Read the original article

8. CloudNC raises $20 million to expand AI tools for precision machining

CloudNC plans to use new capital to broaden adoption of its AI-assisted CNC programming software and develop an automated quoting product, targeting two labor-intensive bottlenecks for machine shops and component suppliers.

Robotics 24/7  |  Read the original article

9. Safety congress puts leadership behavior and hazard technology in focus

The National Safety Council’s annual conference brings workplace-safety leaders to Indianapolis for sessions on safety culture, hazardous-energy control, human-centered technology and regulatory priorities, including an address by OSHA leadership.

Work Safety 24/7  |  Read the original article

Executive takeaways

Vendors, consultants and internal transformation teams routinely propose goods-to-person systems, automated storage and retrieval, robotic picking and AI-driven planning tools to operations under throughput pressure. Some of these investments create real value. Others absorb capital and management attention while the underlying operational problems persist untouched, automated but not fixed. Practitioners have a name for the worst outcome: “automating the mess,” a faster and more expensive way of producing the same defects.

The most extensively documented automation program in the industry offers a useful corrective. Amazon’s fulfilment network has grown from the 2012 acquisition of Kiva Systems to a robotic fleet that crossed one million deployed units in 2025, approaching the size of the company’s roughly 1.2-million person workforce. Because Amazon publishes detailed, if self-interested, accounts of what it deployed and in what order, the trajectory reads as a longitudinal case study in automation sequencing and one worth testing against operations with a fraction of Amazon’s scale and capital, drawing here on operational experience in Middle East and North Africa e-commerce fulfilment and in time-definite airfreight cargo. The central claim: the decisive variable in warehouse automation is rarely the technology itself. It is the maturity of the process the technology is inserted into and the order in which capability is built.

Three failure modes explain why so many automation programs disappoint. Equipment gets specified against processes that were never properly measured, so business cases rest on assumed rather than demonstrated baselines. Automation targets the physical movement of goods when the real constraint is informational; flows collide and exceptions overwhelm supervisors because coordination lags behind capacity. Automation gets conceived as monolithic, end-to-end transformation, creating single points of failure in operations that, by design, cannot stop. Amazon’s program avoided all three, in a specific order.

Standardize the unit of work before you automate it

The foundational move behind Kiva’s goods-to-person model was not robotic, it was standardizing the unit of work. The original drive units did not handle heterogeneous products; they moved standardized shelving pods to stationary workers, turning an unbounded picking problem into a bounded transport problem. Every later generation deepened the logic. The Sequoia system, now scaled to hold more than 30 million items at Amazon’s Shreveport mega site, stores products in uniform totes that robots can retrieve and present at ergonomic workstations. Robotic arms capable of handling individual items only became viable after years of constraining the problem this way.

The lesson generalizes past robotics: containerization, unit-load discipline and consistent units-of-work measurement are unglamorous industrial engineering, closer to Frederick Taylor than to robotics, however, they determine whether automation is even specifiable. An operation that cannot express its work in standardized, machine-addressable units cannot know what a robot is worth to it. Most brownfield operations that struggle with automation struggle first here, not with the technology.

Treat orchestration as its own automation frontier

A striking feature of Amazon’s trajectory is how much of the recent gain comes from software rather than machinery. In 2025, alongside the one millionth robot, Amazon introduced DeepFleet, an AI model functioning as a traffic controller for the robotic fleet reported to improve fleet travel efficiency by roughly 10% with no new physical equipment involved. Coordination, in other words, was the binding constraint on an already heavily mechanized system and relieving it was worth a double-digit efficiency gain on its own.

The same pattern holds even more strongly in less automated environments. In manual and semi-automated operations, simply accelerating the feedback loop, giving supervisors timely, attributable visibility of shorts, overages and exceptions, paired with revised standard work at the identified failure points routinely delivers a substantial share of the benefit that business cases attribute to physical automation at a fraction of the cost and lead time. Informational automation deserves appraisal as its own investment category and in most brownfield operations it should be exhausted before material handling capital gets committed.

Go narrow, not monolithic

Amazon’s fleet is not one system but a portfolio of narrow, task-specific machines: one sorts and transfers packages, another lifts heavy cartons into carts, another consolidates and picks items, another moves package carts to outbound docks. Each automates a sub-task whose variability has already been engineered out and each can fail without stopping the building. Even Amazon’s widely publicized humanoid pilots have been confined to narrow, low-criticality tasks like moving empty totes.

Deployment has been incremental across more than a decade, letting each generation get absorbed operationally before the next arrived.

This is the logic of right-sized mechanization: find sub-tasks that are narrow, stable, high-volume and ergonomically poor, automate those decisively and leave judgement-intensive, high-mix work with people who have better information. The literature often frames hybrid human-machine configurations as a transitional state en route to full automation. The evidence here suggests the opposite, that continuously rebalanced hybrid configurations may be the stable optimum in deadline-driven, high-mix operations, not a stepping stone past them.

What doesn’t transfer

Three caveats matter before borrowing any of this. Scale and capital matter: a program built on owning the robot designer, the software stack and hundreds of largely greenfield buildings is not a template a capital-constrained brownfield operation can copy and what transfers is the sequencing discipline, not the shopping list. Self-reported results: Amazon’s own figures are promotional as well as informational, so any operator evaluating vendor claims descended from this ecosystem should demand baselines demonstrated in their own operation rather than extrapolated ones. And workforce dynamics: the collaborative framing of robotics coexists with legitimate questions about job quality and the long-run substitution of labor; any operation borrowing these methods needs to engage its own workforce honestly rather than importing reassuring language.

The failure modes above remain the norm, not the exception, across the wider industry. Operations continue to buy goods-to-person systems before standardizing totes, deploy vision systems onto undocumented processes and commit to monolithic automated-storage schemes whose payback models assume a demand stability e-commerce never provides. A simple stress test is worth applying before any of it: if the process cannot run acceptably in manual mode with good information, automation will institutionalize its weaknesses rather than cure them.

Warehouse automation succeeds or fails long before the equipment arrives. The order that worked here—standardize the unit of work, automate transport around that standard, extend mechanization task by narrow task, then automate the orchestration layer—is the transferable part, more than any individual machine. A 10% fleet efficiency gain from software alone, layered on top of a million deployed robots may be the clearest available demonstration that information flow is not a supporting function of automation. It is a frontier of it. Automate the learning and coordination loop first, the muscle second and only ever at the pace your process standardization has earned.

About the author

Neal McGuckin is a senior operations leader with a background spanning e-commerce fulfilment and time-definite airfreight logistics across the Middle East and North Africa, including prior roles at Amazon MENA and IAG Cargo, and currently at Emirates Flight Catering in Dubai. He is a Doctor of Business Administration candidate at Edinburgh Business School, Heriot-Watt University, researching supply chain risk management.

Today is Friday, Sept. 11.

1. Dell’s AI supply constraints spread beyond chips to racks and cooling

Dell says shortages now extend from memory, storage and processors to rack-level cooling, networking and power components, prompting major customers to order earlier and share forecasts through collaborative planning tools.

Supply Chain Dive  Read the original article

2. Pentagon considers $5 billion loan to strengthen AI infrastructure supply

The Pentagon is reportedly discussing a roughly $5 billion loan to Fluidstack that would support U.S. manufacturing and supply capacity for data-center components, potentially becoming its strategic-capital office’s largest financing.

The Wall Street Journal  |  Read the original article

3. Hyundai targets 80% North American parts sourcing by 2030

Hyundai plans to raise locally sourced content for North American vehicle production from 60% to 80%, seeking greater supply stability, lower logistics costs and reduced exposure to shifting trade rules.

Supply Chain Dive  |  Read the original article

4. Russian strikes hit Ukrainian logistics corridor and agribusiness facility

Drone and artillery attacks struck Pavlohrad, a key military supply route, alongside a Kyiv fuel station and Bunge facility in Dnipro, intensifying risks to Ukraine’s industrial and distribution infrastructure.

Reuters  |  Read the original article

5. Nvidia and Palantir extend AI collaboration into operational supply chains

The companies plan to combine Palantir’s operational data platform with Nvidia models for industry-specific applications spanning manufacturing, retail, pharmaceuticals and agriculture, with Nvidia also intending to deploy the approach internally.

Barron’s   |  Read the original article

6. Alstom rail order promises work across Britain’s manufacturing supply chain

Alstom secured contracts exceeding €1.2 billion to supply and maintain 29 battery-electric trains for TransPennine Express, supporting more than 350 Derby jobs and an estimated 5,500 positions across its U.K. supply chain.

Reuters  |  Read the original article

7. Mundra container operations normalize after empty-yard operators end strike

Operators ended a 13-day dispute over depot-code restrictions at India’s busiest port, allowing container pickups and yard activity to resume after congestion, security and operating-policy disagreements disrupted cargo flows.

Financial Express  |  Read the original article

8. Vecna raises $31 million to expand flexible material-movement automation

Vecna Robotics plans to use new funding to scale deployment teams and develop dock-to-dock automation for case and pallet movement as warehouses seek adaptable alternatives to fixed material-handling infrastructure.

Robotics 24/7  |  Read the original article

9. Vention combines agentic and physical AI for factory automation workflows

Vention will demonstrate a platform that uses natural-language agents to design and program automation while machine-level AI supports perception and motion, illustrating how manufacturers may simplify deployment and troubleshooting.

Robotics 24/7  |  Read the original article

Executive takeaways

In concert with the Departments of Commerce, Homeland Security, Commerce and Treasury, the Trump Administration is creating a bold initiative to hold both domestic and foreign companies to a higher standard on their imports into the United States.

Thomas Cook

 

CBP is raising the bar and tightening up on several key areas to ensure an even playing field in global trade. This will be accomplished by assuring that fraudulent importers will be discovered and held accountable, and that monies due from duties and tariffs are correctly administered and paid into Treasury.

Looking back

Most trade professionals and seasoned customs house brokers have witnessed CBP’s inconsistent enforcement posture over the last 20 to 30 years.  These two groups are also fully aware that there continue to be foreign and domestic companies who import into the United States under fraudulent circumstances. In some cases, this is done by importers who are unaware of the regulations, while in other cases importers are specifically creating strategies to circumvent regulations.

We often see mistakes, errors and even purposely chosen steps taken to arrive at incorrect HTS numbers, improper valuation on invoices and misdeclared countries of origin. Each of these areas will be a focus of CBP in future enforcement initiatives.

Looking ahead

The current administration believes that importers need to be held to a higher standard and accountable for their errors. If it is found that a mistake was made intentionally, then fraud will be investigated and a more stringent penalty will be assessed.

The administration will also seek out companies that create strategies to avoid paying duties and tariffs and target them with a more aggressive approach.

Forced labor issues will also be scrutinized more closely as the administration evaluates forced labor practices in over 100 countries with tighter surveillance and higher penalties. Additionally, the administration is considering placing tariffs on countries with forced labor concerns, hopefully encouraging better behavior on procurement, sourcing and importing.

The administration is creating a new import culture that will require both domestic and foreign companies to become more diligent, extend their reach into the supply chain, and exercise more responsible behavior in import regulatory compliance.

The impact on importers

Many of these regulations have been in place but not administered or practiced by importers responsibly.  The good news is that the increased scrutiny will eliminate the “bad actors” who are committing fraud while supporting those importers who operate in a trade-compliant manner. Doing the right thing is now more than ever in the best interests of the global supply chain.

CBP is hiring additional enforcement officers to increase staff by 30% and the administration has also created a task force within DHS/CBP, Treasury and the Department of Justice to pursue fraudulent cases. This fact should serve as a serious warning for all importers and foreign suppliers.

Going further

President Trump recently issued an executive order outlining importers’ responsibility to map out their import supply chains.

Executive Order 14415, titled “Securing America’s Defense Supply Chains and Ensuring Domestic Acquisition of Critical Materials,” mandates comprehensive supply chain mapping, restricts adversary sourcing, and phases out critical material waivers. Signed on July 20, 2026, the directive targets national security vulnerabilities by tracking components to their raw material origins.

Core Requirements

• Supply chain mapping. Prime contractors and subcontractors must trace components, equipment, software, and raw materials through every tier of production.
• Waiver reductions. Starting Jan. 1, 2027, the Department of Defense will stop issuing nonavailability waivers for critical minerals and parts from foreign entities of concern.
• Risk screening. Contractors must evaluate lower-tier suppliers for financial distress, sole-source dependencies, and foreign ownership or influence.

Compliance and action

• Mitigation plans. Vendors identifying single-point failures or adversary reliance must report risks and submit formal corrective action plans.
• Domestic sourcing. Companies are urged to qualify alternative domestic or allied suppliers ahead of strict enforcement deadlines

This executive order is basically advising importers to dig deeper into their primary and tertiary global suppliers. More detailed information and scrutiny will need to be incorporated into any global sourcing program.

While this increases the onus on many U.S. buyers, it has been a best practice for over 20 years with well-operated global supply chain managers.

Raising the bar

Importers will have to raise the bar of trade compliance management in their inbound global supply chain.

The following steps need to be considered by all foreign and domestic importers into the United States:

1. Establish a culture of trade compliance management within all the stakeholders of the import process: procurement, sourcing, logistics, manufacturing, distribution and finance.

