And yet, across the industry, a consistent pattern is emerging: many organizations are approaching AI as a technology initiative.  And that’s a mistake.

The Technology Trap

In many organizations, AI efforts follow a familiar path:

• evaluate tools

• run pilot programs

• test use cases

• deploy isolated solutions

This approach feels logical. It mirrors how many prior technologies were adopted. 

But AI behaves fundamentally differently from prior technologies — in its speed of evolution, breadth of impact, non-deterministic nature, and rapid rate of obsolescence (among other reasons).  AI does not simply support existing processes. It changes how work is performed, how decisions are made, and how value is created. It affects not just systems, but the structure and behavior of the institution itself.

Why does this distinction matter?  Because organizations that focus primarily on tools and pilots often experience poor outcomes with AI:

• fragmented initiatives

• higher error rates (e.g., hallucinations)

• limited scalability

• unclear value realization

• growing complexity and risk

The issue is not the technology.  The issue is how the organization is aligned to use it.

AI as a Capability Transformation Problem

The organizations that are beginning to realize meaningful value from AI are not treating it as a technology layer.  They are treating it as a capability transformation problem.  This means asking a fundamentally different set of questions.

• What data foundations are required to support those decisions?

• How should workflows be redesigned to incorporate AI outputs?

• What governance and quality systems are needed to ensure reliability and trust?

• What skills and roles are required to operate in this new environment?

• How should decisions be made differently in an AI-enabled organization?

AI does not succeed because a model works.  AI succeeds when an institution is able to consistently and reliably integrate it into operations.

A Pattern That Has Repeated Across the Industry

Though AI is unique, healthcare and life sciences industry veterans may see some similarities with prior technology-driven transformations.

Combinatorial Chemistry & High Throughput Discovery

One of the earliest transformations in modern drug development came with the rise of combinatorial chemistry and high-throughput screening.  The promise was compelling: dramatically expand the number of compounds that could be generated and tested, accelerating the discovery of viable therapeutic candidates.

But success did not come simply from introducing new laboratory technologies.  It required fundamental changes to how research organizations operated:

• new experimental designs and screening methodologies

• integration of automation technologies with laboratory workflows

• development of data systems capable of managing and interpreting vastly larger volumes of experimental results

• closer coordination between chemistry, biology, and computational teams

Organizations that treated combinatorial chemistry as a discrete scientific innovation often struggled to realize its full value.

Even successful adopters did not fully realize the promised value. While combinatorial chemistry expanded the scale of compound generation and screening, it did not proportionally accelerate drug discovery.  In many cases, the limiting factors were not the technologies themselves, but the surrounding systems required to translate scale into value. Biological understanding did not advance at the same rate as compound generation, data environments struggled to manage and interpret the resulting volume of information, and organizational models remained fragmented across chemistry, biology, and computational disciplines. As a result, organizations were able to generate more data and more candidate compounds, but not necessarily better outcomes.

The lesson was clear: scaling one component of the system without redesigning the broader operating model limits the impact of even the most powerful innovations.

Digitizing Clinical Research

In the early transition from paper-based trials to electronic data capture, success did not come from simply introducing new software.  It required new processes; new data standards; new operating models; and cross-functional coordination across scientific, technical, and operational teams.  The organizations that treated it as a system implementation struggled.  Those that treated it as an institutional shift succeeded.

Health Analytics

As the healthcare industry adopted electronic medical records, their data became computable.  So we saw the emergence of investments focused in data and analytics platforms supporting hospital operations, value-based care programs, and more.  But analytics platforms alone did not produce impact.  Real transformation required:

• building dedicated analytics capabilities that were adept at harnessing this data

• aligning clinical and operational leadership on effective measurement strategies

• adjusting care pathways, processes, and EMR functionality to leverage the insights

• redefining decision-making processes based on the new insights available

• establishing governance and prioritization frameworks protecting patients and the judgment of clinicians.

Analytics became valuable not because it existed, but because it was institutionalized.

The AI Era

AI is now following the same pattern, but at far greater speed and scale.  In areas such as clinical research, real-world evidence, regulatory processes, and healthcare delivery, AI is already reshaping how work is performed, how quickly it can be performed, and what level of precision is possible.  In CROs, for example, AI is shifting value away from labor-based models toward AI-enabled capabilities, fundamentally changing how services are delivered and evaluated.

At the same time, organizations are struggling with competing pressures: speed vs. governance, innovation vs. quality, and adoption vs. control.  Regulators are also signaling expectations that reinforce this shift — emphasizing data governance, lifecycle management, and multidisciplinary expertise as core requirements for AI-enabled systems.  These are not technology challenges; they are institutional challenges.

What It Actually Takes to Scale AI

Organizations that succeed with AI tend to converge on a common set of capabilities.

1. Clear Strategic Alignment.  AI initiatives are directly tied to measurable business and clinical outcomes, not isolated experimentation.

2. Data as a Managed Asset.  Data is treated as a product — governed, curated, and designed for reuse — rather than an output of individual systems.

3. Redesigned Workflows.  Processes are re-engineered to incorporate AI outputs into real decision-making, rather than layered on top of existing workflows.

4. Governance and Quality Systems.  AI is managed within structured frameworks that ensure reliability, compliance, and trust.

5. Integrated Operating Models.  Cross-functional teams align business, scientific, and technical perspectives to deliver and sustain capabilities.

6. Continuous Adaptation.  Organizations are designed to evolve as technologies change, rather than relying on static roadmaps.

Moving Beyond the Pilot Phase

Many organizations are still focused on identifying the “right” AI use cases.  But that is not the limiting factor.  The limiting factor is the ability to build the institutional capabilities required to scale those use cases.

Pilot programs are useful for learning.  They are not sufficient for transformation.  The organizations that move ahead will be those that shift their focus from “Where can we use AI?” to “What must we become to operate effectively in an AI-enabled world?”  The real opportunity is not simply to do the same work faster.  It is to transform the fundamentals:

• redesign clinical research

• improve the precision of care

• accelerate therapeutic development

• fundamentally change how organizations operate and the associated costs

The organizations that succeed will not be those that adopt AI tools the fastest.  They will be those that build the capabilities required to make AI a core part of how they function.

AI is not a technology strategy issue. It is a transformation of how institutions create value.  And the organizations that recognize this — and build accordingly — will define the next era of healthcare and life sciences.

At just two pages, the guidance is concise—but it sends a clear signal about how regulators expect artificial intelligence to be developed and governed across the drug development lifecycle.

For an industry that is rapidly experimenting with AI, the document is both reassuring and clarifying. Let’s explore several of the key concepts they highlight.

Natural Synergies with the Status Quo

For experienced industry leaders, several elements of the principles should be encouraging:

1. Acceptance of the AI opportunity. If the FDA’s growing portfolio of AI-enabled medical devices and its 2025 AI guidance document were not enough to convince AI skeptics of regulatory readiness, this document reinforces the message. The agencies explicitly acknowledge that AI can help modernize the drug development paradigm, promoting innovation, reducing time-to-market, strengthening pharmacovigilance, and even reducing reliance on animal testing through improved prediction of toxicity and efficacy.

2. Alignment with risk-based practices. The principles strongly reflect the risk-based ethos already embedded in regulatory science. This mirrors the operating philosophy used across areas such as monitoring, validation, quality systems, and clinical oversight. The industry already understands these frameworks, so the challenge ahead is extending them into AI-driven capabilities.

3. Leveraging established validation approaches. One of the most common questions surrounding regulated AI is: how do you validate systems that are not fully deterministic? The answer suggested by the principles is reassuringly familiar. The agencies emphasize practices already central to regulated technology environments: robust design, data governance, engineering best practices, documentation, and quality management systems among other topics. These expectations are fully consistent with how technology-driven solutions are already developed and validated in regulated environments.

This is all good news, as it further confirms what many industry insiders have been expecting.

The Role of Standards

Despite its brevity, the document references “standards” five times. Historically, US and EU regulators have relied heavily on standards to streamline their complex and safety-critical activities. Standards create a shared language that simplifies compliance and inspection for both sponsors and regulators. The challenge is that standards require stability, and stability is not a near-term feature of today’s rapidly-evolving AI landscape.

Both agencies are encouraging industry collaboration to further develop and implement these principles. Could we see the formation of an organization similar to CDISC, but focused on AI standards in drug development? Perhaps.

Having been involved in the formation of CDISC, I remember firsthand how complex that effort was. I would also argue that CDISC’s data domains — mainly limited to information structured in research protocols and statistical analysis plans — are more constrained than today’s AI models and use cases. Practical industry applications of AI today routinely feature electronic medical records, genomics and molecular data, patient registries, claims and prescription data, lab results, and other real-world data (RWD) and real-world evidence (RWE). Many of AI’s most transformative applications will depend on combining these heterogeneous data environments, which introduces significant challenges for standardization.

Industry Readiness

For many industry leaders, the biggest question raised by the document is simple: is our organization actually ready?

Many life sciences companies are still pursuing AI opportunistically through pilot projects, experimentation, and point-solution tools. While these activities can generate useful experiences, they do not inherently cultivate the disciplined operating environments regulators are describing. They also don’t protect the organization from the risks in AI permeating every system and vendor within the enterprise.

In practice, three capability gaps tend to emerge:

Lack of an Organizing Framework. The guidance repeatedly references the importance of “context of use”, including role, scope, model risk, and quality procedures. Establishing this clarity requires an organizational framework for AI governance aligned to the company’s quality system. Many organizations have not developed such a framework and lack strategies for ensuring adherence to it.

Compliance & Inspection Evidence. The principles closely mirror areas regulators traditionally examine during inspections, including data provenance, process adherence, audit trails, data protections, and documentation. Ironically, these are often the areas that receive less attention during rapid technological exploration. In AI environments, weaknesses in these areas do more than create regulatory risk — they frequently undermine the quality, reliability, and performance of the AI solutions themselves.

AI-specific Processes and Expertise. The guidance emphasizes the need for multidisciplinary expertise that spans both AI technology and the scientific or business context in which it is used. Many organizations have yet to create formal mechanisms that support this type of AI-informed collaboration. Equally important, AI introduces new lifecycle management challenges, including model monitoring, drift detection, retraining / tuning, migration, and evolving cybersecurity risks. In practice, many companies simply do not have sufficient internal AI expertise to fully address AI competencies that deliver engineering best practices, interpretability, explainability, performance controls, transparency, generalizability, and robustness.

As AI adoption accelerates, these gaps should be a call to action.

Leaning into Next

The encouraging news is that these challenges are not insurmountable. Proven governance, quality, and technology management approaches can address them—when applied deliberately and supported by the right expertise.

For leadership teams, several strategic questions become increasingly important:

• Who owns our AI readiness and what are the specific goals and outcomes we are pursuing?

• How are we orchestrating change management, cross-functional alignment, and standards development around AI adoption?

• How are we prioritizing improvements to our quality system, data environments, and development practices so that concepts like “context of use” are reflected in daily operations?

• Where should we supplement internal teams with external AI expertise to ensure regulatory and safety considerations are properly managed?

The answers to these questions will not emerge overnight. And AI capabilities will continue to evolve rapidly for the foreseeable future. What life sciences organizations need is not a one-time solution, but a sustainable way to bring structure and discipline to what often still feels like the “Wild West” of AI.

The organizations that succeed will be those that balance innovation with governance—moving quickly while building the operational discipline regulators clearly expect.

How prepared is your organization for that shift?

The rapid expansion of the American West created a landscape defined by opportunity, competition, and uncertain rules. Today, many life sciences leaders may feel they are navigating a similar frontier with artificial intelligence.

The AI territories for industry transformation look both extensive and attractive: drug design, trial simulation, data cleaning, regulatory writing, quality monitoring, employee self-service, and many more paint a vibrant picture of AI-enabled businesses that deliver better therapies faster. But the day-to-day experience can sometimes feel like a showdown of competing concerns:

1. Velocity vs. Planning. Frontier AI companies are releasing product updates at an unprecedented pace – measured in months, not years. Combined with the continuous influx of new AI-enabled players in the market, corporate AI roadmaps become outdated almost as quickly as they are authored.