2. Recognize four critical operational stalwarts:

• Due diligence
• Reasonable care
• Supervision & control
• Proactive engagement

These four components are properties of a successful trade compliance program and are the specific expectations of government agencies on how an importer needs to manage trade compliance in their import supply chain.

3. Join the Customs Trade Partnership Against Terrorism (CTPAT), which signals to CBP that you have raised the bar of compliance and security within your business model. It is a voluntary program that offers significant value to your import operations.

4. Develop a point person who will lead the trade compliance program in your company. For small to medium-sized companies, this is likely to be a shared responsibility; for larger companies it is likely to be an independent full-time role.

5. Set up an internal review and audit process for your import process, supported by external independent audits. Areas to review:

• Documentation
• Recordkeeping
• Valuation
• HTS numbers
• Mapping of forced labor with all suppliers and their suppliers
• Assists
• Denied parties lists
• Compliance with free trade agreements
• USA sanctions
• 7501 accuracy
• Correct application of duties, tariffs and other fees. Keep in mind “tiering” of applicable duties/tariffs is occurring
• Import SOP’s
• Stakeholder training

6. Senior management needs to be involved, aware, and must support trade compliance initiatives to ensure internal collaboration, adherence and structure.

7. Develop resources to support knowledge and information availability and trade professionals who can guide you through unchartered waters. Some national options include:

• International Compliance Professionals Association (ICPA)
• American Association of Exporters & Importers (AAEI)
• Council of Supply Chain Management Professionals (CSCMP)
• Institute for Supply Management (ISM)
• Association for Supply Chain Management (ASCM)
• National Customs Brokers and Forwarders Association of America (NCBFAA) Educational Institute
• National Institute for World Trade (NIWT)
• National Association of District Export Councils (NADEC)
• Journal of Commerce (JOC.com)
• Global Trade Magazine (globaltrademag.com)

Summary

The consequences of not paying attention to these new initiatives on import regulations will be fines, penalties and possible loss of import privileges.

Raising the bar of trade compliance on import operations is not difficult, requiring only a strong initiative and potentially some external professional support.

Additionally, we have witnessed numerous situations in which trade compliance management can be not only an internal regulatory control unit but can also assist in providing guidance on strategies that can help reduce risk and cost, business process improvements and in tariff mitigation options.

About the author

Thomas Cook is Managing Director of Blue Tiger International (bluetigerintl.com), an international business consultancy advising on supply chain management, trade compliance, purchasing, trade and disruption management, global business and logistics. Tom was former CEO of American River International in New York and Apex Global Logistics Supply Chain Operation in Los Angeles. He has over 35 years’ experience in assisting companies all over the world manage their import and export operations and is a member of the NY District Export Council and serves on the board of directors of the National Association of District Export Councils (NADEC).  Tom is also the Director of the National Institute of World Trade (niwt.org), a 30-year-old educational and training organization, based in New York.  Both Blue Tiger International and the National Institute of World Trade are strategic partners of the Department of Commerce.  Tom has authored over 20 books on global trade, the latest published in January 2026 entitled “Managing and Mitigating the Impact of Tariffs, Pandemics, and Trade Disruptions in the Global Supply Chain.”  Tom is an Honorably Discharged United States Naval Officer, and Merchant Marine Deck Officer.

Today is Thursday, Sept. 10.

1. Oil clears $100 as Middle East conflict constrains critical shipping flows

Brent settled above $100 as attacks involving Iran and the United States intensified, deepening disruption near the Strait of Hormuz and raising fuel, freight and inflation risks for global supply chains.

Reuters  |  Read the original article

2. EU procurement overhaul would give resilience and regional content greater weight

The European Commission proposed one procurement framework, a shared digital platform and optional European-content preferences, positioning public spending as a lever to reduce strategic dependencies and strengthen supply chain resilience.

Reuters  |  Read the original article

3. Early holiday imports lift Port of Los Angeles to a three-month record

The port processed 2.9 million TEUs from June through August as retailers advanced seasonal shipments to limit exposure to new tariffs, higher marine-fuel costs and potential weather-related delays.

Reuters  |  Read the original article

4. Dow reportedly weighs exit from Saudi chemicals venture amid supply disruption

Dow is considering options for its 35% Sadara stake after conflict disrupted regional petrochemical flows and compounded weak demand, elevated operating costs and global oversupply, according to a Bloomberg report cited by Reuters.

Reuters/Bloomberg News  |   Read the original article

5. U.S. backs Kenyan processing capacity in critical-minerals competition

Washington said it will help Kenya develop domestic mineral processing as bidders pursue the rare-earth and niobium-rich Mrima Hill deposit, linking local value creation with efforts to diversify strategic supply chains.

Reuters  |  Read the original article

6. Italian suppliers seek steep EU duties on Chinese vehicles and components

Industry group Anfia urged an 80% tariff above a proposed import threshold, arguing that rising Chinese vehicle and component penetration threatens Europe's supplier base and long-term manufacturing autonomy.

Reuters  |  Read the original article

7. Mexico’s industrial-property market draws attention as nearshoring and AI expand

BIVA’s chief executive said Mexican real estate trusts could finance industrial parks, logistics assets and infrastructure needed for North American manufacturing growth, although trade-policy, security and energy risks remain.

Reuters  | Read the original article

8. Electrolux workers plan Italian strike over restructuring and plant closure

Italian unions called a one-day strike for September 15 after talks stalled over Electrolux's restructuring plan, adding labor uncertainty to the appliance manufacturer's proposed job reductions and facility closure.

Reuters  |   Read the original article

9. China reportedly raises the bar for humanoid-robotics listings

Chinese regulators are reportedly pressing robotics IPO candidates to demonstrate recurring revenue, narrower losses or meaningful innovation after Unitree's volatile debut, potentially tempering capital flows into the fast-growing automation sector.

Reuters/The Information  Read the original article

Executive takeaways

We spend enormous effort trying to build trust across the supply chain. The joint S&OP offsite where sales and operations finally shake hands over one plan. The shared dashboard everyone agreed to. The “one number” the whole organization signed up to last quarter. And most of us have watched it relax back within a quarter—the handshake fades, the defensive forecast padding returns, the planner quietly starts haircutting sales’ numbers again. We usually blame the people, or the change management, and we schedule another offsite.

We propose a different reading of that pattern, with numbers rather than adjectives. Trust between functions is not an input you install; it is a consequence of the underlying incentive structure. Change the operating model and the reading changes. Leave the model alone, and goodwill fades because the structure is still paying people to behave as before.

That claim is easy to assert and hard to prove inside a real company, where you cannot rerun the same quarter under different incentives and hold everything else constant. So we built a company where you can.

A company in a box

The laboratory is a sales & operations planning process rendered in software: five functional agents—sales, supply chain, manufacturing, procurement and finance—negotiate a monthly plan on top of a deterministic constrained-planning engine. The synthetic company is fully specified—products, bills of material, suppliers, capacities, and customer orders—and decisions become operational consequences: backlog, shortages, overtime, expedite premiums, and working capital. The agents speak through an LLM, but the decisions and economics are deterministic; the language explains the result rather than generating it.

Each agent carries two independent settings, each on a 0-to-100 scale. The first we call locality: it governs whose objective the agent optimizes. At 0, the agent optimizes the end-to-end company; at 100, it optimizes only its own local KPI. The second we call trust: it governs how the agent treats the signals coming from the others. At 100, it takes their forecasts and commitments at face value; at 0, it assumes they are gamed and discounts them accordingly. The two dials are deliberately orthogonal—one is about what a function wants, the other about what it believes—because in real organizations they come apart all the time. A function can be perfectly honest and utterly selfish, or deeply skeptical and entirely aligned.

We then did two things. First, we scored the whole company—EBITDA and working capital—across every combination of those settings applied uniformly, to map the static landscape. Then we let the settings evolve, cycle by cycle, with trust updating from experience: a function that gets burned lowers its trust in the one that burned it; a function that is consistently dealt with straight raises it. That let us watch what happens to trust over four simulated years when the incentive structure is held fixed—the question you can never put to a real company, because you can never hold its incentives still for four years while you watch.

Related Content

 

How I vibe-coded an S&OP app in 30 hours

The gap between what most people believe AI can do today and what it actually can do is enormous. I am not talking about theoretical futures or research papers. I am talking about right now. A planner with deep domain knowledge can build functional business applications in days, not months. I have to say: this genuinely blew my mind.

Read More

 

The point of the laboratory is to remove common explanations. The engine never miscalculates. Information is transparent: every function sees the same statistical baseline, and the planner can quote sales’ historical forecast bias in the meeting. The agents are never tired or political; they follow their settings. Strip away competence, information and personality, and what remains is structure.

The underlying mechanisms are not new. Bullwhip effects, misaligned rewards and the difficulty of coordinating decentralized parties have been studied for decades. What the laboratory adds is quantification and sequence: which mechanisms dominate economically, how they interact with adaptive trust, and in what order interventions work best.

A note on interpretation: this is a laboratory, not an estimate of the average company’s losses. The model is deterministic and calibrated to one synthetic manufacturer. The statistical demand baseline is assumed to be correct, so discounting inflated forecasts is beneficial by construction; inventory, backlog and open orders do not carry from one S&OP cycle to the next; and several behavioral parameters are calibrated rather than estimated from company data. We therefore put more weight on the direction and mechanisms of the findings than on the exact euro amounts. Real organizations are also messier than our five-agent laboratory. A single sales or supply chain function can contain several global, regional and local actors, creating more interfaces where incentives, evidence and trust can diverge.

Locality is the lever; trust is only an amplifier

Start with the static map (Exhibit 1). Move the five functions from end-to-end to purely local objectives, and this company loses about €400,000 of EBITDA a year—roughly 16% of it—and that shift is, by a wide margin, the single largest driver of value anywhere in the map. Trust, by contrast, barely moves the company number on its own. Look down the left-hand edge, where incentives are aligned: whether the organization is fully trusting or fully paranoid, an aligned company earns essentially the same €2.55 million. The profitable column is vertical. Incentives are the axis that matters; trust is nearly irrelevant as a driver of the company’s results.

Where does that €400,000 actually go? Mostly into one behavior, and it is the oldest story in supply chain. A sales function paid on volume and service, and unsure it will get the capacity it needs, does the rational thing: it inflates its forecast to claim a larger share of the allocation. The consensus plan swells, manufacturing builds to it, and the bullwhip that operations researchers have described for decades rolls upstream in the form of orders, inventory, and shortages. In our runs, sales is the single largest value destroyer of the five—not because it is malicious, but because its scorecard rewards precisely the number that breaks the plan. The other functions then face a choice: believe the inflated signal, or discount it. Which is where trust comes in.

That is not what most transformation programs assume. We tend to treat trust as a lever: build enough of it and performance will follow. In our model, trust has almost no independent economic effect when incentives are aligned. Once objectives become local, however, trust becomes an amplifier: believing distorted signals allows the distortion to propagate through the plan unchallenged.

The sharpest illustration is the worst-performing culture on the entire map, and it is not the one you would guess. It is not the cynical, everyone-for-themselves corner. It is the corner where everyone is local and everyone is trusting—every function optimizing its own KPI in good faith, and every other function believing the inflated signal and passing it faithfully downstream. Honest, selfish, and trusting is the most expensive combination a supply chain can run, because nobody is filtering anything. In that light, the low-trust planner who haircuts sales’ forecast is not a bug. She is the immune system. That mechanism also has a real-world precedent: Oliva and Watson documented a consumer-electronics S&OP process in which a demand management office played much the same role, explicitly countering functional forecast bias (Oliva & Watson, 2009; 2011).

Exhibit 1. Company EBITDA across 121 uniform cultures (brighter = more profitable). The profitable column runs down the aligned edge regardless of trust – locality, not trust, is what moves the company result.

Trust is earned by the structure, not chosen

The most useful result comes when you stop holding the dials fixed and let trust move (Exhibit 2). Take two companies. The first is honest-but-selfish: every function local, every function trusting, the expensive corner we just described. Watch its trust over time and it does not stay high. It collapses within a handful of cycles—the gaming becomes visible in missed commitments and inflated asks, functions learn to distrust each other, and trust settles at a mediocre, permanently suspicious level. The organization that started fully trusting ends up chronically wary, and no amount of goodwill at the start prevented it.

Now run the opposite company: aligned-but-paranoid. Every function is incentivized on the enterprise outcome, but every function starts deeply skeptical of the others—trust at zero. Its trust does not stay low. It climbs, steadily, cycle after cycle, because aligned incentives generate nothing to distrust: there is no gaming to detect, no betrayal to remember, so the suspicion has nothing to feed on and slowly starves. Around the tenth cycle, the paranoid-but-aligned company’s trust crosses above the trusting-but-selfish company’s and it keeps climbing toward near-complete trust.

Two features of that climb are worth naming, because they match what practitioners see every day. Trust falls fast and rebuilds slowly: in the model, a betrayal cuts trust by roughly half in a single cycle, while recovery adds it back about a tenth at a time. Trust arrives on foot and leaves on horseback. And it heals unevenly—the planner’s suspicion of sales, the one relationship the structure stresses hardest, is the last to recover, long after the others have mended.