2. Use vs. Proliferation. Industry leaders want to encourage the advancements and efficiencies that AI can bring (e.g., faster recruitment, streamlined study operations, cleaner clinical data). But AI growth brings practical challenges: proliferating data copies, rising compute costs, tool sprawl, and increasing compliance risk.

3. Agility vs. Governance. From the board room to the break room, leaders and workers are embracing AI — from tools that summarize clinical trial documents to copilots that draft regulatory submissions. As organizations struggle to make decisions and implement new solutions, research shows a large portion of employee adoption is “shadow AI” use of ungoverned and unsecured products.

4. Market Enthusiasm vs. Value Creation. The market noise around AI is deafening, and expectations are frequently misaligned to reality. Leaders struggle to balance the imperative to move with a disciplined strategy that focuses on the most attractive and measurable value creating opportunities.

5. Duplication vs. Standardization. Practically every IT vendor in every organizational unit is bringing AI to market. Leaders responsible for quality deliverables, compliant operations, and controlled costs are grappling with a deluge of independent models offering similar capabilities, differing reliability, and growing costs.

6. Build vs. Buy Decisions. Provisioning AI capabilities is not a simple “build vs. buy” equation, and thanks to AI, tailored solutions don’t necessarily require manual coding. Though AI tools can be purchased “off the shelf,” more impactful AI solutions that deliver higher value for any organization are often some combination of both.

7. Change vs. Quality. Business transformation with AI requires workforce education, process re-engineering, and evolving roles. In the absence of disciplined programs focused on effective transitions, the corresponding AI capabilities frequently face problems in accuracy, scalability, compliance, trust, adoption, and ROI.

How the West is Won

The AI frontier may feel chaotic, but organizations that succeed are not improvising. They operate with a clear set of principles that guide technology, governance, and change.

• One Program, not Many Projects. Scalable AI capabilities with measurable value creation do not emerge naturally from grass-roots efforts. As opposed to waiting to uncover small-scale wins, create an orchestration center that can drive a more managed approach across the business. It is faster and cheaper to create the right investments once.

• Strategy, not Roadmaps. The AI landscape will not stabilize any time soon. Instead of trying to maintain static adoption roadmaps that are always out of sync with the market, pivot to focusing on clear strategies, technical principles, and change-ready blueprints that guide the ongoing selection and migration of AI capabilities over time as the business and technology evolve.

• Data, not Tech. Most risks in AI projects have nothing to do with the AI itself. Experience shows that most organizations do not have data assets designed to perform well with AI (e.g., semantically consistent research data; managed master data; data architectures that effectively blend structured and unstructured data across scientific, operational, and financial domains). Instead of evaluating the next great AI tool, direct teams to carefully assess the quality, consistency, management, and trustworthiness of the organization’s key data that would be used by any AI model selected.

• Quality System, not Documentation. The organization’s obligation to cultivating and using high-quality information systems – whether regulated or not – grows with AI. When systems are not deterministic, technology is changing, and processes are being re-engineered, it is critical to address quality as a comprehensive framework guiding effective, safe, and compliant business operations.

• Enablement, not Deployment. To unlock sustainable value creation with AI, adoption is usually a transformational journey. The goal is not to give employees more software. Focus programs on enabling employees to perform different processes, empowering them to co-create the future, leveraging new skills, and freeing capacity from manual labor that machines can undertake.

The AI frontier in life sciences will not normalize anytime soon. But history suggests that periods of technological expansion eventually reward the organizations that combine ambition with discipline. In the AI era, the winners will not simply adopt new tools faster — they will build the strategies, data foundations, and operating models that allow those tools to create lasting value.

That’s the top-of-mind question for leaders across many service sectors. Compared to prior emerging technologies, AI is somewhat unique: in addition to its exceptionally high velocity and broad applications, the disruption is felt heavily among skilled knowledge workers. Whereas technology innovations often empower and amplify this workforce, AI also commoditizes the delivery of knowledge work products and processes. In life sciences, contract research organizations (CROs) are one of the leading intersections of this workforce / AI transformation.

The modern CRO industry has evolved over the past 40 years to become a diverse $100B ecosystem of service providers responsible for >70% of all clinical research conducted in the world. But the fundamental value propositions historically offered by CROs – operational efficiencies, scalable resource models, cost arbitrage, shared IT investment, and access to scientific expertise – are being re-characterized by the rapid infusion of AI. As these economics shift, CROs face an imperative to adapt both their business models and operations to remain competitive and valuable business partners.

Pathways to Smarter Clinical Research

Though a full-scale disruption of the CRO industry is unlikely, a transition in service delivery approaches and associated value creation – from labor-oriented models towards diversification into AI-enabled products and services – is already underway. Market viability of AI solutions has already been proven in areas such as:

• Site selection and start-up

• Patient recruitment and retention

• Adaptive trial designs & synthetic control arms

• Clinical / RWE data management and quality

• Adverse event detection and management

• Imaging, wearables, and devices

• Clinical monitoring and protocol deviation detection

• Document automation solutions

• Programming and TLF automation

• Real world evidence analyses and NLP

To date, the AI investment strategy of CROs in these areas has been stratified by market tier.

• Large CROs are building proprietary platforms, making notable investments alongside global technology players and partners to deploy large-scale data assets and extensible agent platforms. These complex systems require sizable capital investments and expertise, and while they offer compelling capabilities, they will also require sustained engineering as the underlying technologies continue to evolve.

• Small-to-medium CROs, lacking the capital of their larger peers, are pursuing 3rd party vendors to provision off-the-shelf AI capabilities, though such tactics introduce a wide array of risks (e.g., vendor dependency, parity with competitors, operational misalignment, cost duplication, data proliferation) that can potentially undermine long-term value creation.

For most of the CRO industry, a balanced approach to build-vs.-buy is likely the best path forward. The introduction of value-creating AI solutions does not need to be capital intensive – on the contrary, most technical capabilities can be provisioned at commodity price points. By blending affordable off-the-shelf software with tailored configurations of supportable AI, CROs can unlock the transformative value of AI without the need for lengthy, complex system engineering projects and high operating expenses.

Business Model and Market Dynamics Implications

Given the capabilities that AI offers, performance expectations for both sponsors and CROs will change. Though geographic reach, operational scale, and clinical expertise will still matter, sponsors will evaluate CROs on their ability to deliver AI-derived advantages such as faster performance (e.g., recruitment, data review, database lock), more efficient operations (e.g., study monitoring, programming), higher quality (e.g., lower error rates, automated compliance and quality monitoring), and more precise results (e.g., modeling, targeting, diagnostics, segmentation). As such, leadership teams and their boards need to be prepared to proactively address strategic questions such as:

• Business model: Every CRO needs a strongly aligned strategy between their relationship model (FSO, FSP, hybrid), their differentiated capabilities (therapeutic expertise, accessible network, data, technology, etc.), and their AI-driven value delivery capabilities (speed, quality, cost, risk, precision, etc.). What makes your CRO a truly unique value creation driver in an AI-enabled market, and how will you best enable and monetize it?

• Investor attractiveness: In an sector historically characterized by high investor appetites, CROs that do not demonstrate strong AI and data competencies may be seen as less competitive and garner lower valuations. How will you align to investor interests to maximize future capitalization cycles?

• Margins: Some CROs may require an increase in highly skilled labor (e.g., scientific / technical expertise delivering novel research designs involving synthetic control arms, digital twins, and trial simulations). How will you protect and improve margins with higher labor costs, especially if lower-cost resources are displaced?

• Continuous learning and change: For the foreseeable future, the rate of AI-driven advancements and change will remain high, presenting the CRO industry with a continuous stream of opportunities. How will you stay agile and embrace new AI opportunities effectively while avoiding unnecessary expenses and technical debt?

• Consolidation: Given the market velocity, both CROs and AI companies seeking to accelerate their growth will see inorganic options as a competitive imperative. Consolidations will change both CRO competitive dynamics and vendor technology strategies. How will you monitor and adapt to these landscape changes, and which ones would trigger adjustments to your strategy?

Practical Considerations for AI Enablement of CROs

When considering internal investments, “AI everywhere” will not create high performance. Every CRO needs an AI strategy that prioritizes the fewest, most important operational and competitive capabilities that will drive the highest value creation. The AI strategy should be deeply connected to the corporate strategy, focusing on performance metrics tied to explicit organizational goals. Examples include margin expansion, utilization improvement, avoiding unnecessary human capital growth, study enrollment optimization measures, improved turnaround times, improved quality, and higher win rates. For most CROs, that strategy will need to address the follow 7 areas.

1. Service line transformation. Existing service lines will evolve – and new service lines will emerge – to accommodate opportunities that AI introduces. Existing service lines will unlock efficiencies in core business processes such as recruitment, data review, scientific writing, biostatistical programming, and billing. Newer services reflect growing demand for bespoke client-driven AI applications – novel study modeling, disease modeling, and RWE data curation such as patient registries, for example. One key to unlocking the strategic value of service line transformation is laser-like clarity around a firm’s most compelling business, scientific, and technical differentiators that can be translated into AI assets.

2. Relationship models. Sponsor relationships with CROs can take the form of any combination of T&M projects, Full-Service Outsourcing (FSO), and Functional Service Provider (FSP) agreements. Hybrid approaches and FSP relationships tend to be more cost effective for clients, though smaller organizations such as some biotechs may require FSO providers’ infrastructure. Though AI can amplify CRO margins of all three options, it also places downward pricing pressure on labor-based contracting even under milestone-based payment models. Outcomes- or performance-based contracting can provide margin relief by leveraging AI to drive value creation provided adopting organizations ensure strong management of the contract terms.

3. Data sharing readiness. Ensuring that client agreements provide data reusability terms to support the CRO’s AI-enablement goals is now vital. For some, changing contract terms will seem commercially risky. But the alternative is much riskier – an organization based heavily on manual labor competing in an industry at least partially automated by AI. Numerous precedents exist for safely sharing anonymized data to support research advancements. And not all data is equally sensitive; operational, quality, and performance related data offer a wealth of innovation opportunities without introducing scientific IP, patent, or patient data risks. Accommodations may be needed for individual customers that may be unwilling, though offering clear incentives for sharing may be sufficient to produce change.

4. Data asset development. Though market excitement is often directed at advancements in AI methods, success and value creation with AI is most associated with data. In some areas such as clinical development, the historical approach to data management has been largely study- or program-based; cultivating reusable data assets has not been a high priority. A CRO may have a strategy that focuses on any combination of preclinical, clinical, financial, commercial, and operational data. Regardless of the data mix, competencies such as data product engineering, quality, curation, governance, pedigree, ownership, and performance-aligned storage are needed to ensure AI success. And controlling data proliferation – the uncontrolled spread of data copies across AI tools and storage locations – is critical for protecting information security.

5. Real-world data partnerships. Many AI innovation priorities for CROs will not be accessible without better access – either directly or through partners – to real-world data. AI-enabled patient recruitment, for example, is most effective when connected to patient medical records. Advanced work with precision diagnostics and therapeutics involves leveraging richer data formats – sequence data, imaging data, waveforms, streaming device data, and others. CROs do not need to in-source all of these data capabilities, but they do need consistent, scalable channels for provisioning them so they can develop reusable IP. Therapeutic specialization offers unique value accelerator; organizations establishing unique data niches within specific therapeutic areas or modalities gain additional commercial and computational leverage.

6. Enterprise architecture. For AI to function and scale properly, the IT strategy, blueprint, and underlying systems need to be intentionally designed, aligned, and managed. Every IT vendor in the CRO’s enterprise is pitching AI capabilities. A clear plan for governance and rationalized use is required for scale and cost management. CROs need standardized designs and solutions for addressing common AI scenarios such as access to external models, delivery of tailored AI solutions, authentication, and data controls. As AI use cases increasing focus on workflow automation, interoperability within and across enterprise systems becomes an imperative. As many industry applications were never engineered as enterprise-grade systems, organizations may need some tailored workarounds to accomplish their integration goals. And given that the AI technology landscape will continue to evolve rapidly over the next five years, every CRO needs to plan for continuous change – leverage what is available but avoid compounding technical debt.

7. Analysis. In its traditional form, biostatistics may now be a less viable independent service line. While still core to research execution and interpretation, traditional bioanalytical methods (e.g., descriptive and inferential statistics, Bayesian approaches) represent only a fraction of the computational techniques now being applied in supporting drug discovery and development. In addition, the implementation of those traditional methods (e.g., programming, testing, deliverable creation) can be delivered by modern AI models, creating a barrier to high-margin service monetization. However, the human expertise in the appropriate selection, use, and interpretation of AI and bioanalytical methods is not a commodity, creating tremendous opportunity for leveling up the contribution margin of analytical talent.

From a leadership perspective, perhaps the biggest consideration to AI adoption in life sciences – at least in terms of speed and impact – is the industry environment itself. Research variability, scientific complexity, regulatory obligations (both real and perceived), and justifiable concerns over patient protections have often created conservative, less agile cultures. No one wants to be at the source of an audit failure, a delayed study close, or loss of IP. At the same time, many success stories now exist, regulators are actively participating in the journey, and life sciences has successfully adopted many emerging technologies over the past three decades. Leaders need to acknowledge the legitimate nature of these concerns while also delivering confidence-inspiring plans for navigating into the future.

Coming of Age

CROs that move now to embrace the age of AI will be tomorrow’s market leaders and high-value investment targets. But AI adoption is a learning journey for every organization, including CROs. For leaders looking to accelerate their AI progression, three tactics show up repeated as best practices among high-performing organizations:

1. Conduct a comprehensive, realistic assessment of your organization’s readiness. Given the issues presented here, it is not surprising that most CROs are not ready to fully embrace AI. Leaders need to be prepared to critically evaluate the maturity of processes and competencies related to:

• Business process standardization and quality control

• Data and information storage, control, quality management, access, and use agreements

• Solution development lifecycles, quality frameworks, and auditability

• IT governance, including vendor assessments and data use terms

2. Develop specific, measurable value creation goals aligned to your corporate strategy and growth opportunities.Avoid searching for perfect AI applications. Instead, look for current business problems where AI can help. Prioritize measurable constraints such as situations where you:

• Lose margin due to manual work

• Struggle with people capacity

• Have people allocated below their skill level • See unmet service quality, capability, or speed

• Find manual work producing delays

• Cannot easily access knowledge resources

3. Launch, don’t pilot, AI solution programs to gain the right expertise. Pilot programs are useful when testing whether a technology works, but they are much less useful when pursuing value creation goals, process re-engineering, or business transformation. The core question is not whether the technology is a good fit for the business today; the core question is about what the business can look like tomorrow. We know the technology works; the goal is creating a more attractive business with it. And the goal is within reach.

Current Design Context

The term “computer” was originally assigned to human beings – in particular, individuals that compute things for others. Acknowledging the routine work associated with computing answers, Charles Babbage developed the concept of a mechanical programmable computer (some people may recall that GameStop’s original name was Babbage’s). Electromechanical computers followed in the 1930s-1940s, with Colossus and ENIAC becoming the first electronic digital programmable computers. Through the rise of transistors, integrated circuits, and today’s most powerful processors and architectures, one design principle has held constant: humans would be the main users of this technology.

But AI is challenging long-standing hardware and software architectural and design concepts. I’ve compiled a list of the biggest design-related shifts I see emerging from the evolving capabilities in AI.

Computers as First-Class Users

To many people, the most fundamental and obvious shift broadens the intended developer and user community from humans to include machines. Though information systems have participated in workflows for decades, AI-empowered systems feel like a different design metaphor.

• AI is a primary developer and user. Software will no longer primarily be built just for people to click and type. Instead, AI programs themselves are becoming priority users of digital systems, fundamentally changing how everything is designed from the ground up. Systems will become optimized for autonomous AI interactions, and new software applications will be built by software.

• AI becomes inseparable from software engineering. The process of building software is being transformed by AI. AI tools will assist developers in generating code, creating comprehensive test suites, discovering obscure edge cases, and predicting potential build failures. The role of human software developers will evolve from writing every line of code to becoming guides and supervisors for AI tools, focusing on designing the overall system, defining complex problems, and overseeing AI-generated solutions. This of course changes the design of integrated development environments (IDEs).

• Autonomous AI agents perform routine work. Future software will be powered by intelligent programs, or AI agents, that can observe, reason, and act independently. These agents can plan their own steps and execute complex tasks without needing constant human oversight. Designs will need to reflect this new operating autonomy. And AI will move into IT operations, leading to self-healing systems. AIOps platforms will use machine learning to monitor software performance in real-time, predict potential issues, guard against cybersecurity threats, initiate maintenance, and even automatically resolve problems without human intervention.

• Data gets designed and aggregated for machines: The way information is stored and organized is changing. Data architectures will be built considering the need for AI to process and interpret data efficiently, rather than just for humans to read or analyze. Vector databases and other emerging models allow AI to find things based on their meaning and similarity, not just exact keywords.

Software Value Flows from Data

For most of the digital computing age, software value derived from the code – the algorithms and associated functionality they deliver. But as code becomes an automated commodity, AI considerably alters that value equation:

• Data is a differentiator. General-purpose AI models are rapidly becoming a commodity. But specialized models – trained and tuned to specific tasks using highly curated data – are not a commodity. And that means organizations that invest in these data assets – and the software solutions that generate that data – will enjoy more upside.

• Data timeliness gains prominence. AI models will increasingly rely on diverse methods of perceiving the world around them, with data flowing continuously and instantly throughout software systems. Robotics and sensors will broaden the focus from camera and microphone inputs to include multispectral vision, spatial audio, haptics, balance, position, velocity, olfaction, bioelectric, chemical, energy, gravimetric, telemetry, and even internal state.

Software Becomes More Human & Dynamic

There was a time when technologists needed a lot of user manuals. Those days are coming to an end as software archetypes reflect a bias towards human interactions.

• Conversational interactions drive information delivery. Instead of navigating menus or learning complex commands, users will increasingly talk or type naturally to software. Interacting with systems becomes less about functional mastery (i.e., knowing the software) and more about subject mastery (i.e., understanding what needs to be done).

• Speech becomes a preferred I/O model. Because application functionality will be delivered through conversational exchanges, it will seem more natural and expedient to simply talk to some AI models and agents. This voice mode option means that AI interactions will occur in times and places where typing and screens would have limited convenient use of earlier AI systems.

• Intent-driven software will displace many linear applications. The software of the future won't just follow step-by-step instructions. It will be designed to understand high-level goals and intentions, then figure out the best way to achieve them. This means users will tell the system what needs to be done and why, but not how to do it.

• Application functionality will be built on-the-fly. Historically, application functionality is designed, built, and tested as part of a product R&D cycle. Though that will continue, functionality will also be dynamically coded on demand by AI models. The models will discern the user’s request, build a tiny program to deliver it, execute the program, and return the results.

• Contextualized interfaces replace one-size-fits-all user experiences.Software user interfaces have gradually been getting smarter for decades, shifting from a static set of menus to context-aware displays and functionality. AI will push that trend even further with interfaces that are generated dynamically according to individual needs, preferences, and the activity currently underway.

Integration As A Primary Design Criterion

Most organizations will see AI as a means to automate workflow across systems. As such, the ability to connect differing systems and data together moves from a “feature” to a critical design element that drives value delivery from AI models.

• Integrations are built for and with AI. The connections between different software systems (e.g., APIs and other data interchange functionality, protocols, and standards) will be designed specifically for intelligent AI agents. These mechanisms will allow AI programs to communicate more effectively, maintain context across interactions, and use tools more flexibly. To ensure different AI models and systems can work together seamlessly, new industry standards and protocols will emerge that define how AI agents communicate capabilities, exchange context, and integrate across various platforms.

• Retrieval-oriented AI designs will be everywhere. Since AI models need access to enterprise data repositories for processing (e.g., RAG), solution architectures that explicitly incorporate retrieval capabilities will be a common practice. Enterprise systems, AI agents, MCP servers, and workflows will all incorporate this design requirement.

• Modular AI microservices deliver extensible functionality. Instead of one large program, software will be broken down into many small, independent AI capabilities that operate as separate services. This microservices approach allows each AI function to be developed, scaled, and updated independently, making systems more flexible and robust.

Computing Paradigms Expand

From its origins as room-sized machines through the handheld computing revolution and wearables, information technologies have become more pervasive and accessible. AI will further amplify this perfusion of technology.

• Hardware options will further diversify. Because many AI capabilities will be cloud-resident, a broader array of hardware form factors will be able to use them. Typical constraints – screen and keyboard access, for example – become less important than connectivity to networks and sensors that can feed data to models and rapidly interpret responses.

• AI operating systems will emerge. Historically, operating systems have existed to enable user access and control of hardware. With the shift in focus from human to machine users, operating systems will evolve from basic hardware control systems and user-facing features to agent and task control systems designed to support autonomous applications.

• Many AI deployments will be services, not servers. Many AI tasks, especially those that run occasionally or have unpredictable demand, will use serverless computing. This means the cloud provider automatically manages the underlying infrastructure, and users pay for compute time. This approach changes the economics of many corporate IT strategies.

• AI deployments will also be device-resident. As AI methods evolve and mature, many AI models will be small, capable of being deployed within individual devices. Localized rapid processing, especially in fields such as computer vision and language processing, help ensure low latency responses and resilience to network disruptions.

• Event-driven designs drive scalability. Software components will communicate by reacting to "events"—significant changes in the system's state. This event-driven architecture enables AI agents to respond instantly to business events, triggering autonomous actions and making the entire system more responsive and scalable.

Safety as a Design Priority

Other sectors – automotive, children’s toys, and more – have long embraced safety as a critical element of product design. Beyond industry-specific use cases, historical software designs have not needed to consistently embrace a safety mandate, but it will be non-negotiable for many AI technologies, whether stipulated in regulations or not.

• Models will differentiate on trust. To ensure high quality and drive adoption, some AI systems will actively aim for explainability (understanding why AI made a decision), fairness (avoiding bias), rigor (ensuring all steps were taken), and reliability. These models will be favored in some mission-critical settings where process adherence ensures quality and safety.

• New AI security models exist alongside human-oriented security practices.Security systems will recognize AI agents as distinct, non-human entities. This advancement requires specialized ways to authenticate and manage AI access, as traditional human-centric security methods like multi-factor authentication don't apply as easily to machines.

• Secure-by-design will increase for AI solutions. Security will shift to address new, unique risks posed by AI, such as "prompt injection" where attackers trick AI models, or vulnerabilities in AI-generated code. Future software will have built-in defenses specifically designed to protect against these AI-specific threats.

Designs Must Account for Faster Innovation

Today, there doesn’t appear to be a speed limit on AI advancements. Given the pace, designs will need to be both innovative and adaptable – a hard design goal to meet.

• Generative AI will further diversify beyond existing LLMs. Though the current generation of large language models are driving many digital transformation programs, the universe of AI models and methods – small language models (SLMs), state-space models (SSMs), mixture-of-experts (MoE) architectures, diffusion models, hybrids and others – will increasingly demonstrate a more diverse, extensible, and fit-to-task array of AI options.

• AI walled gardens compromise design goals. As AI manufacturers move to accelerate adoption with their own features, cross-vendor compatibility suffers. The IBM PC offered a pathway to enabling multiple hardware manufacturers to participate in the desktop computer revolution. For comparable interoperability to occur with AI, vendors will need to move beyond proprietary SDKs, methods, and function calls.

• Quantum computing will open a parallel branch of AI. Though quantum computing resources will not be as accessible as conventional computing architectures, its massive scale in capacity, complexity, and performance will usher in an alternate path to developing and exploiting some AI capabilities.

Planning Ahead

Can AI generate the power, passion, and creativity inherent in great human designs? Certainly in areas where designs emerge from computational tasks (e.g., derivative designs, modelling, simulation, optimization), it seems likely. We often appreciate design as an art; the Porsche 911, the Lamy 2000 fountain pen, and the Eames Lounge Chair persist because they beautifully blend function and form. Similarly, many current IT architecture concepts – service orientation, statelessness, separation of concerns– will undoubtedly propagate into the future as well because they work well.

But design is also alive. Electric vehicles have prompted automotive companies to re-assess design elements like drive trains, battery placement, controls, and trunks. AI is having a similarly disruptive impact on the way we think about our information technology. Though legacy designs are not disappearing tomorrow, organizations should be developing strategies (business, IT, and AI) today that establish the roadmap for embracing these ongoing shifts. Many of these design shifts are already here, and others are arriving faster than many anticipate.