Read that crossover again, because it inverts the usual advice. You cannot simply decide to be a high-trust organization. You can build a structure that earns trust—and if you do, trust arrives even from a paranoid starting point. Conversely, trust drains away if the structure rewards gaming. The planner’s haircut is the story: not a character flaw to coach away, but a rational filter made necessary by the system. Remove the incentive to inflate, and the haircut becomes unnecessary.

That result also has a people side the model does not price. From a practitioner perspective, the planner who repeatedly challenges a colleague’s number can quickly become “the difficult one” in the room, even when that challenge protects the enterprise plan. Fixing the incentives therefore does more than improve the economics: it removes the need for one person to carry institutional distrust on everyone else’s behalf. Leadership has to frame the change as learning from the system, not as a search for who is to blame.

A low-trust reading is therefore a diagnostic signal, not proof of bad intent or even bad incentives: a genuine service failure can also make a reliable organization look less trustworthy for a time. This distinction matters: trust research finds that competence failures and integrity violations repair differently. Our model does not yet distinguish between them (Kim et al., 2004; 2006).

Exhibit 2. Trust over 48 planning cycles. The honest-but-selfish company (starts fully trusting) loses it; the aligned-but-paranoid company (starts at zero) earns it, overtaking around cycle 10.

Why the dysfunction survives every review

If misalignment is this expensive, why does it survive quarter after quarter of management attention? Because from the inside it does not look dysfunctional. It looks like performance.

In a dysfunctional cycle, four of the five functions look better on their own scorecards than in the aligned company. Sales reports higher attainment against its inflated plan; manufacturing runs fuller lines at lower unit cost; supply chain shows slightly better turns; and finance sees slightly leaner inventory. Procurement is the exception. Yet the company loses about €300,000 of EBITDA. Nobody in the S&OP review appears to be failing because the review examines local scorecards. The loss appears only in enterprise EBITDA and working capital, where no single function’s report explains it. Dysfunction survives because the instrument used to inspect the chain is partly blind to the failure mode. For practitioners, this may be the most recognizable result: a dysfunctional S&OP meeting does not necessarily feel dysfunctional; people can leave believing they did exactly what their function asked of them.

And before anyone reaches for the obvious remedy: more visibility will not fix this. In the model, visibility is already perfect—every function sees the same statistical baseline, and the planner can quote sales’ bias to its face in the meeting. The gaming continues anyway, because sales is not misinformed; it is paid. A control tower answers an information problem. S&OP dysfunction is an incentive problem wearing an information costume. Visibility helps at the edges—it simply cannot neutralize an incentive to game the very number everyone can already see.

The disruption ends, but the bill keeps running

There is one more finding that matters especially to anyone running a resilience program, and it only appears once trust is allowed to move (Exhibit 3). We hit the aligned company with a three-cycle disruption—a demand spike in one run, a capacity crunch or a supplier failure in others—then removed it, and watched what happened next.

Service recovered almost immediately: back within two points of baseline by the ninth cycle, six cycles after the shock ended. But trust did not recover with it. The shock taught the functions to hedge and to distrust, and those behaviors outlived their cause. Across the tail—the cycles after the disruption was over—the company lost a further €0.6 to €1.0 million through the defensive buffers and haircuts that earned distrust leaves behind. The disruption had a behavioral half-life measured in quarters, long after the operational event was gone.

The practical implication is uncomfortable for how we report resilience. A recovery measured by service level declares victory roughly ten cycles too early. The chain looks recovered—the fill rate is back—while it is still quietly paying for the disruption in trust it has not yet rebuilt. If your post-mortem closes when service returns, you are booking a win in the middle of the loss.

Exhibit 3. An aligned company hit by a three-cycle demand spike. Service recovers within two points by cycle 9 – but EBITDA keeps bleeding for another ~€1.0M after the shock ends: the behavioral tail of distrust.

Paranoia is insurance you don’t need to pre-buy

If a skeptical planner is the chain’s immune system, should the whole organization remain paranoid as a hedge? Our experiments say no. Distrust behaves like insurance against distorted signals: it can help when a demand spike or allocation scramble amplifies an inflated ask. But it does little against a genuine operational failure—a supplier is late or a line is down regardless of what anyone believed—while permanent suspicion carries a cooperation cost.

Because trust is adaptive, the organization does not need to pre-buy that insurance. In repeated shocks, skepticism emerged where evidence deteriorated and relaxed as conditions normalized; standing paranoia won only when shocks were almost constant. The lesson is targeted challenge, not generalized distrust: place skepticism where gaming enters the process.

What to do: align the incentives first—and fully

So how do you actually fix a chain whose incentives reward gaming? We ran a tournament of transformation strategies, starting each contender from full dysfunction and scoring their cumulative results over two years (Exhibit 4). Three lessons held up.

Complete alignment beats partial alignment economically. Under our deliberately frictionless transformation assumption—no implementation cost, adoption delay or failure risk—moving all five functions to the enterprise objective at once beat every phased alternative, worth roughly €6.8 million in cumulative plan value over two years versus doing nothing. A halfway move captured only about 29% of the full-alignment value in this model. Read the result as the economic cost of leaving misalignment in place, not as a universal prescription for a big-bang reorganization.

The offsite pays—but only as the second step. The trust-building gesture was not worthless. Layered on top of the structural fix, it added real value. On its own, it added nothing that lasted. Sequence is the whole story: align the incentives first, then convene the offsite, and it reinforces a structure that now supports it. Do it in the other order—offsite first, structure later or never—and you are back to the handshake that relaxes within a quarter.

If you must sequence, fix the filter before the offender. In this model, aligning the planning function first beats aligning sales first, even though sales was the louder problem. An aligned planner stabilizes the consensus while other functions are still optimizing locally; fixing sales first can leave defensive filtering in place. The direction is useful, but the size of the advantage depends on calibration.

Exhibit 4. The transformation tournament: cumulative EBITDA gain versus doing nothing, over 24 cycles. Big-bang alignment wins; a halfway compromise captures barely a third.

Seven moves, in the order that works

For a leader who wants this as a sequence rather than a set of findings, here is what the model prescribes, in order.

1. Fix the incentives first, and fully. Put a meaningful enterprise component—company EBITDA and end-to-end service—into every function’s scorecard, translated into its operating language: schedule adherence rather than raw utilization for manufacturing, total cost of ownership rather than price variance for procurement, and so on. Remove KPIs that actively pay for gaming.
2. Make it the CEO’s mandate, not the S&OP team’s. More on why below—but this is the one move that cannot be delegated.
3. Measure the hidden corrections and audit the scorecards. Track the recurring gap between submitted sales forecasts and the consensus released by planning, controlling for genuine new information and constraints. Then ask about every major conflict: whose local KPI improved while the company’s results suffered? That is your gaming map.
4. Trace signed bias, not only forecast error. Publish persistent forecast bias by function, channel and decision point alongside forecast value added. Once the planner’s haircut becomes a shared, auditable correction rather than a private act of distrust, it can disappear when the underlying bias disappears.
5. Build end-to-end understanding deliberately. Put the company scorecard next to each local scorecard in the same S&OP deck, so “dysfunction looks like performance” becomes visible in the room, not just in the year-end results.
6. Then, and only then, hold the offsite. It pays as reinforcement of a structure that now earns trust—never as a substitute for building one.
7. Budget the rebuild time, and resist the halfway compromise. After a disruption or reorganization, track behavioral recovery—forecast haircuts, hedge levels, override rates and expedite spend—alongside service recovery. Trust can recover over quarters, not weeks. And treat partial alignment as an intermediate state, not a destination: in this model it captured only a fraction of the full value.

Whose job this actually is

This also explains why many S&OP transformations stall. No function can rewrite a peer’s scorecard. Supply chain can see the end-to-end consequences, but it cannot change how sales is compensated or how manufacturing’s utilization target is set. It can haircut an inflated forecast; it cannot remove the incentive to inflate it.

That leaves the CEO with the authority to realign cross-functional incentives. This is not a facilitation task for the S&OP team; it is an enterprise operating-model decision. The problem hides because functional dashboards can stay green while the loss appears only in enterprise EBITDA and working capital. The CEO therefore has to look deliberately for conflicts in which a local KPI improved while the company’s results deteriorated.

A warning sits alongside that mandate. When we handed one function the power to keep the peace—letting finance enforce a tight cap that shut down the arguments—escalations fell to almost nothing, and it earned the least of any strategy we tried. A calm S&OP calendar is not evidence of a healthy chain. Meeting peace and chain health are different things, and it is entirely possible to buy the first at the expense of the second. A CEO looking for signs that the realignment is working should watch the enterprise numbers, not the quiet of the room. The practitioner corollary is that the referee role itself must be carefully designed: if governance sits entirely within one function, that function’s incentives can quietly become the enterprise’s decision rule.

One last reason this matters now. As companies hand planning to AI agents, this structure question does not go away—it sharpens. Every agent in this simulation gamed with a fixed level of cunning: sales inflates the number, but never learns to anticipate the planner’s haircut and adjust against it. The next question is what happens once agents start anticipating each other. Game theory suggests an uncomfortable answer: intelligence sharpens local optimization rather than dissolving it. That is the subject of our next piece.

Trust was the consequence. If we want more of it between our functions—and we should, because the durable version is worth a great deal in EBITDA, in working capital, and in the resilience to absorb the next shock without a year-long behavioral tail—then we have to stop trying to build it directly, and start earning it, by fixing the structure underneath.

Literature

• Kim, P. H., Ferrin, D. L., Cooper, C. D., & Dirks, K. T. (2004). “Removing the shadow of suspicion: The effects of apology versus denial for repairing competence- versus integrity-based trust violations.” Journal of Applied Psychology, 89(1), 104–118.
• Kim, P. H., Dirks, K. T., Cooper, C. D., & Ferrin, D. L. (2006). “When more blame is better than less: The implications of internal vs. external attributions for the repair of trust after a competence- versus integrity-based trust violation.” Organizational Behavior and Human Decision Processes, 99(1), 49–65.
• Oliva, R., & Watson, N. (2009). “Managing functional biases in organizational forecasts: A case study of consensus forecasting in supply chain planning.” Production and Operations Management, 18(2), 138–151.
• Oliva, R., & Watson, N. (2011). “Cross-functional alignment in supply chain planning: A case study of sales and operations planning.” Journal of Operations Management, 29(5), 434–448.

About the authors

Knut Alicke is a partner emeritus at McKinsey & Company and teaches supply chain management as a Professor at the University of Cologne and KIT, Karlsruhe.

Ida Hedman is Director of Supply Chain Planning & Logistics at Seco Tools, part of the Sandvik Group.

Author’s note: The simulation described here was built and run with the assistance of AI coding tools, and AI writing tools were used in drafting this article. The views expressed in this paper are those of the authors and do not necessarily reflect the view of Seco Tools

Today is Tuesday, Sept. 9.

1. Expanded tanker attacks deepen the Hormuz shipping crisis

Iran and the United States struck multiple tankers near the Strait of Hormuz as the conflict intensified, further constraining a corridor that normally carries about one-fifth of global oil supplies.

Reuters | Read the original

2. GE Aerospace moves to control a critical engine bottleneck

GE Aerospace agreed to acquire precision-castings supplier Consolidated Precision Products for $11.75 billion, bringing a persistent engine-production constraint in-house as commercial and defense demand stretches industry backlogs into the next decade.

Reuters | Read the original

3. Federal procurement becomes the next U.S.-Canada pressure point

President Trump directed the General Services Administration to begin removing Canadian-origin products from federal purchasing schedules, potentially extending the bilateral trade dispute into a government procurement market exceeding $110 billion annually.

Reuters | Read the original

4. ASML starts major capacity expansion for AI chip equipment

ASML broke ground on a 35-hectare Dutch manufacturing campus that could support 20,000 workers, with logistics and cleanroom capacity scheduled for 2029 as advanced lithography production remains largely booked through 2027.

Reuters | Read the original

5. U.S. scrutiny raises sourcing questions for Ford’s China ties

The Transportation Department urged Ford to sever relationships with CATL, Geely and BYD, challenging the automaker’s battery-technology licensing and continued Chinese production while Ford defended its control of U.S. operations.

Reuters | Read the original

6. Coal constraints push a key Indian steel input to a two-year high

Indian sponge-iron prices reached a two-year high as expensive imported coal, lower domestic availability, monsoon disruptions and power-sector demand tightened feedstock supply for the world’s largest sponge-iron producer.

Reuters | Read the original

7. Ukraine retailers disperse inventory after warehouse attacks

Russian strikes have reportedly damaged food, e-commerce, medical and humanitarian warehouses around Kyiv, prompting retailers to consider smaller, dispersed or cross-border storage networks as product availability and logistics costs come under pressure.

The Guardian | Read the original

8. Japanese manufacturers strengthen outlook on semiconductor demand

Sentiment among large Japanese manufacturers reached its highest level since December 2021, led by electronics and AI-related capital spending, although companies identified Middle East tensions and raw-material costs as continuing risks according to polling conducted by Reuters.

Reuters | Read the original

9. Demand planning must adapt to products with shifting uses

A new Institute for Supply Management article argues that products serving multiple purposes amid rapid software change require demand planners to replace traditional forecasting assumptions with more responsive and adaptive methods.

Institute for Supply Management | Read the original

Executive takeaways

Early-bird registration for the 2026 NextGen Supply Chain Conference will end Monday, Sept. 14, marking the final opportunity for attendees to take advantage of this special discounted-registration price for the Oct. 21–23 event at the W Nashville.

Registration includes access to the conference’s keynotes, awards presentations, featured presentations and interactive Small Group Sessions. It also covers the Wednesday welcome reception; breakfast, lunch, breaks and evening reception on Thursday; and breakfast and a morning networking session on Friday.

Built around the theme “Innovate. Upskill. Transform.,” NextGen brings together supply chain executives, operational leaders, academics and technology experts to examine how organizations are turning emerging ideas into measurable results.

The conference will address transformation across logistics and fulfillment, retail, food and beverage, chemicals and pharmaceuticals. Sessions will explore artificial intelligence, automation, robotics, supply chain planning, workforce development, resilience and the leadership skills needed to guide increasingly digital operations.

Practitioners share what happens after implementation begins

NextGen is designed around candid discussions of execution—not just what a technology promises to do, but what organizations encounter when they attempt to deploy it across real-world supply chain operations.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

The 2026 agenda features executives and practitioners from companies including Wayfair, Tractor Supply Company, Eli Lilly, Mars Snacking, CVS Health, Ryder, BJC HealthCare, Target, Apple, Amazon, Fanatics, Penske Logistics, Johnson & Johnson, DP World, Stanford Health Care, GE HealthCare and Southern Glazer’s Wine & Spirits.

Nitin Kapoor, vice president and general manager of technology, data and innovation at Wayfair, will join SCMR Editor-in-Chief Brian Straight for a keynote fireside chat on Wayfair’s logistics evolution. The conversation will examine how the retailer has used technology to build an integrated logistics network capable of delivering speed, reliability and scale across a complex home-delivery operation.

Craig Ledbetter, senior vice president and chief supply chain officer at Tractor Supply Company, will participate in the Visionary Award keynote, “Supply Chain as a Growth Engine.” The discussion will explore how investments in network expansion, fulfillment, operational scalability and last-mile capabilities can turn supply chain performance into a source of business growth and competitive advantage.

Dr. Mar Gimeno, associate vice president of U.S. supply chain and global launches at Eli Lilly, will deliver a keynote on agentic AI and the future of supply chain decision-making. The session will bring a pharmaceutical supply chain perspective to one of the industry’s most closely watched technologies.

Other presentations will examine how Fanatics is applying agentic AI to demand forecasting, how GE HealthCare is connecting AI investments to cash flow and financial performance, and how Stanford Health Care combined control-tower capabilities, intelligent planning and automation to improve supply chain performance.

Sponsors support the NextGen experience

The 2026 NextGen Supply Chain Conference is supported by technology providers and service organizations looking to connect with senior supply chain decision-makers. Current sponsors listed in the conference materials include:

• Diamond Sponsor: Zion Solutions Group
• Platinum Sponsor: Gather AI
• Gold Sponsors: Cycle Labs, Dematic, Geek+, Dexory and Zimark
• Bronze Sponsor: Verity
• Associate Sponsors: AutoScheduler and Argano

Sponsorship opportunities remain available, including a limited number of Gold Sponsorships. Gold Sponsors receive a 30-minute customer case study presented jointly with an end-user customer, giving attendees a practical look at how supply chain technology is being implemented in real-world operations.

Smaller sessions encourage deeper conversations

In addition to general-session keynotes, fireside chats and panel discussions, NextGen features 30 Small Group Sessions organized across five concurrent rooms.

These interactive sessions are intended to create a different conference experience. Rather than watching every presentation from the back of a large ballroom, attendees can select topics that align with their industries, responsibilities and current transformation priorities.

The smaller format gives participants an opportunity to ask detailed questions, compare experiences and discuss implementation challenges directly with presenters and fellow supply chain leaders.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

Topics range from warehouse intelligence and computer vision to AI-enabled talent, automation testing, healthcare supply chain orchestration and distribution-center transformation. Case studies will include the lessons organizations learned during implementation, the operational and change-management challenges they encountered, and the results they achieved.

Recognizing supply chain transformation

The conference will also recognize the winners of the 2026 NextGen Supply Chain Awards.

End-user honorees include Mars Snacking for Intelligent Transformation, CVS Health for Autonomous Operations, and Ryder and BJC HealthCare for Partnership in Execution. Tractor Supply Company will receive the Visionary Award.

Solution-provider honorees include Netstock for Intelligent Transformation, Pickle Robot for Autonomous Operations and Robust.AI as the Startup Award recipient.

Award presentations are integrated into the conference program so attendees can hear directly from the organizations and leaders behind the initiatives. The emphasis is not simply on recognizing innovation, but on sharing the decisions, partnerships and implementation lessons that moved each project from an idea to an operational result.

Networking with a Nashville soundtrack

NextGen’s educational program is complemented by several opportunities for attendees to build relationships in a more informal setting.

The conference begins Wednesday evening with a welcome reception and networking event featuring a performance by Nashville singer-songwriter Nick DeLeo.

On Thursday, attendees will gather for lunch and networking at Zaytinya, with a performance by Nashville country-pop artist Emma White. The day concludes with a rooftop reception at the W Nashville featuring songwriter Travis Hill, who writes under the name Scooter Carusoe.

Carusoe has more than 20 No. 1 songs associated with his work as a songwriter and publisher. His songs have been recorded by artists including Kenny Chesney, Darius Rucker, Brett Eldredge, Tim McGraw, Taylor Swift, Keith Urban and Eric Church.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

The receptions, meals, breaks and Small Group Sessions are all structured to give attendees time to connect with peers confronting similar workforce, technology and operational challenges.

For supply chain leaders, those conversations can be as valuable as the presentations. They provide an opportunity to compare strategies, exchange lessons and build relationships with executives from different industries and points across the supply chain.

Register before the price increases

The 2026 NextGen Supply Chain Conference will take place Oct. 21–23 at the W Nashville in downtown Nashville.

Discounted early bird registration is available through Monday, Sept. 14. The rate increases after that date.

Register for the 2026 NextGen Supply Chain Conference or view the complete conference agenda.

1. Canada activates broad counter-tariffs on U.S. imports

Canadian importers now face surtaxes of 15%, 25% or 50% on selected U.S.-origin goods, including steel, dairy, appliances, agricultural equipment, paper products and electronics, as the bilateral trade dispute escalates.

Canada Border Services Agency | Read the original 

2. Attacks on Saudi energy sites deepen fuel-supply risk

Houthi attacks damaged Saudi energy facilities and prompted operational shutdowns at some sites, pushing crude prices higher as shippers and manufacturers confront another potential source of fuel and transportation-cost volatility.

Reuters | Read the original

3. China trade surge changes the demand signal for global suppliers

China reported August export growth of 25% and import growth of 28.2% from a year earlier, with high-technology shipments leading gains and the monthly trade surplus reaching $119.09 billion.

Reuters | Read the original

4. Bombardier threat puts a major U.S. supplier network in play

President Trump said Bombardier must manufacture in the United States to retain market access, but offered no enforcement mechanism; the Canadian jetmaker says it supports 2,800 U.S. suppliers and spends more than $2.5 billion annually with them.

Reuters | Read the original

5. Hapag-Lloyd restructures ZIM bid around Israeli control

Hapag-Lloyd is revising its proposed $4.2 billion purchase of ZIM to preserve Israeli ownership of a carved-out 16-vessel operation, seeking government approval while addressing national-security and maritime-access objections.

Reuters | Read the original

6. India steel buyers brace for another input-cost increase

Indian hot-rolled coil prices reached a four-year high after rising 4,000 rupees per metric ton since August, as higher coking-coal costs, mill maintenance and lean distributor inventories tighten near-term supply.

Reuters | Read the original

7. Chinese automakers lean harder on exports as home demand falls

China’s August passenger-vehicle exports rose 77.5% to 894,000 units while domestic sales fell 23.7%, intensifying overseas capacity, logistics and trade-policy considerations for automakers and their suppliers.

Reuters | Read the original

8. Malaysia adds two data centers to the AI capacity race

Nvidia-backed Firmus said it signed a multi-year agreement to supply OpenAI with computing capacity from two Malaysian data centers, expanding regional infrastructure demand while energy and water constraints draw scrutiny.

Reuters | Read the original

9. Jaguar Land Rover pairs workforce cuts with technology investment

Jaguar Land Rover plans to eliminate 4,000 jobs over two years and save £1.7 billion, while directing £15 billion to £18 billion toward electrification, digital systems and manufacturing upgrades.

Associated Press | Read the original

10. Latvia proposes steep grain tariff as Baltic routes gain importance

Latvia plans a 300% tariff on grain from Russia and Belarus, threatening an alternative export corridor that has gained strategic importance as attacks disrupt Black Sea and Sea of Azov shipping.

Reuters | Read the original

Executive takeaways

Supply chain leaders are right to automate repetitive planning, procurement, transportation, and inventory work. The danger starts when companies assume that because AI can perform more junior tasks, they need fewer junior people.

The latest labor-market evidence suggests that this substitution is already reshaping early careers.

Stanford Digital Economy Lab researchers report that the employment shortfall for workers ages 22 to 25 in highly AI-exposed occupations widened from 15% in the July 2025 data vintage to 19% by June 2026. The widening is driven primarily by reduced hiring of young workers.

Supply chains should be especially cautious about that trend because much of the field’s value comes from handling exceptions. Forecasts are easy until demand changes suddenly. Procurement is straightforward until a supplier misses a commitment. Transportation planning works until weather, labor, capacity, customs, or a customer requirement disrupts the plan.

SCMR recently argued that AI is changing supply chain skills and talent and that companies need to redesign early-career development paths. That should become an operating requirement, not merely an HR recommendation.

Automate the task, not the entry-level job

A junior planner should no longer spend hours assembling a routine forecast that AI can generate faster. But that planner should still exist. Give the employee the forecast and ask what assumption would break it. Have the employee investigate where the system’s historical pattern no longer matches current business conditions.

A junior buyer should not manually compare every supplier quote. Let AI normalize the proposals. Then have the buyer analyze supplier reliability, switching costs, operational dependencies, capacity risk, and the consequences of a seemingly cheaper choice.

A transportation analyst should not spend all day producing standard route plans. Let the system optimize the baseline. Then give the employee the disrupted network and require a decision that balances service, cost, inventory, customer priority, and operational reality.

Those are stronger entry-level jobs than the ones automation removes.

They also address a basic supply chain truth: resilience comes from judgment under imperfect information. AI can process more data than a human. It does not eliminate the need for people who know when the data is incomplete, when a supplier’s promise is unrealistic, when a customer constraint matters more than the mathematical optimum, or when a small operational problem is about to become a large one.

SCMR’s recent coverage of AI-enabled supply chain operations makes the same broader point: technology only works when organizations strengthen coordination, decision-making, visibility, and workforce capabilities.

Make independent competence the new talent metric

The missing metric is time to independent competence.

How quickly can a new planner explain why the AI forecast is wrong? How soon can a buyer make a credible sourcing recommendation when cost, resilience, and supplier performance point in different directions? When can an analyst manage a disruption without a senior employee rewriting the plan?

If AI shortens those learning curves, it is solving two problems at once: productivity and talent development.

If AI instead reduces junior hiring, supply chain organizations may save money while weakening the bench that produces future category managers, planning leaders, logistics directors, and chief supply chain officers.

That risk compounds across the industry. One company can decide to hire only experienced people. The entire industry cannot. Someone has to create the experienced people everyone later wants.

The widening 15% to 19% early-career gap says the collective pipeline is already under pressure.

Turn every AI deployment into an apprenticeship

Supply chain leaders should therefore treat every AI deployment as both a technology project and an apprenticeship redesign. Automate the routine work. Capture expert overrides. Turn disruptions into teaching cases. Require junior employees to explain disagreements with the system. Rotate them across planning, procurement, logistics, and operations so they understand how one decision creates consequences elsewhere.

The future of AI adoption at work in supply chains should not be fewer chances to enter the profession. It should be a faster path from beginner to capable decision-maker.

About the author

Gleb Tsipursky, PhD, is a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).

Executive takeaways

Economists are finding it increasingly difficult to predict the direction of the economy. One reason is the growing importance of the so-called K-shaped economy, a concept popularized during the COVID-19 pandemic.

Initially, the K shape illustrated how industries experienced sharply different recoveries. Some sectors rebounded quickly and expanded, forming the upper arm of the K, while others struggled and recovered much more slowly, forming the lower arm. Today, the concept increasingly describes a structural divide in consumer income and spending.

The upper part of the K represents higher-income consumers, who are generally better positioned to absorb inflation and maintain discretionary spending. The lower part represents lower-wage consumers, who are more exposed to rising costs for housing, food, energy, and other basic needs. While affluent households may continue spending, lower-income households may reduce purchases, trade down to less expensive products, or eliminate nonessential expenses altogether.

This divergence creates a significant challenge for traditional economic models. Many models rely on averages for income, spending, inflation, and unemployment. However, an economy that appears healthy on average may conceal substantial weakness within particular income groups, product categories, or geographic markets.

Why the K shape complicates forecasting

A K-shaped economy makes forecasting more difficult because different consumer groups respond differently to the same economic conditions. Interest rates, inflation, wage growth, and changes in employment may have a limited effect on affluent consumers while creating considerable pressure on lower-income households.