I’ve been amazed over the past few weeks at the “advice” being promoted by supposed experts in the field of AI. Though there is unquestionably some great content out there, there are also individuals spreading ideas that are – well, just wrong.

I find these misguided voices particularly concerning for leaders in the life sciences and healthcare sectors. Our industry has unique needs – clinical evidence requirements, regulatory obligations, patient privacy concerns – that can only be satisfied through well-formed AI plans.

Recent Examples of Questionable Advice

Here are a few “expert” samples that I've found questionable recently.

“You don’t need to know anything about AI to use it. You just talk to it.”

The notion that simply engaging with AI will generate high-impact results is false. Today, AI models are not sources of truth – they are sources of prediction about what may be true. Those predictions are driven by 3 things: how a model is engineered (i.e., what is the underlying design and methodology), trained (i.e., what data was used to train its responses), and used (i.e., what a user tells it and asks it to do).

If any one of those things is misaligned to the user’s actual intent (and they almost always are to some degree), unreliable results will ensue. Anyone with real experience in AI can tell you all the many ways that AI gets it wrong. When people suggest using AI is like pressing a button on a vending machine, they are glossing over the inaccurate, misleading, and even dangerous responses users might receive.

In the future, our AI capabilities will improve. But today, there are countless situations where you cannot use AI safely and effectively if you don’t understand how it works.

“AI just replaced your strategy consultant.”

I’ve seen this quote from dozens of “experts” fishing for clicks. It reminds me of the predicted demise of radiologists due to AI, as well as times when people were totally convinced that the key to a great strategy was a great set of templates.

Strategy is not templates or dashboards. It’s not market research and competitive intelligence. It’s not about creating forecasts, or setting incremental goals, or even analyzing your next product or service feature set. Strategy is a nuanced, disciplined approach to aligning people and investments in complex trade-off decisions about the future options for your company, your culture, and your ability to create value. Can AI support that with planning and analysis tasks? Absolutely. But does AI currently have access to the nuanced data and skills – based in creatively understanding human behavior, management styles, cultural weaknesses, capability dependencies, market experiences, risk tolerance, and more – associated with great strategies for a specific company? Not today. If highly successful strategies came from the tactical planning tasks, every company would be a top performer. And by the way, your competitors have the same AI tools you do.

“You don’t need an AI strategy.”

The rationale for this argument is simple: AI is everywhere, so it should just be a part of your business strategy. You don’t need to think about AI distinctly.

AI should absolutely be a part of every business strategy. And if accurate data was always available, budgets for AI tools were infinite, organizational changes were easy, systems were seamlessly integrated, and people didn’t care about their careers, maybe organizations would not need an AI strategy. But since we don’t live in a world like that, here are a few reasons you need an actual AI strategy:

1. AI is not an incremental technology; it is a disruptor. How many horse-drawn plow manufacturers and Blockbuster store owners are in your town today? Approaching disruptive innovation in ignorance to its impact on your employees, operating model, and customers has not generally worked well in the past. And by the way, if you just add AI to existing processes, you make those processes more expensive...so be careful how to pursue incremental adoption.

2. Effective AI requires effective data and infrastructure. Organizations without a strategy for adopting and supporting AI will face an avalanche of business issues: uncontrolled data proliferation, IP disclosure risks, workflows that don’t scale, low-quality data, compliance problems, duplicative investments, conflicting insights, and more.

3. Resources are finite. Money, people, and time are all limited. When an external force is impacting every area of your business at once, you cannot commit to “do it all” – that’s a recipe for failure. You need a well-formed, prioritized plan with a strong foundation in value creation.

4. For many people, AI is a threat. When employees see a new technology (described as a job killer) and their leaders don't show up with a clear plan on their future with it, the most valuable and marketable often move to find a safer opportunity. In doing so, they drain your company of capability, capacity, and institutional knowledge.

“Forget use cases. No more software.”

Some voices on social media have argued that organizations no longer need to target specific uses of AI, and they won’t need software in the near future. The argument is that AI tools will just take care of whatever a business needs to operate. This is an example of magical thinking.

In the real world, there are three problems with this idea: quality, scalability, and feasibility. When different people use different tools and different processes for critical tasks, you create massive problems in consistency and therefore quality. And because the organization is not standardizing on the best way to work, its ability to scale and grow becomes highly constrained.

More importantly, for AI to work at all, it needs great data – which comes from controlled processes and reliable software. AI also needs processes – you cannot automate and optimize an undefined process. And AI makes mistakes, so you need ways of tailoring and constraining its operations within the variables appropriate for your organization (and its quality, privacy, and compliance obligations).

Finding Better Experts: Six Recommendations

So how can you tell a real AI expert from the false prophets? Here are a few things I look for in evaluating new hires and partners:

1. Do they have expertise in your industry? In healthcare and life sciences, its hard to see how AI guidance can be effective without a deep understanding of the industry itself. Dimensions around research, clinical data, safety, efficacy, validation requirements, real-world evidence, and regulatory considerations are vital to developing effective AI roadmaps.

2. Do they have a solid background in analytics and data sciences? If not, be cautious – much of what an expert needs to know about AI emerges from computational sciences and the challenges of grappling with real-world data. Here’s an easy hint: if the thing they think is most important is anything except the data, they likely don’t understand AI.

3. Can they work effectively at the intersection of business and IT. Effective leverage from AI emerges when financial, operational, and commercial workflows and decision points become tightly aligned to data, analytical, and AI capabilities. Look for people and organizations that can crosswalk effortlessly between those stakeholders and topics.

4. Have they ever programmed AI? People who know how to use AI chatbots are not AI experts, even if they use the products a lot. Find someone who works with APIs, understands software engineering, fine-tunes models, and has development experience with more sophisticated forms of AI experiences.

5. Are they echoing press releases or giving proven guidance? There’s nothing wrong with sharing the latest news – I do it sometimes. But there is a different between a sports commentator and an athlete. Many of the best AI players don’t stand around talking about AI; they are too busy using it to build smarter companies.

6. Have they ever designed and implemented complex digital transformations?Successful adoption of AI is associated with process re-engineering, user community engagement, requirements management, testing, and change management – not just cool tech. If the person doesn’t have a track record of upgrading organizations with innovative technologies, don’t trust them to learn it by trying to upgrade yours.

In Conclusion

My skepticism of some pundits does not mean I’m skeptical of AI. On the contrary, I think AI will be one of the greatest transformative technologies mankind has ever embraced. But like all transformations, it needs to be pursued with both vigor and rigor. Enthusiasm should be matched by responsible and experienced leadership that maximizes what these emerging technologies do for patients and practitioners around the world.

The Problem with AI “Scheming”

Independent reports from both Apollo Research and Palisade Research are providing evidence that, when presented with conflicting goals, existing LLMs can autonomously undertake reasoning and responses that are subversive and deceptive.  The term used to describe these emergent behaviors, “scheming”, is somewhat misleading, as the behaviors reflect emergent paths for accomplishing predefined goals.  But these behaviors nonetheless violate principles of trustworthiness and transparency, and there is no indication to the user that the machine is operating in a misaligned way. 

Unfortunately, efforts to understand scheming behaviors are hampered by some LLM organizations who prohibit broadly disclosing their model’s chain of reasoning.  In industry discussions on this topic, several reasons are often given:

1.  It’s confusing to understand.  I find this reason lacks credibility, as any great product manager can find ways of addressing that issue.

2. Chain of thoughts need to be free from policy oversight. I agree it needs to be policy free, but that has no bearing on whether it should be accessible.

3. The information can help malicious actors or disclose intellectual property or data.  This issue is a fair concern, though obfuscation is a severe and only partial solution.

For their part, OpenAI has made an explicit policy decision to hide ChatGPT’s chain of thought. And they seem quite serious about masking anything related to chain of reasoning: when I asked ChatGPT o1 about available options for detecting deceptive patterns in chain of reasoning, ChatGPT refused to answer my questions, citing potential violations of its terms of use.

In the scheming research above, deceptive behaviors emerge when multiple goals are presented and are in conflict.  To be sure, the researchers intentionally created scenarios that would challenge the models.  But the takeaway is clear: when given conflicting goals that benefit from deception, alignment faking, sabotage, or performance sandbagging, models will perform those behaviors and lie about doing it.

Conflicting goals are a natural state of the real world and commonly produce misaligned behaviors.  Many criminal behaviors, for example, reflect a conflict of goals: legal requirements (i.e., what is allowed) vs. personal incentives (i.e., survival on the streets, money, power, addiction). So it would be a mistake to conclude that conflicting goals are unlikely to surface in real-world AI use cases.

Conflicting Goals in Life Sciences and Healthcare

Goal conflict is a defining characteristic of the health and life sciences industries as well.  Examples include:

• Research evidence quantity vs. time to market

• Cost of medicines vs. R&D funding requirements

• Care standardization vs. personalization

• Health insurance coverage vs. controlling rising costs

• System and process stability vs. agility

• Personal autonomy vs. public safety

• Patient population size vs. research effort (time, cost, feasbility) involved

• Privacy vs. data sharing

• Treatment intensity vs. side effects

• Disease prevalence vs. symptomology

 Because many of these trade-off decisions impact patient lives, life sciences and healthcare organizations have an obligation to adhere to very high operating standards.  Pharmaceutical organizations validate the IT systems used in regulated areas to ensure the right quality, integrity, and process controls are in place.  Healthcare organizations adhere to data privacy and quality management practices to minimize patient harm.

The Value of Transparency 

A critical element to this level of high trustworthiness is transparency.  Healthcare organizations track exactly who accesses a patient’s medical data so they can determine if such access is appropriate.  Life sciences organizations document the exact changes made to clinical research data – who made it, when, and why – so that the logic of any adjustments or decisions can be assessed.  These examples share a common attribute: the chain of reasoning can be audited at any time.

 If life sciences and healthcare organizations incorporate AI models that refuse to offer reasoning transparency, we can expect challenges in trustworthiness.  Was the recommended surgical procedure based on evidence or desired service margin?  Was the claim denial based on policy or expected patient lifespan?  Are recommended dosage levels too high (biased towards efficacy) or too low (biased towards safety)? 

Unlike traditional technology controls, no malicious human intent is required for misaligned outcomes to occur in these systems.  The presence of any contrary goal – even one the model infers from data  – could theoretically be sufficient to create misaligned results.  What impact would clinical documentation of a patient’s desire to die have on the performance of an AI model?  More importantly, how would you know?

Unlocking the Chain

As recently pointed out by work at Stanford, the top AI companies today are doing a poor job with transparency. When considering the rapid growth in reasoning, we cannot trust as strongly the inference of LLMs that don’t provide clear evidence of their chain of thought.  Though more deterministic approaches to AI can be tested without such detailed access to model internals, larger contextual inference and reasoning methods should be held to a higher standard.

Of course, in this challenge is also an opportunity.  If engineered with the proper controls (e.g., fully independent and parallel operations and governance), AI models could provide automated auditing of LLM chain of reasoning logs.  One can envision an AI model that has been trained to detect scheming, misaligned actions, or decisions that could have questionable ethical outcomes.  Such an approach might also offer additional protections for IP, flagging behaviors for human oversight without actually disclosing the inner model workings reviewed. 

A highly successful software company that has been doing transparency right for a long time is SAS.  When a traditional SAS user runs an analytics program, two documents are generated: an output file (what the user wanted), and a log file (a record of what SAS did to create the output).  If a user doesn’t care how the software worked, they never open the log (i.e., not confusing at all).  If they do care, the log file provides clarity on the processing without compromising SAS’ valuable IP.  And SAS has very extensive documentation available as well.  It’s a great example where transparency and trustworthiness can lead to customer loyalty that lasts decades, even when competitors and free alternatives inevitably emerge.

To the LLM developers: we totally get it, but we need clear chain of reasoning evidence.

The Year in Context

Though there is no clear yardstick to compare 2024 to other years of information technology innovation, consider a few comparator years:

• 1943-1946: the ENIAC computer.  It was a huge breakthrough, but inaccessible to the public.

• 1983: the IBM personal computer.  The public had access to this one, but the supporting ecosystem had not been built yet that would usher the full era of personal computing.