Aggregate results can therefore be misleading. Growth in premium products may offset declining demand in value or mid-tier categories, hiding important changes inside the overall numbers. Premium and value products may even grow simultaneously while mid-priced products decline. Demand can also vary significantly by region, depending on employment patterns, local industries, household income, and living costs.

Companies can no longer assume that a single macroeconomic indicator will affect every customer or product in the same way. The economy has become more heterogeneous, and supply chain planning must reflect that complexity.

Resilience begins with segmentation

This environment is developing alongside persistent supply chain disruptions. The pandemic, material shortages, supplier shutdowns, transportation constraints, energy-market volatility, geopolitical tensions, and other events have demonstrated that stability cannot be taken for granted.

The starting point for resilience is segmentation. Companies need to understand where their products, customers, and markets sit within the K-shaped economy. They must identify which categories are benefiting from stronger spending, which are under pressure, and which could move rapidly as economic conditions change.

Organizations must also acknowledge that they cannot make every part of the business equally flexible or resilient. Attempting to protect every product, material, and supplier relationship would be prohibitively expensive. Some exposure is unavoidable. The objective is to protect the areas that are most critical to profitability, customer relationships, and operational continuity.

Segmentation should consider factors such as margin contribution, demand volatility, customer importance, substitution options, supply risk, lead time, and geographic exposure. This creates a stronger foundation for deciding where resilience investments will generate the greatest value.

Forecasting at the segment level

In the past, companies could often apply broad economic indicators across the business or use historical trends to estimate future growth or decline. In a polarized economy, those methods may no longer provide sufficient detail.

Forecasting models should be refined by product, customer, income group, channel, and geography whenever the available data supports that level of analysis. A national consumer-spending forecast, for example, may have limited value if premium demand is rising in one region while value-oriented demand is weakening somewhere else.

Forecast accuracy should also become an organization-wide priority. It is not solely the responsibility of demand planning or supply chain teams. Poor forecasts can create excess inventory, product shortages, manufacturing inefficiencies, wasted materials, inaccurate financial projections, and higher transportation costs. The effects can spread across supply chain, finance, sales, procurement, and manufacturing.

Companies should continuously measure forecast accuracy, identify recurring sources of error, and refine their assumptions. Planning for continued economic polarization can be a useful baseline, but it should not become a fixed conclusion. Forecasts must remain flexible enough to recognize when consumer behavior begins to change.

Connecting segmentation with scenario planning

Segmentation and scenario planning must work together. Companies should develop a manageable number of credible scenarios that reflect how the K-shaped economy could evolve. These might include continued premium growth, increased trading down among middle-income consumers, regional demand deterioration, higher input costs, or constraints involving critical suppliers and materials.

Each scenario should answer practical questions. What inventory levels would be required? Which supplier or material constraints could emerge? How would the cost to serve change? Which products deserve stronger protection? Where could inventory be reduced and additional risk accepted? Which materials require multiple sources to create flexibility?

Every scenario should also include an actionable playbook. A common planning failure is to create numerous scenarios without defining what the organization will do if any of them occur. Effective playbooks establish decision triggers, responsibilities, response options, and expected financial consequences. They convert scenario planning from an analytical exercise into an operational capability.

Improving long-term decisions

Many supply chain decisions are costly, difficult to reverse, and dependent on long lead times. Adding production capacity, qualifying a new supplier, relocating sourcing, increasing strategic inventory, or redesigning a distribution network may require commitments based on uncertain demand projections.

For that reason, organizations should evaluate the confidence level associated with each forecast and scenario. Leaders need to understand not only the expected outcome but also the range of possible outcomes and the risks attached to each one.

Scenario planning allows companies to evaluate trade-offs before conditions force an immediate response. Higher inventory may improve service and reduce stockout risk but increase working capital and obsolescence. Local suppliers may be more expensive but provide shorter lead times and greater reliability. Multiple sourcing may improve flexibility but add complexity and reduce purchasing leverage. These decisions should be assessed by segment rather than applied uniformly across the business.

Combining AI with human experience

Artificial intelligence can perform much of the analytical heavy lifting. It can process large datasets, identify patterns across demand segments, evaluate economic indicators, and generate potential scenarios more quickly than traditional methods.

However, AI-generated outputs still require interpretation. Experienced supply chain professionals understand operational constraints, supplier behavior, customer priorities, and risks that may not be fully represented in the data. Their judgment is essential when converting analysis into long-term decisions.

Resilience in a K-shaped economy ultimately depends on combining detailed segmentation, accurate forecasting, disciplined scenario planning, and experienced human judgment. Companies that develop these capabilities will be better equipped to protect profitability, allocate resources intelligently, and respond when different parts of the market move in opposite directions.

About the author

Vinicius Giarola is the Executive Director of Consumer Supply Chain, Logistics & Supply Chain at Bridgestone Americas, Inc.

Executive takeaways

A supply chain control tower can show hundreds of late orders, constrained parts and shifting arrival dates before breakfast. The dashboard may be accurate. The predictions may be sophisticated. Yet the planner still faces the same question: Which problem should I work first?

That question is where many AI business cases become fragile. Organizations invest in better forecasts and faster alerts, but the operating model around exceptions remains manual. Every warning enters the same queue. Teams investigate the same facts in different systems. The loudest customer or the most senior escalation often determines priority.

Supply chain AI earns its return when it changes that daily decision. The real product is not another prediction. It is an exception queue that helps people understand impact, choose the next action and learn from the outcome.

An alert is not an exception

An alert says that a threshold was crossed. An exception says that a business commitment is at risk and requires a decision.

The distinction matters. A shipment that is two days late may have no customer impact because inventory is available at the destination. A four-hour delay may stop a production line. Treating both events equally creates noise and trains planners to ignore the system.

A useful exception record connects the signal to the business context: affected order, customer promise, inventory position, production dependency, financial exposure and recovery window. Without that context, AI only accelerates awareness. It does not accelerate action.

Rank by consequence and time to recover

Most teams naturally rank exceptions by severity. They should add a second dimension: time to recover.

Some problems are expensive but can wait. Others look small but will become irreversible within hours. A part shortage may be manageable before the production sequence is frozen. A carrier delay may be recoverable before a cutoff time. A supplier quality issue may require immediate containment even when no customer order is late yet.

The queue should therefore answer two questions for every exception: What happens if no one acts, and how long does the team have before the best recovery option disappears? That combination turns a list of risks into a work plan.

Show the evidence behind the recommendation

Planners will not trust a recommended action if they cannot see why the system proposed it. The explanation does not need to expose every model calculation. It does need to show the operational evidence.

If the recommendation is to expedite, the planner should see the customer commitment, available inventory, alternate lanes, cutoff time and cost difference. If the recommendation is to hold, the planner should see the buffer that makes delay safe.

This is especially important when data conflicts. The system may see one arrival date in transportation, another in the supplier portal and a third in the ERP. A confident recommendation built on unresolved conflict is more dangerous than a cautious request for review.

Design the queue around decisions

A practical exception queue should separate four states. The first is observe: the risk is visible, but no action is required. The second is investigate: evidence is incomplete or inconsistent. The third is decide: viable recovery options exist and an owner must choose. The fourth is execute: the action is approved and needs to move across systems and partners.

These states prevent the queue from mixing information work with decision work. They also make ownership visible. A data steward may resolve conflicting attributes. A planner may select a recovery option. Procurement may negotiate with a supplier. Logistics may execute the change.

When every exception has a state, owner and next decision, leaders can see where work is stuck instead of merely seeing how many alerts exist.

Capture the recovery outcome

Many systems close an exception when the alert condition disappears. That is not enough. The organization needs to know what action was taken and whether it worked.

Did expediting protect the customer promise? Did a substitute part create a quality issue later? Did the planner reject the recommendation because the inventory data was stale? Did the team discover that the alert arrived after the recovery window had closed?

Those outcomes are training data for the operating model, not only for the algorithm. They reveal broken master data, unrealistic thresholds, unclear approval rights and supplier processes that create recurring failure.

Start with one expensive exception family

Organizations do not need to redesign every planning process at once. Start with an exception family that is frequent, costly and recoverable: supplier commit changes, late inbound shipments, inventory allocation conflicts or production constraints.

Map the evidence planners gather, the decisions they make, the approvals they need and the actions that follow. Then measure whether the new queue reduces time to decision, repeated investigation, missed recovery windows and unnecessary premium cost.

This approach keeps the AI program close to operational value. It also builds confidence because planners can see the system improving a decision they already understand.

The operating system for disruption

Supply chains will never eliminate exceptions. Volatility, data gaps and competing priorities are part of the work. The advantage comes from seeing the important exception sooner and responding while options still exist.

A forecast tells the organization what may happen. An exception queue tells people what deserves attention now, why it matters and what can still be done. That is the point where AI moves from an analytical feature to an operating capability—and where the return begins to show.

About the author

Hemang Upadhyay is a senior product and AI leader with more than 16 years of experience across enterprise AI strategy, digital commerce, product data governance, customer experience and scalable platform transformation. His writing focuses on the operating controls that help organizations move AI from promising demonstrations into reliable business workflows. LinkedIn: https://www.linkedin.com/in/hemang-up/ Website: https://www.hemangai.com

Those cabs and related components are then moved from Kings Mountain into Volvo Group’s production network in Virginia and Pennsylvania, where they have to arrive on time and in the right sequence to keep production on schedule. Any delay in that flow can affect build plans, customer commitments, and the broader truck market that Volvo Group serves.

Before Volvo Group acquired the Kings Mountain operation, the site was part of the company’s supplier network. Volvo initially set out to help the supplier improve the flow of materials and get cab production back on track. Starting with an assessment of the logistics operation, Volvo looked at the constraints around the site and what it would take to support better movement in and out of the facility.

“We were looking at it as a supplier issue at first,” says Dave Yancey, project manager, production logistics, at Volvo Group North America. “They were behind on our cab production, and it was affecting our markets. We sat down to assess the operation and figure out how to fix the logistics flow, including the warehouse capacity we needed to support the site.”

A familiar partner for a fast-moving problem

Volvo needed a logistics partner that could step in quickly, understand its production environment, and help stabilize the warehouse side of the operation. The company had options, but this project didn’t come with months to plan from scratch. Material was already coming in, cab production needed support, and the Kings Mountain site was still changing.

A. Duie Pyle (ADP) already had a track record with Volvo Group, having provided regional less-than-truckload (LTL) service and value-added logistics support. It also handled cab storage for Volvo Group’s Mack Trucks operation in Macungie, Pa., during an earlier production challenge.

“When I was in Macungie, we had a major success story with the ADP team,” Yancey says. “They were storing our cabs at the time and helped us through some production issues. I liked the system they used and thought this would be a perfect model.”

A nine-day sprint

What started out as a supplier-support effort became part of a larger transition when Volvo Group acquired the Kings Mountain operation. ADP already had agreements in place with the supplier, so the acquisition added contract complexity at the same time the warehouse had to keep receiving, organizing and moving material.

“Once we got the agreements straightened out, it was full speed ahead,” Yancey says. “The attitude from the ADP team was: ‘Tell us what we need to do and we’ll make it happen.’”

And with that, Volvo took over a 180,000-square-foot warehouse that still needed the basics before it could fully support the operation. Material had to be received, staged, and stored while utilities, equipment, and warehouse processes were still being put in place.

“It started as just a warehouse with no electricity,” says Yancey. “We used pallet jacks to unload inbound material, and then we started getting power, forklifts, racking, shelving, and scanning. It went from being an empty shell to a functioning warehouse.”

Working together, Volvo and ADP moved the initial material into the warehouse over roughly nine days, even as new inbound shipments continued to arrive. Once electricity, forklifts, and other equipment were in place, the operation started to function more like a true warehouse. By the following spring, ADP and Volvo were adding racking to increase storage capacity and make better use of the space.

Yancey says ADP brought in a capable team to help Volvo get through that stretch. “We have a lot of good people inside Volvo Group, but we can also be spread very thin,” Yancey says. “ADP was able to jump through hoops, make things happen and bring in the right people to get this done.”

Building the warehouse around the parts

With this project, both Volvo and ADP had to carefully assess where parts belonged inside the warehouse and how quickly those parts needed to move. The operation was organized around truck model and product line, including Class 8 highway trucks, Class 6 trucks and cab-over models.

Each had its own parts profile, and some of the existing supplier’s part numbers didn’t follow Volvo’s normal numbering conventions. “This added complexity to the setup,” says Yancey.

As production patterns became easier to see, ADP and Volvo used that information to set up more directed putaway and improve turn times. “We set the operation up by model so we could move the fast-moving parts to the front and make our turn time quicker,” Yancey says. “If something didn’t move as often, we could put it deeper in the warehouse.”

There were also repacking requirements to consider because some of the manufacturer’s parts came in on one-way pallets or in packaging that didn’t match its standards. Some pallets also arrived with multiple part numbers or incomplete identification, which made them harder to receive, store and move efficiently. ADP had to move those parts into Volvo-approved containers before they could move through the warehouse and back to the plant.

Finally, the new warehouse had to meet Volvo’s compliance requirements, including both ISO and environmental standards. That meant handling cardboard, plastic, wood pallets and other waste streams properly rather than sending everything to a landfill. It added another layer of discipline during the early months, when ADP and Volvo were still improving the physical setup, training new employees, and bringing the warehouse up to Volvo’s standards.

Finding Its footing quickly

Going into the project, Volvo’s supplier had its own warehouse system, but ADP could generate inventory reports multiple times a day and help Volvo compare physical inventory against what the plant expected to see. That gave the teams a better handle on inventory accuracy while the operation was still settling in.

Volvo and ADP also built key performance indicators (KPIs) around the parts of the operation Yancey wanted to measure, including repacking activity, pallet handling and inbound and outbound volume. The prior day’s KPIs were available by noon the next day, followed by weekly and monthly summaries. That gave the team a way to spot bottlenecks, track warehouse performance, and make decisions as the operation matured.

With its first wave of inbound material under control, the newly-acquired operation started to find its footing. Power came online within the first few days, forklifts followed and the warehouse began taking on the pieces it needed to function properly, from racking and shelving to scanning.

By the following spring, ADP and Volvo were adding more racking to increase storage capacity and make better use of the space. “The facility went from being a shell to a functioning warehouse,” says Yancey. “The early stages of the transition were challenging, but overall it was a huge turnaround. The warehouse runs extremely well.”

A relationship built on trust

For ADP, Volvo’s Kings Mountain project grew out of a relationship built on both execution and trust. Chris Incudine, VP of solutions design at ADP, works closely with Yancey and says Volvo was asking the company to step into an unusual situation: help take over a supplier’s logistics operation because the manufacturer needed help stabilizing it.

“If you take the relationship out of it, you basically have a customer asking us to go run their supplier’s operation for them because they had more trust in us than they did in the supplier,” Incudine says. “That’s a unique scenario. There has to be a level of trust there, and that trust came from the partnership we’d already built.”

ADP, which wasn’t operating in North Carolina at the time, also had to step outside its own regional network to support the project. Incudine says the strength of the Volvo relationship made the decision possible. Regional LTL carriers like ADP can be a strong fit for this kind of work because they tend to know their customers closely, understand the freight and can respond quickly when a transportation issue becomes a broader supply chain challenge.

“We wouldn’t have been in North Carolina three years ago if we hadn’t gotten an ask from such a strong partner,” Incudine says. “I think it speaks to the level of partnership and trust we’d built through execution and transparency.”

That transparency had to flow both ways, of course. He says Volvo was direct about what it needed, while ADP was equally direct about what was working, what needed attention, and where the operation needed support. That open communication helped both companies work through a project that was anything but simple. “It wasn’t all easy. There was blood and sweat—and some tears,” he says, laughing, “but we got Volvo what they needed.”

Executive takeaways

Food and beverage supply chains operate under a particularly unforgiving set of demands. Products must move through complex networks quickly and accurately, inventory decisions can carry shelf-life implications, and changing consumer expectations require companies to support new channels without losing control of cost or service.

Those pressures are making artificial intelligence, automation and better-connected fulfillment operations increasingly important across the sector. They will also make food and beverage one of the industry paths attendees can follow at the 2026 NextGen Supply Chain Conference, taking place Oct. 21-23 at the W Nashville in downtown Nashville.

Across the conference’s awards program, main-stage panel discussions and interactive Small Group Sessions, leaders from Mars Snacking, Southern Glazer’s Wine & Spirits, Target and Berry Direct will offer practical perspectives on how food and beverage organizations are improving decisions, modernizing distribution and responding to a more demanding fulfillment environment.

Mars turns AI into faster supply chain decisions

The food and beverage conversation will begin Thursday morning when Mars Snacking receives the NextGen Supply Chain End User Award for Intelligent Transformation.

Kristen Daihes, senior vice president of analytics, digital and data at Mars Snacking, will represent the company during the awards program and discuss how Mars is embedding artificial intelligence into supply chain decision-making.

At the center of the transformation is V2C, or Volume to Customer, an AI-powered platform that brings sales, customer care and supply chain teams into a shared workflow. The platform replaces fragmented tools and manual processes with predictive analytics, machine learning and SAP integration.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

Mars said that V2C reduced a decision-making process that previously required more than 50 labor hours to seconds. The platform has also supported sales enablement, customer service, working capital performance and cross-functional alignment and is now being scaled globally.

The project illustrates an important shift in enterprise AI: The value does not come simply from generating another forecast or dashboard, but from connecting information and decisions across functions so teams can act faster.

Modernizing beverage distribution

Southern Glazer’s Wine & Spirits will bring the beverage distribution perspective to Thursday’s Small Group Sessions in a joint case study with Dematic, “Modernizing Beverage Distribution: How Southern Glazer’s and Dematic Built a Scalable Fulfillment Network.”

Karli Sage, vice president of supply chain management technology and engineering at Southern Glazer’s, and Paul Havens, director of project management at Dematic, will take attendees inside the work required to develop a more scalable beverage fulfillment operation.

The session adds a sector-specific dimension to one of the conference’s central themes: how companies can translate automation investments into operational capabilities that support growth, improve execution and accommodate changing customer requirements.

Like the other Small Group Sessions, the presentation will be offered during both the morning and afternoon blocks, allowing attendees to incorporate the case study into a personalized conference schedule.

Food fulfillment meets the new retail economy

Food and beverage fulfillment will also be represented during Thursday afternoon’s executive panel, “Retail Reinvented: Automation, Omnichannel Execution & the New Fulfillment Economy.”

Eric Watts, vice president of food supply chain operations at Target, and Jay Di Sieno, senior supply chain manager at Berry Direct, will join Jeff Kellan, division president, omnichannel retail in AmAPAC at GXO Logistics. Norman Katz, president and CEO of Katzscan, will moderate the discussion.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

Berry Direct supports the delivery and fulfillment network behind Edible Arrangements, giving Di Sieno a view into a specialized operation where product handling, customer experience and last-mile execution converge. Watts will bring the perspective of a major retailer managing food supply chain operations at scale.

Together, the panelists will examine how consumer expectations, omnichannel demand, automation investments, labor challenges and regulatory complexity are reshaping retail networks. The discussion will focus on the practical tradeoffs involved in balancing service, speed, cost, compliance and profitability.

A broader celebration of supply chain innovation

Mars is one of seven organizations being recognized through the 2026 NextGen Supply Chain Awards, sponsored by Zion Solutions Group. The program is designed to do more than recognize innovation: Winners will share the projects, implementation lessons and measurable results behind their achievements.

CVS Health will receive the End User Autonomous Operations Award for a robotic fulfillment ecosystem that increased daily processing capacity from 150,000 to more than 400,000 units, achieved greater than 99.9% pick accuracy and reduced picking costs by 40%.

Ryder and BJC HealthCare will receive the Partnership in Execution Award for a healthcare logistics collaboration that improved fulfillment, inventory visibility and service while reducing order-processing costs. Tractor Supply will receive the Visionary Award, with Chief Supply Chain Officer Craig Ledbetter discussing how supply chain can become an engine for growth.

To view the latest agenda, click here

To register for the conference, click here

Organizations interested in sponsoring the conference, click here

Friday morning’s Solution Provider Awards will recognize Netstock for Intelligent Transformation and Pickle Robot for Autonomous Operations. Robust.AI will receive the Startup Award for Carter, its collaborative mobile robot designed to work alongside warehouse associates while improving productivity and flexibility.

Nashville networking—with a soundtrack

NextGen’s educational program will be paired with networking opportunities throughout the three-day event, beginning with a Wednesday evening welcome reception and continuing through breakfasts, breaks, lunch and Thursday evening’s rooftop reception at the W Nashville.

The rooftop reception will feature a performance by Nashville songwriter Travis Hill, who performs under the name Scooter Carusoe. His songwriting credits include No. 1 songs recorded by Kenny Chesney, Darius Rucker and Brett Eldredge, along with songs recorded by artists including Tim McGraw, Taylor Swift, Keith Urban, Rascal Flatts, Eric Church, Lady A, Uncle Kracker and Dierks Bentley.

The mix of education, peer discussion and informal networking is designed to give attendees opportunities to continue conversations with speakers and fellow supply chain leaders beyond the formal sessions.

Sponsors support the NextGen experience

The 2026 NextGen Supply Chain Conference is supported by technology providers and service organizations looking to connect with senior supply chain decision-makers. Current sponsors listed in the conference materials include:

• Diamond Sponsor: Zion Solutions Group
• Platinum Sponsor: Gather AI
• Gold Sponsors: Cycle Labs, Dematic, Geek+, Dexory and Zimark
• Bronze Sponsor: Verity
• Associate Sponsors: AutoScheduler and Argano

Sponsorship opportunities remain available, including a limited number of Gold Sponsorships. Gold Sponsors receive a 30-minute customer case study presented jointly with an end-user customer, giving attendees a practical look at how supply chain technology is being implemented in real-world operations.

The 2026 NextGen Supply Chain Conference will bring together leaders from supply chain, logistics, procurement, operations and technology for three days of executive education, networking and peer-to-peer learning. Registration is open, with additional speakers and session details to be announced as the conference approaches.

APQC identifies response latency—the time from disruption to informed operational action—as a critical, often overlooked performance measure. Its model tracks how quickly companies identify impact, make decisions and execute responses.

Benchmark data shows critical risk information can take weeks to influence decisions, adding an average 16.4 days to the response cycle. Organizations can reduce these delays by clarifying ownership, eliminating unnecessary approvals, standardizing response processes and learning from disruptions to improve future supply chain speed, resilience and agility.

For more on this topic, visit www.apqc.org

 

Executive takeaways

Artificial intelligence has become nearly impossible to avoid in supply chain conversations. Planning platforms now include AI-assisted forecasting, logistics providers are embedding predictive models in visibility tools, and manufacturers are connecting equipment sensors with maintenance systems.

Yet activity is not the same as scale. A June 2025 Gartner survey found that only 23% of supply chain leaders had a formal AI strategy for their function. In the 2025 MHI Annual Industry Report, developed with Deloitte, 28% of respondents reported that AI was already in use, while another 54% expected to adopt it within five years. Together, the findings describe a market with strong momentum but uneven operational maturity.

That unevenness helps explain why the most convincing applications today are focused decision services rather than autonomous systems attempting to run an entire network. They address practical questions: Which purchase order is beginning to fail? Will this shipment meet the next cutoff? Is a critical asset showing signs of failure?

These systems extend familiar processes rather than replace them. They draw on ERP, WMS, TMS, asset-management, and IoT data and deliver a risk estimate or recommendation while people still have meaningful alternatives.

To understand how this works in practice, we interviewed two experienced supply chain technology practitioners. Anupam Bandyopadhyay is a senior supply chain technology leader with more than 19 years of global experience in warehouse modernization, logistics systems, automation, and AI-enabled distribution. Ramachandra Handaragal is a senior manager and solution architect at Peloton Consulting Group with 20 years of experience leading digital-transformation projects across aerospace, manufacturing, high technology, retail, food, utilities, and other sectors.

Their perspectives are complementary. Handaragal brings experience from procurement, production, and maintenance programs, while Bandyopadhyay focuses on how warehouse, transportation, automation, and real-time data systems turn AI outputs into action. Their observations point to three areas where AI is beginning to produce tangible value.

1. Supplier risk is moving from scorecards to open-order prediction

Supplier management has traditionally been retrospective. Measures such as on-time delivery, quality, responsiveness, and cost remain important, but they answer a historical question: How has the supplier performed? They are less effective at identifying which open purchase order is beginning to fail.

AI-supported supplier-risk systems move the analysis closer to the individual order, component, lane, and delivery commitment. A model can consider promised-date changes, advance shipping notice timing, receipt patterns, supplier history, transit variability, border delays, port congestion, weather, and external disruptions. No single signal proves that an order will be late, but several weak signals can become meaningful when considered together.

Handaragal first encountered this gap roughly a decade ago while working with an aerospace tier-one manufacturer in India. An Asian supplier repeatedly remained confirmed in the ERP system until an expected shipment failed to arrive. The experience predated today’s generation of AI tools, but it captured a persistent problem: the system displayed a supplier commitment almost as if it were an objective forecast.

“The system was green. The line was down,” he said.

Drawing on that earlier experience, Handaragal sees how AI can close the gap between reported status and probable performance. A model can combine purchase-order-line history, commitment slippage, ASN timing, border dwell, holiday calendars, supplier-specific patterns, and lane performance. In implementations using this approach, he reports that risk flags can emerge four to six days before scheduled delivery.

In daily execution, four days can separate a manageable problem from a fire drill. A buyer may still be able to verify the shipment, transfer inventory, reserve premium transportation, alter the production sequence, or find another source. Once the line is waiting, many of those options disappear.

Bandyopadhyay identifies a related challenge earlier in the inbound process: critical information is often trapped in inconsistent documents. When ASNs are missing or incomplete, facilities may depend on bills of lading whose layouts vary by carrier and vendor. AI-powered document-intelligence systems can extract purchase-order numbers, carton counts, pallet quantities, and weights; assign confidence scores; and route uncertain values to an employee for review. The extracted data can then be compared with the purchase order to flag missing references, quantity discrepancies, or unusual weights before receiving is completed.

Bandyopadhyay also connects AI-enabled inbound information with warehouse replenishment. Demand signals and near-term consumption patterns can help identify high-velocity SKUs, allowing the WMS to reprioritize replenishment and protect product availability for urgent demand. In this way, AI-enabled inbound information becomes an input into warehouse replenishment and inventory decisions rather than remaining isolated within the receiving process.