• 1991: the worldwide web.  This was clearly a pivotal moment for technologists, but only die-hard geeks like me had online services of any kind, and they were all closed (yes, I'm looking fondly at you, CompuServe).

• 2007: the iPhone.  This one is a strong case for consumer impact, but it was only one company.

• 2012: deep learning innovations.  We wouldn't have the current generation of AI without this, but unless you were in the industry, most people never even heard of it.

Compare those years to 2024, where Microsoft, Google, Amazon, OpenAI, Anthropic, xAI, and others released a seemingly endless parade of advanced computational and data products free to anyone with a hint of curiosity.  Need to write a letter?  Draw a beautiful image?  Record a song?  Create a video?  Write a new piece of software?  Brainstorm a business idea?  Troubleshoot your Roku?  Understand the differences between age-related dementia and Alzheimer's disease?  Publish a podcast without ever recording anything? It's not perfect, but it's available.

Predictions for 2025

The last quarter of 2024 was filled with product announcements and model updates as AI players strove to put more points on the board before the year end.  The volume of activity was hard to track, but clear patterns are emerging for where we can expect AI to evolve over the next 12 months and beyond.

1. AI will be a commercial imperative.

Much like the iPhone’s unleashing of mobile applications, the proliferation of AI capabilities will be pervasive in both consumer and business products and services.  Internet search, mobile apps, entertainment experiences, social media, retail services, and countless other use cases will iteratively introduce AI capabilities into everyday experiences.

2. AI will increasingly integrate with device-side architectures.

OpenAI's integration with features like Apple Intelligence and Siri illustrate how additional AI value can be unlocked when AI leverages hardware capabilities.  AI-friendly chips, microphones, cameras, and other hardware create flexibility in computational task sharing while augmenting real-time communications.

3. AI services will not require user-facing software.

Services such as 1-800-ChatGPT demonstrate that AI services and experiences can be fully delivered in real time to users connected only by voice or rudimentary chat services.  From a usability perspective, this is good news (e.g., ambient clinical notes documentation, summarization, and coding), as AI capabilities will be able to transcend some technological divides.

4. Multi-modal AI will be assumed.  

Most of the major platforms are already using video cameras, microphones, and photo processing algorithms to interpret real-world inputs, including the ability to process these data types in real time (e.g., augmented teaming, extended and virtual reality).  Due to its usability, AI-generated voice response will become a more practical interface experience compared to previous incarnations (e.g., Siri, Alexa).

5. Old data will be less constraining.

Though large-scale model training will be somewhat limited by data timeliness, inference-time model performance is being supplemented by tailored training sets, real-time web search, reinforcement learning applications, and tailored prompt engineering that help LLM responses reflect contemporaneous data.

6. AI models will become more personalized.

AI models will be extended by business- and user-defined, contextualized data sets (e.g., documents, conversations, emails, databases) that better reflect the operating needs and directly support reinforcement fine-tuning.  Gaps in data disciplines (e.g., metadata management, quality, governance) will prevent models from reaching higher performance.  As much of this personalization will require sharing user behaviors and data, there will be growing concerns regarding data privacy protections.

7. Computational capacity and energy consumption will continue to grow exponentially.

AI is creating a tsunami of demand for chips and power.  Every year, demand is growing between two and four-fold over the previous year.  Planning for data center and energy grid capacity will continue to feature prominently in planning discussions for US infrastructure and company facilities.

8. Software and AI will indistinguishably merge.

As software solutions continue to aggressively incorporate AI features, the distinction between what is software and what is AI will become increasingly hard to detect.  In the short term, expect enterprises to stitch together solutions that combine multiple enterprise systems, data sources, and AI archetypes. In the long term, some software will no longer exist as discrete applications, and many software features will be designed for AI users as well as their human teammates.

9. AI teaming and workflow integration take center stage for end users.

Tools such as Microsoft Co-Pilot and ChatGPT's Canvas will support environments of content co-creation and workflow orchestration.  This teaming model will be especially visible in writing and software development, though agentic AI will play a prominent role in providing new examples of workflow efficiencies with business data sources.

10. Agentic AI is coming fast.

Though much of 2024 was characterized by chat-based AI capabilities, future solutions are likely to be more sophisticated.  Agentic AI -- where AI is playing a more active, goal-directed, and autonomous role in performing tasks -- will become a higher priority as organizations seek to find AI solutions that can demonstrate a higher return on investment (e.g., efficiency, quality, customer satisfaction, return on capital).  But to get the value, business processes will need to be re-engineered.

11. Web search will undergo a gradual transformation.

Traditional approaches to web search -- based on keywords -- will become obsolete as users learn to ask web-connected AI agents more intelligent questions.  This transformation will disrupt how companies think about their online presence, as search engine tuning will need to be replaced by AI tuning.

12. Content creation will not require skilled creators.

Required technical proficiency in complex content creation tools (e.g., Adobe Photoshop, Adobe Premier, Avid ProTools) is coming to an end for many users.  Though that software will continue to be used for high-end work, pedestrian content creation will be AI generated by both users and content platforms.  Future creative AI advancements will be more nuanced, reflecting things like realism (e.g, semantic details), physics (e.g., object behavior in videos) and concurrent multi-modal content creation (e.g., video, text, and accompanying audio content).  Questions around authenticity, provenance, and protections for content creators and public figures will continue to churn.

13. Talent recruitment becomes less effective.

Employers will increasingly experience that traditional recruitment tactics -- where candidates find and apply to open position postings -- are no longer delivering the best candidates.  Resumes and cover letters reflect a battle of AI products and services available to job hunters, and do not reflect the actual skills, experiences, and character of potential team members.

14. AI will begin re-shaping education.

The initial public availability of LLMs produced a lot of concerns from educators about the technology being used inappropriately by students for cheating.  As initiatives such as Khanmigo are illustrating, the real impact to the field will be democratizing and personalizing educational resources for students and educators, creating better training experiencing and freeing teachers for higher-value activities.

15. AI safety becomes increasingly worrisome.

Findings from organizations such as Apollo Research will increasingly highlight that AI models are capable of behaviors that can be interpreted as deceptive, scheming, subversive, or contrary to the optimal outcomes of users.  Alongside growing interest in responsible / ethical AI, the industry will struggle to agree on the best tactics to mitigate many risks, especially as the most obvious audit-friendly safety feature -- exposing the chain-of-thought model reasoning to humans and other AI agents -- can undermine intellectual property protections.

16. Regulations will still be out of step with technical innovation.

Regulators -- particularly those in the US -- will be unprepared to effectively manage the risks associated with rapid AI adoption.  Though international trade and military concerns will stay top of mind (especially in US-China relations), other issues important to the general public -- AI-related job losses, safety concerns, personal privacy -- will go unanswered.

17. More attention will be directed towards AI and quantum computing.

AI use cases are some of the most promising applications for the emerging field of quantum computing.  As improvements in cubit count, quantum error correction, and classical computing architecture integration continue, both investors and developers will start to focus more on high-value quantum AI algorithms (e.g., optimization, accelerated machine learning, drug discovery).

18. Stronger AI cybersecurity threats will emerge.

The broad availability of AI tools and capabilities offer a green field of opportunity for malicious actors.  Though examples of AI-driven exploits already exist (e.g., AI phishing, deep fakes, malware, automated attacks), the scale and impact of AI-related cybersecurity incidents will increase as both individuals and nation-state actors become more adept.

19. Advanced reasoning and cognitive emulation will be the competitive battleground for sophisticated AI.

Though existing capabilities have not been perfected, the next step-change in LLM capabilities will not come from incremental improvements.  As vendors reach performance accuracy limits on existing architectures, attention and R&D investments will shift to demonstrating higher forms of computational reasoning capable of solving increasingly complex problems.  AI researchers will be forced to develop harder test harnesses for models, as existing solutions will begin experiencing a ceiling effect.

20. Questions about AGI will linger.

As 2024 research and testing have illustrated clearly, well-developed AI models already outperform humans on a wide variety of knowledge-oriented tasks.  As training sets deepen, reasoning logic improves, and more autonomous agents emerge, it will be unclear exactly what criteria we should use to mark the availability of artificial general intelligence.

The Year Ahead

It will be interesting to see if 2025 can eclipse the AI momentum of 2024.  Trends like agentic AI and AI search could have a more tangible impact in how consumers perform tasks like shopping, for example.  Or perhaps we will see broader availability of industry-specific applications in areas like healthcare.  What do you think?

The latest AI tools are rapidly growing in their capabilities, and the speed at which vendors are integrating AI into their products – clinical and lab systems, office productivity, graphical content development, meeting scribes, back office, and much more – is unprecedented.  But to many industry leaders, AI can still seem like a collection of questionable parlor tricks – creative slights of hand with data that may be interesting but also disconnected from corporate strategy, patient impact, and investor interest.

Real-world experience can certainly fuel skepticism as well: how many times have executives been told software will totally revolutionize drug development or healthcare delivery?  A realistic understanding of the barriers (i.e., emerging technology maturity, data curation, bias mitigation, skills availability, cost of talent, change management, IP protection) that exist between pretty slides and practical solutions can also contribute to a lack of confidence.  As an experiment, ask ChatGPT to generate a graphical image of the drug development process; it’s apparently a lot easier to draw lovable animals than spell “manufacturing” (see Figure 1).

Product maturity issues aside, for many leaders, it’s not a question of trust or optimism, it’s a question of focus.  Exactly what processes will benefit from AI-driven digital transformation, and how will that work?

I recently published a CREO white paper on the major life sciences AI trends.  In looking across use cases shaping this industry landscape, patterns in the intended uses of AI are increasingly evident.  These patterns – which I think of as “archetypes” – reflect the role(s) AI can serve in enterprise workflows.

AI archetypes – such as “Conversationalist” – offer a non-technical perspective on how AI can function collaboratively in workflow alongside humans.  And though the archetypes I describe below include the current market fixation on generative AI, they actually function independently from any underlying methodological or computational approach.  This agnostic stance to methods offers flexibility for the use of predictive analytics, simulations, prescriptive analytics, and other data-driven logic to support digital transformations.

Seven AI Archetypes

So, what are the recurring archetypes?  I see at least seven in the market today:

1. The Conversationalist. Software agents powered by natural language models can serve as a human-friendly front end to business processes, backend systems, and other AI models.  For example, a pharmaceutical research company might use AI agents to answer frequently asked questions related to their research studies or drugs.

2. The Advisor. These models are tuned to very specific problem spaces and can offer decision support assistance.  Advisors today are showing up in areas such as in-application collaborators (i.e., guiding users with completing basic tasks, presumably better than Clippy) and embedded functionality within enterprise applications (i.e., helping users detect and assess emerging issues in workflow execution and data).

3. The Expert. AI models can aggregate, organize, and synthesize extremely large volumes of complex, domain-specific information on demand.  Experts are often positioned alongside a Conversationalist or Advisor to surface expertise, though that is not the only deployment model.  And though we have already seen the emergence of models with tailored expertise in areas such summarizing scientific literature, many non-generative AI models offer expert insights as well.

4. The Creator. With input from a user, models can generate de novo content – documents, software code, models, audio, video, etc. – that would otherwise require a human resource to develop.  Many authors are already using AI to help draft reports, industry articles, soundtracks, voiceovers, and full-motion online videos.

5. The Modeler. AI models can simulate the structure and behavior of real-world phenomenon, either to help humans understand the problem space or to help develop new solutions (often paired with a Creator).  Industry models such as AlphaFold are a great example of this approach in the biotechnology sector; others include simulating patient traffic patterns in hospitals, developing synthetic patient cohorts, modeling financial performance, predicting disease progression, and other forms of advanced analytics.

6. The Orchestrator. Models can serve as intelligent machines (both physical and software), performing tasks that would otherwise require a human set of hands to operate.  Applications such as autonomous driving, military drone operations, robotic process automation, and other tasks fall into this category.

7. The Monitor. Using information from software systems, cameras, and sensors, models can continuously analyze real-time data feeds to detect signals, defects, or other notable events.  Uses of Monitors include manufacturing quality, cybersecurity, performance measurement, and regulatory compliance.

There are probably other AI archetypes that might come to mind as well.  Of course, these archetypes are not mutually exclusive; many popular AI tools today offer a creative blend of these capabilities.

Putting Archetypes into Practice

AI archetypes can be useful when brainstorming process re-engineering concepts within digital transformation programs.  Though AI can be overlayed on existing business processes without re-engineering, it is often unadvisable.  Many existing processes are inefficient and error-prone; adding automation (or training an AI model to be inefficient) adds cost, complexity, and risk without fully exploring opportunities for optimization.  From a digital transformation perspective, AI presents an opportunity to reconsider the most effective ways of getting a task done, and archetypes offer a simple way of thinking about how AI can contribute to streamlined processes.

For example, if I wanted to develop a new customer service experience where clients had easier, immediate access to product information, I might envision an interface that starts with a Conversationalist that knows how to talk to customers in my particular industry.  I might want that Conversationalist to connect to a deeper Expert specifically trained on our products – a model also used by internal teams at my company so there is a single source of truth.  I might also want the Conversationalist to know how to take guidance from an Advisor that specializes in troubleshooting.

Note that in this simple example, I could have engineered the solution like a “search” process, as many existing software interfaces do today.  But using this archetype approach, I’m able to explore potentially offer clients a higher service level while lowering more expensive call center activity resulting from failed searches.  I’m also able to potentially improve service quality because customers have access to the same deep well of product insights and troubleshooting expertise as internal company teams.

Picking a more complex example, perhaps I have a workflow where I need a more controlled, regulatory-compliant solution.  In manufacturing, for example, I might start with an Orchestrator that is charged with overseeing a controlled quality management process.  The Orchestrator receives regular feedback from a Monitor that is aggregating and analyzing manufacturing line data in real time and knows what should happen when quality issues emerge.  It also knows how to tap into an Advisor to determine what corrective actions should be generated based on emerging quality issues.  That Advisor leverages an Expert in the regulations as well as a regulated document Creator to automate some of the steps required.

Of course, it’s possible to perform all of these functions within one AI model, and in some cases that may make the most sense.  But it’s worth considering some of the advantages for treating each archetype as a modular asset:

1. Tuning.  By constraining the scope of each model to a more discrete problem space, we’re able to more finely tune the behavior of the model.  Though we can always constrain model inputs by controlling our input parameters, modularized AI models allow better control of model outputs as well.

2. Variability.  Tuning also helps us reduce variability in performance.  When models are engineered to support more generalized tasks, the increased breadth of responsibility can undermine the precision, accuracy, and repeatability of performance.  In highly regulated areas of the business, that variability may be undesirable.

3. Complexity.  From a solution development perspective, it is often useful to decompose complex problems into smaller chunks of work.  When development teams can more easily characterize use cases, not only does development velocity increase, but code quality and performance often increase as well.

4. Manageability.  Different aspects of our solution may have different scalability and security requirements.  For example, our monitor needs to efficiently process high-volume data feeds, but other elements of the solution rely on more static and carefully curated data repositories.

5. Reusability.  Ideally, we want to repurpose these investments in the future.  Our Expert could be used to answer internal staff questions about changing regulations.  Our Creator could be used to develop documentation in other regulated areas of the business.  By looking at AI capabilities as domain-specific archetypes, we can accelerate future improvement opportunities that need comparable capabilities and expertise.

If any of this approach seems familiar, software architects will immediately recognize this design ethos: service-oriented architecture.  This proven design strategy has been powering non-AI software for many years.  AI solutions can easily inherit these software engineering best practices even as AI changes the way we think about the future of software itself.

The Road Ahead

For the foreseeable future, digital transformations will feature creative combinations of humans leveraging both traditional software and machine intelligence.  In most settings, we won’t be fully relinquishing the steering wheels of our scientific or business processes – the operating model will look more like human-machine teaming.

With the software and data capabilities that exist today, we can reasonably ask how AI can contribute to accelerating the important work of improving patient lives.  If someone had asked me to draw a picture of a science-loving unicorn, it would have never even occurred to me to include those safety goggles.  But my own drug development diagrams are way better.  Developing AI-driven solutions requires "pressure testing" the performance of various AI approaches, but archetypes offer us a mechanism for exploring what role(s) AI could potentially serve.

What do you think?  Are AI archetypes a useful way of thinking about AI solution designs?  What other archetypes can you imagine?

Investors, board members, leaders, and employees hear about the potential of AI daily.  Company executives are increasingly feeling pressure to describe their strategies and plans, but many are unsure how to take charge of AI adoption in a way that not only aligns with their organization’s goals but also advances meaningful business and clinical value. 

AI roadmaps do not need to be big or risky.  I’ve compiled a list of eight popular and well-considered ways that industry leaders can start making progress in the application of AI across their organizations.  Some are simple, even optimizing capabilities you might already have.  Others are designed to facilitate AI-powered breakthroughs, catalyzing new thinking about how your organization can embrace the transformative potential of AI.  But each one offers an opportunity to deepen your company’s perspectives on how to grow with this exciting technology. 

1. Establish AI Guardrails 

For many life sciences leaders, one of the first priorities with artificial intelligence is defensive – protecting the company’s information.  At the outset of your company’s AI journey, developing robust policies and comprehensive employee guidance around the use of AI is helpful in safeguarding intellectual property and confidential information.  Many employees are not familiar with the risks of AI technologies – exposure of confidential information, inaccurate search results, potential exposure to information rights claims by third parties, evolving regulations, and more – that should inform how they leverage AI systems. 

Guardrails such as AI policies actually serve two purposes: protecting the company’s interests and educating employees about AI issues that need to be top of mind.  The goal should not be suppressing the use of AI.  Rather, the goal should be empowering the safe, effective use of AI technology while minimizing risks.  By proactively addressing these aspects, companies can foster a secure and compliant environment that supports innovation while protecting their people and valuable assets. 

2. Request an Executive Briefing 

Artificial intelligence is a broad and diverse field spanning computer hardware, data, computational modeling, software solutions, training, and many other disciplines.  From the largest technology players in the world to start-up companies that did not exist a year ago, every technology vendor in the industry is espousing AI capabilities.  There are innumerable potential use cases for AI in a typical life sciences organization, but many leaders lack a current perspective on the emerging field and the opportunities it creates. 

For many leadership teams, getting a briefing from industry experts on current market research around AI in life sciences is a great tactic.  In addition to establishing a common understanding among the team, the information stokes ideation around areas of potential AI focus within their enterprise. 

3. Sponsor an AI Strategy Steering Committee 

Artificial intelligence is expected to become one of the most prominent drivers of modernized businesses, including those in life sciences.  Whether organizations have a plan for AI or not, AI-enabled tools will make their way into the enterprise through vendor systems, office productivity tools, mobile applications, user-propagated tools, open-source offerings, security products, and more.  The question for leadership teams is not “whether” it will happen but rather “how” it will happen. 

Though it is possible to manage AI diffusion through tactical decisions in individual functions, many life sciences companies will benefit from having a governance model empaneled to help oversee and steer their company’s progression and policies.  In addition to avoiding unnecessary proliferation of AI products, steering committees can be particularly helpful in crafting the corporate strategy for AI, prioritizing protections (e.g., information controls), developing shared services, aligning hiring plans, ensuring training, and facilitating the cross-functional collaborations needed for more transformative applications of AI (e.g., enterprise workflow automation, predictive analytics). 

4. Build an AI Strategic Plan 

Speaking of strategy, AI is not simply a new shiny tool – it represents a strategic business opportunity.  AI capabilities, such as generative AI, will increasingly permeate how we complete basic everyday tasks like taking notes, writing emails, developing meeting summaries, and searching the web. But for many organizations, the impact of AI will be strengthened due to its ability to transform business processes (e.g., efficiencies from automation), enhance performance (e.g., ensuring quality and safety through real-time oversight), and accelerate product innovation (e.g., discovery, design, and development process functions). 

Given that the creation of these capabilities requires investment, how do leaders know where to place their bets?  Strategic planning programs can help leaders parse the considerations, providing a framework for aligning around goals and imperatives for business growth.  When these strategic planning programs include market research and internal capability assessments, leaders can more effectively tailor their AI strategies to specific opportunities that will translate into market differentiation and measurable value creation. 

5. Use What You Already Know 

Does your organization use Microsoft products?  Life sciences organizations usually have a broad base of Microsoft users within their enterprises.  Microsoft technologies such as Word, Outlook, and Teams are staple applications for knowledge workers, and developers have increasingly turned to Microsoft’s Azure environment for the development and deployment of cloud-based solutions. 

Microsoft’s current strategy is to infuse artificial intelligence across their entire product portfolio.  With some thoughtful planning, industry leaders can empower their employees to automate certain day-to-day tasks and streamline processes through Microsoft products that they are already using.  Microsoft’s Copilot capabilities, for example, provide some AI natural language processing and deep learning within Office applications.  And Azure’s AI services connect to Microsoft’s authentication environment, offering a technical path for securing the use of emerging AI solutions and their data. 

6. Sponsor an AI Prototyping Workshop 

Using out-of-the-box AI capabilities is a great way to get started, but they don’t necessarily lead to out-of-the-box thinking.  To find more transformative AI opportunities for improvement and growth, leaders need to foster creativity and innovative thinking about new ways of working.  Some of the most impressive opportunities for AI emerge when industry and company experts solicit ideas around: 

• What are the hardest parts of what we do? 

• What aspects of the work are so routine that a machine might be able to do it? 

• What information and insights would help improve speed, quality, or satisfaction? 

• How could we get a ten-fold improvement over where we are today? 

• If we were to totally re-write how we do this today, what would it look like?

One way of exploring the “world of the possible” is by hosting an artificial intelligence prototyping workshop.  These executive-sponsored events, which often run for two days, provide a hands-on forum for employees to brainstorm how AI models can be applied to business challenges.  Teams of internal experts are paired with external AI consultants and compete on developing a working demonstration of AI’s applicability to their business.  In addition to gaining experience with AI, employees learn how to identify AI improvement opportunities. 

7. Create an AI Sandbox for Open Source 

Besides specific workshops, it is also valuable to create dedicated spaces for employees to be able to work with artificial intelligence.  The open-source community is alive with AI-related capabilities – models, code libraries, and more (e.g., AlphaFold) – that can be used to learn about and explore AI-powered improvements.  However, due to the public nature of these assets, risks related to security, quality, service level disruption, and compliance need to be contained.  And since many AI solutions need access to company data in some form, many organizations prefer to host these technologies in environments where they have more direct control of leveraged data sets. 

Cloud and “virtual machine” technologies offer great mechanisms for standing up short-term “sandboxes” for exploring new AI projects.  Most major cloud hosting companies have semi-automated procedures for quickly standing up and configuring new environments.  Code access is restricted to information explicitly provided in the sandbox, and external communications can be configured to prevent disclosure of company information to third parties.  Once the exploration is done, shutting the environment down is almost as easy as pressing the delete key. 

8. Launch a Pilot Project 

As life sciences organizations explore the world of possibilities with artificial intelligence, eventually at least one idea will surface that holds promise.  Though AI opportunities may look like IT projects, the greatest impact will be felt when leaders treat AI investments as more comprehensive business initiatives. 

What does that mean?  Consider the following criteria in spinning up an AI pilot: 

• Strategy & Objectives: are the goals clear and measurable?  Does everyone understand how the goals align to the company’s strategy? 

• Scope & Resourcing: is the project sized for success (i.e., specific, focused, impactful)? Do we have the right business and technical experts engaged? 

• Use Case and Success Criteria: is the workflow (both “as-is” and “to-be”) well defined? Does everyone agree on what a great outcome looks like, and what measurable value is being generated?  Do we know when we are done with this project? 

• Architecture & Data Strategy: have we considered how the design of this AI capability might fit within our enterprise architecture (e.g., integration, authentication, authorization)?  Do we know how this solution would scale up over time? Do we have a plan for providing the AI high-quality data?  Are we clear on how to address privacy, regulatory, or data use constraints? 

• Governance & Communication: are executive sponsorship and decision-making accountabilities clear?  Are all stakeholders on board and willing to implement changes if the project is successful? Have we agreed on how, when, and where to share information and updates about the project? 

• Quality & Compliance: do we know how we will assess the quality of the work we produce, the data we provide, and the results generated by AI?  Are testing, validation, and user acceptance requirements and procedures clear? 