The same logic applies to expediting. Handaragal recalls an organization spending $8,000 to expedite components worth $3,000 without a shared framework for comparing intervention cost with operational exposure. A low-value part may protect a high-value operation, so the expedite was not necessarily unjustified. AI can help compare the probability and consequence of delay with the cost and likely effectiveness of each possible response.

Supplier AI is therefore moving beyond more elaborate scorecards. It is bringing together order behavior, transportation risk, inbound documents, inventory needs, and economic consequences to show where intervention matters most.

2. Predictive logistics is turning visibility into action

Traditional track-and-trace systems report milestones, showing where freight was at the last update but not necessarily whether it will meet the next production, dock, or customer cutoff.

Predictive ETA systems combine movement and milestone data with route characteristics, carrier performance, traffic, weather, congestion, and historical dwell times. DHL’s Smart ETA service illustrates the approach. DHL reports that its data-driven ocean-freight forecasts have improved ETA predictions by up to 48% compared with carrier estimates. Because this is a company-reported maximum, it is best understood as evidence of the technology’s potential rather than an independent benchmark.

A predictive ETA is useful not because it creates certainty, but because it represents uncertainty more honestly and gives planners a better basis for deciding whether to intervene.

Bandyopadhyay emphasizes that an ETA creates value only when it changes execution. Distribution centers and cross-docks use arrival information to schedule labor, assign dock doors, prepare staging space, and coordinate unloading. In food-service replenishment, these decisions span multiple systems: demand may originate in the ERP, routing commitments in the TMS, and picking and loading in the WMS. A changed ETA can move an urgent order forward in the picking queue, alter an outbound loading sequence, or give a temperature-sensitive delivery priority over freight with a wider window.

He describes the enabling architecture as a decision layer that receives real-time events from ERP, WMS, TMS, and IoT systems through APIs. It evaluates what has changed and returns an updated priority or recommendation to the system where the work is being performed. At that point, predictive visibility becomes orchestration rather than another number on a control-tower screen.

Handaragal’s field experience illustrates the human side of that transition. At a Midwestern distribution center, an inbound coordinator refreshed a TMS every 20 minutes and called carriers whenever a load stopped reporting. The practical visibility process relied on two whiteboards, a telephone, and years of accumulated judgment.

When the coordinator saw a probabilistic ETA model that updated every four hours, he reportedly said, “This is what I have been doing in my head for 15 years.” According to Handaragal, the difference was scale: the model could apply comparable reasoning across approximately 4,000 loads.

Exception management is developing alongside ETA prediction. AI can classify an exception, assemble relevant order and shipment data, estimate its potential impact, and recommend a response. In one unnamed consumer-goods engagement, Handaragal reports that three analysts were handling approximately 300 logistics exceptions per week. After a triage layer was added, roughly 60% were handled through predefined logic or routed to the appropriate employee with a recommended action before an analyst manually investigated them.

The analysts spent less time gathering routine information and more time on unfamiliar or high-impact disruptions. Bandyopadhyay sees the same pattern in warehouse execution: operators still need to understand why a task moved, which operating condition changed, and when human intervention is required. AI can prioritize and recommend, but consequential decisions still require accountable human judgment.

3. Predictive maintenance is becoming a supply chain capability

Predictive maintenance is often presented as an engineering application, but its effects reach across the supply chain. A critical failure can mean lost production, delayed orders, emergency parts purchases, reduced warehouse capacity and, in a cold chain, threatened inventory and food-safety exposure.

Predictive maintenance uses sensor and operating data to evaluate the condition of an individual asset. Models may analyze vibration, temperature, pressure, electrical current, acoustic signals, and other measures for deviations from normal behavior. ABB reports that its condition-monitoring system at Tenaris’s continuously operating Dalmine pipe mill monitors high- and low-voltage motors, helps predict maintenance needs, and supports alignment with planned production stops.

Handaragal describes a manufacturing engagement in which unplanned downtime on a critical conveyor was estimated to cost approximately $2 million annually. The maintenance team was completing its scheduled work, but the schedule had been established 15 years earlier and no longer reflected the equipment’s age or operating intensity.

“The PMs were happening on time,” he recalled. “The failures were happening anyway.”

The organization connected vibration and temperature data with two years of maintenance history. Handaragal reports that the team identified a pattern that preceded motor-bearing failures by eight to 12 days, and that the first prevented failure would otherwise have caused an estimated 16-hour stoppage. Those figures come from his project account rather than a named public case.

The warning created time to locate the bearing, reserve labor, and schedule the repair during a planned production gap. At another client, Handaragal worked on an IoT-enabled predictive-maintenance process connected with Oracle enterprise asset management. The approach reduced mean time to repair by approximately 60%, improved mean time between failures by roughly 45%, generated more than $2.5 million in annual savings per installation, and resulted in zero unplanned downtime for monitored assets during the first year.

Bandyopadhyay approaches the same issue from the warehouse and cold-chain perspective. Refrigeration systems and compressors protect temperature-sensitive inventory, while failures in interconnected warehouse equipment can disrupt fulfillment capacity. An alert may therefore require more than a maintenance work order. Inventory may need to move, inbound loads may need to be redirected, and maintenance timing may have to account for order volume, labor capacity, and transportation departure schedules.

For Bandyopadhyay, integration and trust are inseparable. The alert must connect with systems that know where inventory is located, what shipments are arriving, and what alternative capacity exists. Operators also need to understand why the alert appeared, how urgent it is, and what action is expected.

Handaragal makes the same point through model performance. A system that detects 95% of failures but raises false alarms 40% of the time may create less operational value than one that detects 80% with a 10% false-positive rate. The figures reflect one engagement, but the lesson is broadly applicable: repeated false alarms teach technicians not to trust the system.

The real shift is from AI tools to AI-supported work

The supply chain AI market can appear fragmented because the same label covers forecasting, computer vision, document extraction, optimization, generative interfaces, and autonomous agents. A more useful distinction is between AI that produces an interesting output and AI that changes an operational decision.

Supplier-risk models matter when buyers and warehouse teams can intervene before material is late. Predictive ETAs matter when they change dock, labor, production, inventory, and customer-service decisions. Predictive-maintenance models create value when an early warning mobilizes parts, technicians, capacity, inventory, and contingency plans before an asset fails.

Handaragal’s field experience shows how AI can formalize knowledge traditionally held by buyers, coordinators, planners, maintenance teams, and technicians. Bandyopadhyay’s experience shows how that intelligence becomes executable through document processing, connected sensors, warehouse automation, and integration among ERP, WMS, TMS, and asset-management platforms.

Their perspectives lead to the same conclusion: value does not come from prediction alone. It comes from connecting the prediction with the systems, decision rights, and people capable of changing the outcome. AI will not predict the future perfectly, but it can give supply chain teams enough time—and sufficiently useful information—to respond before the less costly options disappear.

About the author

Sara Hsu is a clinical associate professor of supply chain management at the University of Tennessee, Knoxville’s Haslam College of Business. Previously, she was an associate professor of economics at the State University of New York at New Paltz. Hsu specializes in supply chain disruptions and supply chain fintech.

Procurement teams invest heavily in the RFQ process. Suppliers submit detailed cost breakdowns across materials, labor, overhead, logistics, tooling, packaging, freight, tariffs, and margin. In advanced environments those breakdowns are checked against should-cost models, commodity indices, engineering estimates, and historical spend.

The data exists. The transparency exists. The analytical effort exists.

The problem is what happens next. Findings are based on responses from 100 procurement and supply chain professionals (approximately 80% manufacturing; 60%+ large enterprise with global operations). Company sizes ranged from fewer than 500 employees to more than 5,000. The sample reflects a broad mix of roles and seniority across cost-intensive industries where supplier cost transparency is critical. The research shows a consistent pattern. RFQs are treated as transactions rather than long-term data assets. Once the business is awarded, the underlying cost structure is often archived in spreadsheets, stored as PDFs, buried in email, or reduced to a single unit price in the ERP. The richness disappears, and the organization resets its knowledge base with every new cycle.

That is the missed opportunity. Companies don’t need more RFQs, they need a system that turns RFQs into continuous cost intelligence.

How companies use cost breakdowns today

The research reveals heavy tactical use and a sharp drop-off when the conversation turns strategic:

• Negotiation/price validation: 85% of respondents
• Benchmarking (supplier/region): 65%
• Design/engineering input: 50%
• Should-cost modeling: 45–50%
• Risk management (T2/T3, geo, tariffs): 40%
• Strategic decision-making (make/buy, reshoring): 30%

Teams are effective in the moment. They use cost breakdowns to challenge assumptions, identify outliers, benchmark quotes, and negotiate better pricing. They spot inflated labor rates, high overhead, outdated material assumptions, and freight or tariff drivers that inflate total landed cost.

Beyond that moment, usage falls off. Sixty to 75% of organizations treat RFQs as one-time events once business is awarded. Intelligence is created in the moment, but not carried forward. The result is strong tactical execution and weak institutional learning.

Maturity matters: where organizations stand

Cost-intelligence maturity can be viewed as a five-phase journey. The research places the large majority of organizations well below strategic maturity:

1. Phase 1: Tactical/documentation (35%). RFQs used for recordkeeping, quoting, and spreadsheets.
2. Phase 2: Negotiation support (40%). Cost data used reactively for price challenges and PPV savings.
3. Phase 3: Structured benchmarking & should-cost (20%). Manual benchmarking, early cost modeling, and some design input.
4. Phase 4: Integrated cost intelligence (5%). Data connected across systems.
5. Phase 5: Predictive & collaborative (~0–1%). World-class: continuous, forward-looking use of cost data.

More than 75% of organizations remain in Phases 1 and 2. They collect data and use it for documentation and negotiation. Few have built the systems and routines that turn transparency into continuous learning.

What organizations gain as cost-intelligence maturity improves

Over more than two decades of working with manufacturing procurement organizations, Advanced Purchasing Dynamics (APD) has consistently observed that as companies move from price-based purchasing to structured cost intelligence, the gap between quoted prices and economically supportable pricing narrows. This improvement comes from better cost visibility, stronger benchmarking, more informed supplier discussions, and the ability to capture, retain, and leverage supplier cost information for better sourcing and business decisions.

Based on APD's experience across manufacturing industries, organizations relying primarily on piece-price comparisons may leave an estimated 8% to 18% gap to economically supportable pricing. Consistent use of supplier cost breakdowns can reduce that gap to approximately 6% to 13%. Organizations using company-specific cost models often narrow it further to roughly 3% to 7%, while industry-based cost models can reduce it to approximately 2% to 5%. The closest alignment typically occurs when buyers and suppliers collaborate using open-book cost information and a shared understanding of cost drivers. These ranges are directional observations from APD's experience, not guaranteed savings, and actual results vary by category, market conditions, supplier dynamics, and data quality.

As organizations advance in cost-intelligence maturity, they typically achieve:

• Closer alignment to economically supportable pricing through greater visibility into cost drivers.
• Stronger negotiations focused on facts rather than quoted prices.
• Better supplier collaboration and a more strategically aligned supply base.
• Earlier identification of cost risks, market shifts, and sourcing opportunities.
• Greater organizational knowledge by preserving supplier cost history.
• Better cross-functional alignment across procurement, engineering, finance, and operations.
• Higher buyer productivity by reusing existing cost intelligence instead of rebuilding analyses.
• Improved visibility into supplier, commodity, tariff, and logistics risks.
• A stronger foundation for predictive analytics, AI-enabled decision support, and continuous improvement.

The greatest benefit is that value compounds over time. Each supplier cost breakdown strengthens future negotiations, benchmarking, sourcing, product design, and risk management. As cost intelligence accumulates, organizations create a continuously improving knowledge base that supports faster, better-informed procurement decisions.

The biggest barriers to strategic use

The issue is not effort, it is infrastructure. Respondents consistently pointed to six barriers:

• Data fragmentation (Excel, PDFs): 80%
• Lack of a central repository: 65%
• Poor ERP/system integration: 60%
• Limited cross-functional alignment: 45%
• Compliance and data-governance gaps: 40%
• Lack of tools and analytics capability: 35%

Cost knowledge often remains personal rather than institutional. As one respondent put it: “There is no place to know supplier history. It is based on what individuals remember.” When insight lives in buyer memory, old folders, and individual spreadsheets, turnover erodes capability, just as decisions grow more complex.

Three patterns we see repeatedly

Strong transparency, no continuity. A mid-sized manufacturer required detailed breakdowns and used them aggressively in negotiation. After award, files were archived by event. The next cycle required rebuilding much of the analysis. The company had transparency; it did not have intelligence.

The ERP disconnect. Teams built rich Excel models covering material, labor, overhead, freight, and duties. When data entered the ERP, everything collapsed into a single price field. Future increase requests could not be compared to original assumptions.

Lost benchmarking potential. A global organization collected breakdowns across regions but stored them inconsistently, different templates, depths, and locations. Scale produced more disconnected files rather than a powerful comparative asset.

From events to systems

Companies do not need more RFQs. They need systems that treat every RFQ as an input into a broader cost-intelligence capability. When cost data is structured and reusable, it supports design-to-cost, supplier development, risk identification, strategic sourcing trade-offs, and better financial planning.

Cost breakdowns can become a shared language for understanding business trade-offs, not merely a negotiation tool.

AI is the next multiplier, but data is the limiter

Seventy percent of respondents see AI-driven cost analysis as the next frontier. Interest in automated supplier comparison, outlier detection, and scenario analysis is real. Those use cases are promising, but AI depends on structured data.

AI is a multiplier, not the starting point. Without standardized formats and connected historical data, advanced analytics cannot deliver strategic value. Organizations that first standardize and integrate cost data will be positioned to benefit; those that skip the foundation will struggle.

What leading organizations do differently

• They standardize cost-breakdown formats so submissions are comparable.
• They centralize cost data so teams can retrieve, compare, and reuse it.
• They integrate cost insights across procurement, engineering, finance, and operations.
• They connect internal cost structures to external signals such as commodity indices, labor trends, exchange rates, tariffs, and freight markets.
• They build internal analytical capability, not just data storage.
• They use cost insight earlier, during design, supplier selection, budgeting, and strategy, not only at final negotiation.

Where to start

Most organizations do not need a dramatic transformation. Practical steps move the needle:

1. Standardize supplier cost-breakdown templates. Consistency beats perfection.
2. Create a centralized, searchable repository. A shared location is better than disconnected spreadsheets.
3. Preserve cost structure in core systems. Avoid collapsing every RFQ into a single unit price.
4. Connect procurement with engineering and finance. Cost data should inform design, margin planning, and strategy.
5. Start with simple analytics. Compare suppliers, track changes over time, identify outliers, and build basic category benchmarks before pursuing advanced AI.
6. Establish governance. Ownership, definitions, and rules keep the system usable.
7. Build toward predictive capability. Once the foundation exists, scenario modeling and AI become far more powerful.

Final thought

Procurement teams already collect the data. The challenge is not collection, it is utilization. RFQs are one of the richest sources of cost intelligence available to the organization.

Companies that treat them as such will build better supplier strategies, make better design decisions, improve risk visibility, and support stronger financial planning. They will not just negotiate better. They will make better decisions.

About the authors

Dr. Sime Curkovic is a professor of supply chain management and Lee Honors College Faculty Fellow at Western Michigan University. His research focuses on supply chain management, sourcing, operations, logistics, and risk management.

Jeoff Burris and Mike Wynn are principals at Advanced Purchasing Dynamics, where they help organizations build cost models, improve sourcing processes, and turn supplier cost data into actionable intelligence.

Executive takeaways

Artificial intelligence is rapidly transforming the way organizations plan, source, manufacture, move, and deliver products around the world. From demand forecasting and procurement optimization to logistics management and customer service, AI is helping companies make faster and smarter decisions. However, as organizations accelerate AI adoption, they are also becoming more interconnected through cloud platforms, digital infrastructure, operational technology, and increasingly complex supplier ecosystems.

This convergence creates tremendous opportunity, but it also introduces a new leadership challenge: trust.

Can organizations trust the data, systems, suppliers, and infrastructure on which AI-enabled decisions depend? Can organizations trust their partners and suppliers are cyber-ready and cyber-resilient? As organizations become increasingly connected, trust becomes the foundation that determines whether AI can securely scale.

Trusted AI depends on the entire supply chain ecosystem

A simple question illustrates the scale of the issue: How many companies help run your supply chain? Increasingly, these organizations do more than move products or provide services. They also generate data, operate digital platforms, and influence the information that AI uses to make recommendations. Enterprise AI is therefore only as trustworthy as the ecosystem on which it depends.

For many organizations, the answer ranges from hundreds to tens of thousands. Each supplier, partner, technology provider, and service organization is increasingly becoming part of a broader AI-enabled ecosystem. As connectivity expands, trust, resilience, and cyber readiness become critical business capabilities rather than technical considerations.

For years, cybersecurity was often viewed as a defensive function focused on protecting systems and preventing attacks. Today, that perspective is no longer sufficient. Organizations seeking to leverage AI at scale must ensure that their data, infrastructure, supplier networks, and decision-making processes are trustworthy. In this environment, cyber resilience is no longer simply a risk management discipline. It has become a business capability that enables trusted AI, resilient operations, and confident executive decision-making. It is about enabling innovation, maintaining operational continuity, and preserving stakeholder confidence.

The challenge is too large for any single organization to solve alone. Governments, technology providers, infrastructure operators, and private-sector enterprises each play a vital role in building trusted and resilient supply chains. This makes public-private collaboration increasingly important.

Supplier cyber readiness strengthens enterprise AI readiness

This challenge is particularly important across supplier ecosystems, where thousands of small and medium-sized businesses often support critical operations while operating with fewer cybersecurity resources than their larger customers. As enterprises increasingly rely on AI-driven decisions, the cyber readiness of these suppliers becomes a direct contributor to enterprise resilience. Strengthening supplier cyber readiness is therefore not simply a compliance exercise; it is an investment in trusted AI, operational continuity, and long-term competitiveness.

Governments establish policy and incentives that encourage secure innovation. Technology providers embed security, identity, and explainability into the platforms organizations rely upon. Infrastructure providers strengthen the resilience of the digital backbone that supports modern commerce. Large enterprises can use procurement and supplier engagement to raise cyber readiness across entire ecosystems.

The executive questions that will shape trusted AI adoption

As AI adoption accelerates, executive leadership teams face several critical questions. Where should organizations invest first to safely accelerate AI adoption? What role should trust play in supplier selection and ecosystem management? What infrastructure capabilities are required before businesses can confidently rely on AI at scale? How should cyber readiness influence supplier selection and ongoing supplier performance?  What happens when a critical supplier lacks the cyber capabilities required to support trusted AI? And, perhaps most importantly, how should leaders determine when AI can make decisions autonomously and when human judgment must remain in the loop?

Organizations that answer these questions effectively will be well positioned to compete in the next era of global business. Those that fail to address trust and resilience may find it increasingly difficult to scale AI initiatives, manage risk, and maintain customer confidence.

Looking ahead, the organizations that lead in the age of AI will not necessarily be those deploying the most advanced algorithms. They will be the organizations that build the strongest foundation of trust across their data, technology, operations, and supplier ecosystems. They will recognize that cyber resilience enables innovation, supplier readiness strengthens enterprise readiness, and collaboration across the public and private sectors creates competitive advantage.

In the age of AI, trust is no longer simply a security objective. It is a business strategy.

About the author

Sasha Pailet Koff, is managing director of the Cyber Readiness Institute. She is also founder and president of So Help Me Understand and a former senior vice president of digital supply chain at Dell. 

Executive takeaways

Retailers are not short of returns data. They can rank categories, brands, products and customers by return rate; forecast reverse-logistics volume; and estimate the cost of processing an item after it comes back. Yet the metric at the center of most returns conversations remains too blunt.

Saurabh Bahree

The National Retail Federation projected that the value of retail returns would reach $849.9 billion in 2025 and estimated that 19.3% of online sales would be returned. At that scale, a percentage-point change can affect transportation, inspection labor, recovery speed, markdown exposure and working capital. But not every percentage point means the same thing.

The raw rate can point management in the wrong direction

A fitted dress, a standardized phone charger and a fragile kitchen appliance do not carry the same inherent return risk. Categories differ in fit uncertainty, product complexity, price, channel mix, seasonality and customer expectations. Comparing them solely on raw return rate is like comparing hospitals solely on readmissions without considering the patients they treat.

This can create a management error. Research shows return propensity varies across products and can be predicted from product characteristics. A high-return category may be near its expected level, while a lower-return product may underperform comparable items because of defects, misleading content, fulfillment errors or damage. Raw rates reveal volume, not abnormal performance.

Return prediction does not solve this by itself. Models estimate whether an order is likely to come back, as ASOS researchers have demonstrated using GraphReturns. Most applications target individual transactions, customer messaging or volume forecasts. Healthcare offers another model: aggregate probabilities into expected outcomes and compare them with actual performance. Returns teams could apply this risk-adjustment logic.

Borrow a measurement discipline from healthcare

The Centers for Medicare & Medicaid Services does not assess hospital readmissions using raw counts alone. Its Excess Readmission Ratio compares predicted with expected unplanned readmissions after adjusting for patient case mix. The objective is not to excuse poor outcomes, but to avoid treating different underlying risk populations as though they were identical.

Retail can adapt that logic. For a product, supplier, category or operating node, calculate an expected return count by summing the pre-purchase return probability of every transaction in the group. Then compare observed with expected merchandise returns:

Excess Return Ratio = Observed merchandise returns / Expected merchandise returns

A ratio near 1.0 indicates performance broadly in line with the selected benchmark. A ratio materially above 1.0 identifies more return demand than the model expected. It is an investigation signal, not a finding of fault. Confidence limits and minimum-volume thresholds should prevent small or statistically insignificant deviations from triggering action.

Consider two hypothetical groups with 10,000 purchases each:

Illustrative comparison; figures are hypothetical.

The raw-rate dashboard prioritizes dresses. Risk adjustment points to appliances. If appliance returns also cost more to transport, inspect and recover, the financial priority becomes clearer still.

Risk-adjusted monitoring is not new. Manufacturing researchers have applied risk-adjusted control charts to warranty claims to detect emerging product-quality problems, while healthcare has long used observed-to-expected measures. The opportunity is to apply that discipline to e-commerce return demand across commercial and supply-chain owners.

The benchmark must follow the accountability question

The hardest decision is not which algorithm to use. It is which variables the benchmark should treat as legitimate context.

For group g, expected returns are the sum of its purchase-level probabilities: Eg=ipi. Its Excess Return Ratio is ERRg=Og/Eg, where Og is observed returns. A value above 1 indicates more returns than expected.

The variables used to estimate pi must match the accountability question. A supplier benchmark might adjust for category, price band, fit exposure, season, channel and market, but not supplier-controlled factors such as workmanship, sizing consistency or defect history. Otherwise, persistent poor performance raises the benchmark and begins to look normal. Apply the same rule to fulfillment and product content: adjust for conditions outside the owner’s control while leaving its own performance visible.

Expected also does not mean unavoidable. A model describes what is likely under specified assumptions. Only investigation can determine whether excess demand is preventable and who owns the cause.

Give excess return demand a budget

Measurement becomes operational when it creates a decision rule. Google popularized error budgets in site reliability engineering: teams define an acceptable level of service failure, then use budget consumption to decide whether releases can continue or reliability work must take priority. Retailers could apply an excess-return budget above the risk-adjusted benchmark.

A category, supplier or launch would consume that allowance when observed return demand credibly exceeds expectation. A slow, persistent burn could create an investigation ticket. A rapid burn associated with a launch, supplier batch or fulfillment node could trigger inventory inspection, content correction, containment or a replenishment pause.

This is not a target of zero returns. Legitimate returns reduce purchase risk and support conversion. The budget instead creates a shared tolerance for abnormal demand and a common escalation rule across merchandising, quality, e-commerce, finance and supply chain.

Close the signal, not just the transaction

When a threshold is breached, the organization still needs a cause-closure process. A practical sequence is to classify the suspected cause and confidence, attribute excess units and cost, assign the accountable owner, apply a cause-matched intervention, protect legitimate customers, and verify that performance improves.

The intervention should follow the evidence. A sizing problem belongs with product and supplier teams. Misleading imagery belongs with content. Damage may belong with packaging, fulfillment or the carrier. Customer friction should not be the default response to failures created upstream.

Leaders can test the approach in 90 days. Select one high-volume category, define the accountability question, agree on permissible adjustment variables, and compare raw with risk-adjusted rankings. Choose two or three statistically credible excess clusters, investigate them and track corrective action through an effectiveness check. Monitor conversion, complaints and defective-product indicators alongside return outcomes.

Retailers have spent years improving what happens after an item enters the reverse network. The next step is to improve how return demand is measured before assigning accountability and deciding where to intervene. The raw return rate describes workload. A risk-adjusted excess return measure can direct action.

About the author

Saurabh Bahree is an enterprise e-commerce and digital transformation leader with more than 17 years of global experience in consulting, program leadership and technology-enabled business transformation. He has led complex, cross-functional programs spanning digital commerce, platform operations and enterprise technology. He holds an engineering degree and an MBA and is a certified PMP and SAFe practitioner.

What’s missing is execution.

Agentic AI represents the next step in supply chain evolution, shifting systems from passive insight generation to active decision support and, in some cases, autonomous execution. Instead of simply surfacing disruptions, these systems can recommend or even take actions across planning and execution workflows, helping organizations respond faster to volatility, reduce manual intervention, and improve overall performance.

In this session, we’ll move beyond the hype and examine:

• How agentic capabilities are being applied today
• Explore what differentiates true “decision-oriented” AI from traditional analytics
• Where early adopters are seeing measurable impact
• How to move from insight to execution without overhauling your tech stack

Whether you’re experimenting with AI or looking to scale beyond pilot programs, this discussion will provide a practical framework for turning insight into action.

FEATURING:
Ben Dussault, Director, Product Management, Supply Chain, Anaplan; Dipti Gupta, Head of Product Strategy & Solutions, Logility and Caique Zaniolo, SVP of Product, Tradeverifyd and Adrian Wood, DELMIA Strategic Business Development Director, Dassault Systèmes