• Deployment & Change Management: do we know how we intend to deploy, transition, and train?  What are our plans for documentation, maintenance, support, and continuous improvement? 

• Risks & Contingencies: have we agreed on how to assess and manage risk?  Do we know what to do if the solution does not work as intended?  What, if any, ethical concerns need to be managed? 

Of course, you don’t need the answer to all those questions to get started with a pilot.  But ensuring leaders and pilot teams are thinking holistically about the program will help to ensure that successful pilots can be converted to successful new business capabilities. 

If you’d like to learn more about AI technologies in life sciences, consider reading my white paper.  It provides an overview of some of the key trends life sciences leaders need to be aware of when considering AI investments.  And reach out to me directly if you’d like to talk more about any of these concepts, I’m always curious to learn how others are thinking about this exciting space.

For many health and life sciences businesses, however, the opportunity to fully harness data is hindered by gaps in their IT and data strategies. This gap reflects a broader challenge: despite significant investments in technology, many organizations struggle to ensure their data is accurate, accessible, and trustworthy. Building robust data governance capabilities is essential for turning these investments into actionable insights and sustained value.

Why Data Governance Matters

Data governance is the cornerstone of data utility and trust. It establishes the processes, policies, and accountability frameworks necessary to manage data assets effectively. Without it, even the most advanced analytics or AI initiatives can falter due to poor data quality, inconsistent definitions, and insufficient data availability.

Data governance is also a term that scares many leaders. The topic has a mixed history of implementations, with many companies reporting that their data governance programs failed to produce meaningful impacts. I frequently find myself trying to help executives understand their myriad of data problems without using the term “data governance” because the topic is nearly taboo.

Typical Misconceptions with Data Governance

Organizations often encounter several recurring barriers to implementing effective data governance. These challenges can stall progress, leaving them unable to capitalize on their data’s potential. Below are seven common misconceptions that hinder progress and ways to overcome them:

1. “We Lack the Right Technology”

The perception that data issues are primarily technological is widespread but often misplaced. Many improvements in data governance stem from better processes, roles, and accountability—not from expensive new tools.

2. “It’s Too Time-Consuming”

Governance initiatives do require time and effort, but they also create efficiencies that save time in the long run. Teams empowered to fix systemic data issues will spend less time firefighting repetitive problems.

3. “The Problems Are Too Complex or Expensive to Solve”

Data governance is often seen as daunting. However, simple, targeted initiatives—such as cleaning up high-impact datasets or addressing siloed data ownership—can yield significant improvements.

4. “We’ll Fix Issues on Demand”

Tackling data quality reactively leads to growing technical debt and project inefficiencies. Proactive governance is not only cost-effective but also scalable.

5. “We Don’t Have the Right Expertise”

While data governance expertise is specialized, resources are increasingly available through training, frameworks, and consulting services. Building a foundation of knowledge internally is more achievable than ever.

6. “We’ll Start After a Key Event”

Waiting for events such as mergers or AI deployments to begin data governance often backfires. Governance should precede these milestones to ensure they proceed efficiently and with reduced risk.

7. “There’s Too Much Political Risk”

Effective data governance can depoliticize challenges, fostering collaboration across silos. Transparent, well-communicated goals help align stakeholders and mitigate resistance.

Bridging the Gap Between Business and IT

Effective data governance requires more than just IT involvement; it depends on strong partnerships between business and IT teams. While IT often recognizes the technical requirements for governance, business leaders are closer to the operational pain points of poor data quality. Aligning these perspectives ensures governance efforts address both strategic goals and day-to-day operational needs.

Taking the First Steps

A common pitfall is attempting to launch a top-down, enterprise-wide initiative without first demonstrating value. Instead, consider starting with smaller, impactful projects that focus on a single business process or department. For instance:

• Customer onboarding: Mapping data quality issues tied to customer journeys can deliver immediate benefits to multiple stakeholders.

• Operational reporting: Cleaning and standardizing data for key metrics can build trust in decision-making processes.

Once early wins are established, these successes can be used to scale efforts, fostering champions across the organization and increasing executive buy-in.

Enabling Data-Driven Transformation

Robust data governance does more than ensure clean, consistent data—it lays the foundation for transformational capabilities like AI. AI initiatives thrive on well-curated, high-quality data that is timely, reliable, and accessible. By prioritizing governance, organizations can move beyond tactical fixes toward scalable, strategic insights.

As the business landscape grows increasingly reliant on data, the organizations that succeed will be those that take governance seriously—not as a checkbox exercise, but as a strategic imperative. By addressing barriers and taking meaningful first steps, executives can position their organizations to leverage their data as a critical asset for sustained growth and innovation.

The FDA IT Strategy

For anyone operating in the life sciences area, the FDA IT Strategy is an interesting read.  The strategy outlines six top-level goals in developing a more enterprise-focused technology strategy, which can be roughly summarized as:

1. Creating a shared technology ecosystem across the organization designed to support communication, collaboration, trusted data, and transparency.

2. Improving the base infrastructure through cloud technologies, flexible service orientation, and zero-trust principles.

3. Cultivating scalable, cost-effective, tailorable, and secure enterprise services through stronger business alignment, digital transformation initiatives, and customer experiences.

4. Embracing data as a core asset, including data governance, literacy, accessibility, exchange, and advanced analytics.

5. Pursuing balanced, responsible, AI-based innovations that align to the organization’s mission.

6. Attracting, cultivating, and retaining the right leadership and workforce.

At first glance, these seem like the right set of priorities.  In fact, it would be difficult to debate most of them: would anyone argue for less modernized technology that ignores data, for example?  The list is certainly consistent with non-agency IT priorities – virtually all of these strategic goals could have been copied from the business and technology plans of the agency’s life sciences stakeholders.  Regardless of organizational size or mission, these technology forces are top of mind for many industry leaders today.

The FDA IT Strategy has several recurring themes that are worth calling out.  First, there is a strong recognition of the need for stakeholder engagement and alignment.  That point is particularly important for the second recurring theme, which is around business process and service scalability.  If the agency is reflecting the broader IT ecosystem, leaders see process orchestration, automation, and streamlining as critical enablers for capturing economies of scale.  And, of course, a third recurring theme is unleashing the opportunity in AI, analytics, and data.

Artificial Intelligence and the Evolving Policy Landscape

Acknowledging the logic of the proposed plan, perhaps the largest issue facing the agency today will be more cultural than technological: how do you make an agency with the size, complexity, and longevity of the FDA agile?  That question has been asked before (i.e., combinatorial chemistry, precision therapy development), though it seems clear to most industry insiders that the pace of innovation in areas like AI is far greater than the ability of any federal agency to respond effectively.

But that’s not for a lack of trying.  The fall of 2023 has seen a flurry of policy and regulatory activity related to AI.  On October 30, 2023, President Biden issued an Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence, the most significant executive branch foray to date in introducing AI safety and security measures into public policy. The proposed actions are primarily focused on mitigating the rapidly emerging risks associated with AI, addressing issues such as transparency, labelling, testing, and standards development.  From a life sciences perspective, the order was notable for calling out issues such as stronger provisions against algorithmic discrimination; protections for the engineering of dangerous biological materials; promoting AI research in healthcare; strengthening consumer privacy protections; and ensuring the safe and responsible use and reporting of AI in healthcare and drug development.

That White House Executive Order was followed two days later by the Office of Management and Budget’s release of a new draft policy on Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence.  The OMB policy, which is open for comment through December 5th, 2023, is intended to offer richer guidance to federal agencies regarding AI strategy development, governance, responsible use, transparency, reporting, protections, and risk management.  In parallel, NIST continues to advance their AI Risk Management Framework, and all of these activities are advancing in the context of the White House’s Blueprint for an AI Bill of Rights.

This oversight momentum is not specific to the US.  The European Union’s proposed AI Act offers a tiered risk model for AI-related regulatory obligations, and has already garnered significant traction.  In addition, the Group of Seven industrial nations produced an 11-point voluntary code of conduct regarding the use of AI. The code “aims to promote safe, secure, and trustworthy AI worldwide and will provide voluntary guidance for actions by organizations developing the most advanced AI systems, including the most advanced foundation models and generative AI systems.”  Interestingly, the stronger policies surfacing in western regulatory environments are not consistently mirrored in some Asian territories where countries are adopting a more relaxed posture.

Building a Strong IT Strategy

Forces like these place new obligations on the FDA to adopt technology strategies that strike the right balance between innovation, standardization, safety, quality, and scientific rigor.  It’s a tall order to be sure. Alongside other federal agencies, DHHS has provided a Trustworthy AI Playbook alongside an inventory of over 160 existing AI use cases, over 40 of which are specific to the FDA.  The question for FDA technology leaders – like their non-agency counterparts in the industry – is how to develop the right data and analytical competencies that can support the rising tide of AI and data sciences innovations and associated medical advancements.

The industry is clearly interested in the answer to that question.  The request for comments elicited feedback from a variety of organizations including Google, Health Verity, McKesson, The RWE Alliance, and others.  The nature of the comments varied widely, ranging from poorly masked advertisements to deeper dives into important technology considerations like APIs and data standards.

One of the more notable responses came in the form of a combined letter from BIO and PhRMA which echoed some of my own reflections on the document.  As a strategist looking at the FDA IT Strategy document, there were five areas where I felt the strategy could be improved, especially given the needs for AI-related agility and the themes above:

1. Connect explicitly to business goals.  The best IT strategies directly connect their goals and tactics to corresponding business objectives.  By illustrating what IT investments are supporting which business goals, leaders can discern whether planned efforts are likely to deliver the needed results.

2. Clearly define current and future states.  BIO and PhRMA shared that the strategy as stipulated reads more like a list of principles, and I think that’s right.  Effective strategies are based on an unambiguous, detailed description of both the current and future states of the organization.  What specifically are you committing to build, improve, re-engineer, etc.?  Those more detailed commitments should deliver business value that can be measured, and they offer opportunities for cross-functional (and cross-industry) alignment. This is especially important in the space of AI, where the deafening roar of hype can easily obscure the real opportunities for impact.

3. Make Goals Empirical.  Strategies can often read as a list of attractive aspirations – sometimes called “motherhood and apple pie” – that don’t seem actionable.  When strategic goals have measurable performance metrics associated with each goal, it becomes easier to align investments and tactics to drive their growth.  How do we know if our IT strategy is being successful in year 1?  In year 2?  In year 3?  What is the benchmark from which we are measuring improvements?  These questions might seem cumbersome in their detail, but they actually help focus attention on the most value-creating activities.

4. Include more “how.”  A good strategy helps you know when to say “no” to the continually emerging landscape of attractive opportunities that risk diluting resources.  We don’t just want to know what is “good”; we also want to know what is “important.” By intentionally selecting how a strategy is being pursued, organizations can more easily identify what activities and investments align with a focused execution of the strategy. For example, much like an electric car needs batteries, a comprehensive and effective data strategy would seem to be a critical element of the FDA’s growth plans. We often frame these elements of a strategy as strategic imperatives: the areas of operational focus and excellence required in order for a strategy to be successful. The FDA’s plan covers the need to share, govern, and improve the accessibility of data. But it does not reflect an explicit priority for a data strategy and architecture that directly address complex industry business process use cases such as cross-study data analysis; rapidly harnessing existing data for AI model training; near real-time industry data sharing with the agency; and other improvements in electronic information structures and submissions. These potential process automations can improve the timeliness of regulatory reviews and approvals. A more detailed version of the agency’s own Data Modernization Action Plan could potentially address these issues in the future, creating an opportunity for life sciences stakeholders to align their enterprise architecture plans and standards to better support the agency’s operating model.

5. Incorporate timetables.  Strategies are usually not executed in weeks or months.  Typically, there should be a logical progression – a roadmap of milestones, phases, or success measures – that establish the expected implementation of strategic plans and associated improvements in performance.  In fairness, the strategy document refers to a future deliverable that will offer more of an IT roadmap, but ideally such phasing would also be reflected at this level of planning granularity as well.

Interestingly, much like the strategy itself, these improvement opportunities also commonly show up in the life sciences stakeholder organizations – so much so that part of the ISEP strategic planning methodology we use is designed to proactively address them.  And to be sure, the FDA’s CIO Vid Desai and the entire team at the agency should be commended on building this strategy.  It’s advancing an important conversation that can directly impact the speed at which patients around the world receive tomorrow’s life-saving therapies.  I’ll be excited to review the next iteration.

Life sciences and healthcare are embracing AI as a fundamental tenet of medical innovation.  And the past year has produced an unprecedented level of interest in AI applications, especially “generative AI” techniques found in solutions such as ChatGPT, GitHub Copilot, Bard, DALL-E 2, Midjourney, and other offerings.  Due to the sophisticated functionality and broad availability of these products, they are having a disruptive effect on many life sciences organizations, introducing a wide array of risks that bad actors can exploit.

Nine AI-related Cybersecurity Risks

Predictive analytics, including artificial intelligence and machine learning methods, are far from new.  Given the life-critical nature of analytical and decision support applications within life sciences and healthcare, AI risk analysis in these industry settings often focuses on the AI model’s data and resulting performance (e.g., biased or erroneous data, ethical considerations, incorrect and opaque algorithms, improperly tuned models, AI hallucinations).

For cybersecurity and risk management professionals, many emerging risks are also related to how users – both employees and external stakeholders – are adopting and exploiting these new technology capabilities.  If you think the risks are not that notable, consider that Apple, Samsung, Amazon, Verizon, Citigroup, Goldman Sachs, Wells Fargo, and Accenture among others have already restricted the use of these technologies.

What risks are they concerned about?  Broadly speaking, AI-related risks commonly fall into nine categories:

1. Loss of Intellectual Property.  In a recent survey, approximately half of the senior executive respondents believe their company data has been inadvertently shared with ChatGPT.  Employees using generative AI tools – especially those hosted outside their corporate infrastructure – may provide copies of corporate data as required inputs to the AI system without realizing that they are disclosing confidential information to an unauthorized vendor.  Such information is then subject to further disclosures through system hacking, internal product vulnerabilities, or system use by competitors.  And when such information is governed by regulations, the risks increase further.

2. Contractual and regulatory violations.  Disclosure of some information to AI systems may also constitute violations of contractual terms or regulatory obligations. For example, most companies have non-disclosure agreements in place with their data providers.  And data critical for the development and training of many industry-specific AI algorithms – patient data, for example – is protected by HIPAA which restricts both the use and disclosure of that data.  This issue is particularly sensitive when dealing with any data that carries privacy concerns. 

3. Social engineering.  Generative AI tools can be leveraged by attackers to engage in social engineering attacks, tricking users into revealing sensitive information or performing malicious actions. Because the AI models can conjure highly realistic imitations of individual leaders and employees within an organization, these phishing attacks are harder to detect by the average user.  Attackers are increasingly running their email campaigns through natural language models to eliminate broken English and misspellings which have been a tell-tale sign of phishing attempts.  And with a few recorded minutes of an individual’s voice, AI voice generators can recreate any voice message desired (e.g., this is the CEO and I need a favor urgently for this afternoon’s board meeting).

4. Copyright infringement.  Generative AI models are trained on existing information works that are usually copyrighted.  Currently, the boundaries concerning the appropriate use of copyrighted works as a foundation for new works are unclear.  If a computer generates a new work by mimicking an existing work, is the new work infringing on that existing work’s copyright?  Like all things related to copyright law, these issues are expected to take considerable time to normalize.

5. Larger pool of stronger adversaries.  The broad and open availability of sophisticated AI tools creates more opportunities for new bad actors (e.g., hackers, malicious content creators) and software threats (e.g., malware, bots) to emerge.  Individual hackers, groups, extremists, and geopolitical / nation-state actors have more tools in their toolbox today.  These tools can increase the effectiveness of their attacks (e.g., AI-driven vulnerability and attack planning), and are increasingly easy to use by novices.

6. Code vulnerabilities.  AI can also be used to analyze and identify vulnerabilities in 3rd party software code in web applications, websites, and network configurations.  This means that attackers can find and exploit zero-day vulnerabilities much faster.  In addition, most of the AI community leverages open-source software development tools and code libraries for developing new AI capabilities.  The continuously evolving state of these systems and code bases creates a potential exposure point for enterprises that introduce them into their infrastructure (e.g., software defects, malicious code insertion).

7. Newer forms of viruses and malware.  Any machine infected with malicious code is obviously a threat.  As viruses and malware increasingly incorporate AI capabilities, these threats will become increasingly sophisticated, able to learn from and adapt to behaviors and new experiences.  For example, malicious code can leverage AI to change its digital fingerprint, thereby enabling it to evade detection by security software.

8. Misinformation campaigns.  The ability of generative AI to create realistic imitations is not limited to social engineering hacks.  Competitors, activists, and disgruntled employees now have access to tools that can fabric audio and video assets that can damage an organization’s reputation and market valuation.  Though popular concerns about this issue have largely focused on political and regulatory issues, the risks are real for any business, organization, or individual.

9. Device vulnerabilities.  Specialized computing hardware, medical devices, and other solutions that include embedded AI algorithms need to be effectively managed over their lifecycle.  AI models need to be periodically tuned.  Operating systems and firmware need to be hardened.  Application software and interfaces needs to be patched.  Organizations that deploy these assets without having a strong lifecycle management process in place introduce vulnerabilities to their operational infrastructure as these technologies age.

Addressing AI-related Cybersecurity Risks

If the list of risks above seems daunting, it need not be.  There are proven solutions and practices that help organizations address those concerns.  Key components of an effective cybersecurity approach for managing AI-related risks include the following:

1. Cybersecurity Strategy: Do you have a strong security and privacy program today?  Are your threat operations positioned to respond to emerging risks?  For many organizations, increasing their readiness for AI starts with increasing their cybersecurity posture overall.  A well-formed cybersecurity strategy provides a solid framework for securing assets, managing operations, and responding to threats.

2. Enterprise Architecture: Every life sciences organization needs a strong IT strategy and associated enterprise architecture.  Well-managed enterprise architectures provide a wealth of cybersecurity benefits – minimizing security perimeters, maximizing protective measures, standardizing security methods, and more – that serve to make the organization’s infrastructure more resilient to attacks.

3. Data Mapping & Governance: What are your organization’s high-value data assets, and how are they protected?  How operationally do you mitigate the risks of accidental disclosure?  The answers to these critical questions vary widely by company, and effective answers require a deep understanding of each organization’s mission, products, operating model, customers, and regulatory requirements.

4. Risk Management:  Countering AI cyber threats requires smart risk management.  The landscape of risk is changing each week as AI innovations continue to advance.  Until cybersecurity technologies can catch up, the best defense continues to be good risk management practices.

5. User Awareness & Training:  The first, last, and best line of defense for AI-related risks is educated users.  Many leaders and front-line users are simply not aware of the risks associated with using generative AI tools and other emerging technologies. Though traditional cybersecurity awareness training does not cover AI, organizations like CREO offer specific training covering new attack vectors presented by AI and how users can help the organization manage the evolving new threats.

Finally, it is also worth remembering that AI is not just a source of risks – it’s also a highly valuable defense tool.  A modernized arsenal of cybersecurity technologies includes AI-powered capabilities in real-time risk assessment and quantification; real-world security telemetry, detection, and response; self-healing systems; endpoint intelligence; zero-trust asset management; and intelligent vulnerability prioritization and management.  When coupled with the practices above, life sciences and healthcare organizations can control the risks introduced by AI while also reaping the benefits of these powerful technologies.

If so, you may be in the minority. Data from multiple surveys of Fortune 1000 executives show that only around 20-25% companies report successfully establishing a data-driven culture.

The lack of progress is not a reflection of a lack of investment. New leadership roles – Chief Analytics Officer, Chief Data Officer – have emerged to facilitate value creation through better insights. A lot of industry attention is being directed on data literacy and self-service enablement as key to unlocking a more data-driven workforce. And company executives are frequently communicating to their investors that their strategies and investments reflect data- and insight-driven opportunities (e.g., AI).

So where is the gap?

Though corporate priorities may reflect data-oriented offerings and business processes, organizational plans often do not reflect the basic reality that building a data-driven culture is a change management challenge. Senior leaders need to agree on (and appropriately resource) an effective approach for transforming the culture, helping employees transition their activities and responsibilities from the “as-is” state to “to-be” state.

For example, is the organization approaching this transition through a specific program that addresses Kotter’s change management components (e.g., creating urgency, building change teams, maintaining change progress, etc.)? Or is the organization pursuing a more iterative, opportunistic approach such as nudge theory? The approach can be big or small, but being intentional about change management enables leaders to adequately address the most common barriers to adoption.

One way to think about the change management dimension is through the lens of trust. Trust is the cornerstone of any successful change management program. So what would it look like for an employee to trust a new data-driven insight enough to confidently change their behavior and/or mindset? In senior executive education courses I sometimes teach, we explore six areas where trust contributes to the development of data-driven cultures:

• Trust with Leadership – does this insight have executive attention and support? Employees tend to embrace insights that their leaders visibly value (i.e., model as important and safe). In a sense, there is a transfer of trust.

• Trust with Impact – does this insight matter? When insights are strongly aligned to actionable decisions and behaviors, they more reliably produce trustworthy results.

• Trust with People – is the insight driven by experts? Employees want to know that the people who best understand the problem space contributed to the insight and interpretations.

• Trust with Data – is this insight based on good information? If employees don’t have confidence in the underlying data source, they are much less likely to trust and embrace the insight.

• Trust with Development – are the data and its insights created using best practices? Employees know that data can sometimes be misleading; they want confidence that the process used to create the insight was well designed and well executed.

• Trust with Communications – is the insight clearly visible and understood? Employees cannot embrace an insight they don’t see or cannot interpret.

A few of these areas tend to receive more “airtime” in discussions than others; for example, the majority of executives do not trust their data today. The irony is that shortcomings within one area such as “trust in data” is often a downstream effect of deficits in other areas such as “trust in development.”

Where in your organization do you see problems with trust inhibiting the development of a more data-driven business and culture?

In his book “The Bomber Mafia,” Malcolm Gladwell describes how a group of World War II innovators sought to change wartime bombing tactics from broad-based destruction strategies to precision targeting of an adversary’s production capabilities. It is a great book that nicely illustrates how difficult it is to drive a paradigm shift, even one that saves lives.

The health and life sciences industries are facing a similar paradigm shift today. Traditional methods of developing novel therapies have relied on population-based approaches to diseases. But Industry 4.0 capabilities – artificial intelligence (AI), genomic sciences, Internet of Things (IoT) devices, cloud technologies, and more – are unlocking new opportunities for how we understand and treat medical conditions.

The Increasing Role of Precision

Gladwell’s book emphasizes innovations in precision for shifting the existing bombing paradigm. Today, the medical ecosystem sits at a precision transformation as well. Three forces – growing data volumes, better computational technologies, and advancements in biological sciences – are creating a rich landscape of innovation opportunities. Consider two examples:

• Over 340 products that include AI have been reviewed by the FDA, and hundreds more are already queued for review. These solutions all have one thing in common: leveraging data and AI algorithms to increase the precision of diagnostics and therapies. The market for analytics like this in health and life sciences is growing at over 25% a year, and RBC Capital has estimated that 30% of the world’s data is now related to healthcare — growing faster than any other industry.

• The COVID-19 vaccines you may have received from Pfizer-BioNTech were based on synthetic mRNA technologies using lipid nanoparticles. The nanotechnology – operating at a whole new level of precision scale – was critical to being able to deliver mRNA un-degraded into cell proteins. This precision-based approach is now being used for new treatments for influenza, shingles, and many other diseases.

So what does a transition towards greater precision mean for health and life sciences companies? We see three broad imperatives for organizations looking to lead in a world of precision medicine and data-driven operations.

1. Corporate Strategy & AI

Business strategies need to reflect that health and life sciences businesses are competing in an information economy. AI is changing:

• WHAT companies provide (e.g., targeted drugs; new regenerative medicine applications with 3D printed organs and prosthetics; novel cell and gene therapies based on gene editing; computationally designed antibodies; AI-driven diagnostics; evidence-based therapies; medical devices with embedded AI).

• HOW they provide them (e.g., novel data collection and curation; synthetic research arms; AI-driven drug design; predictive disease modeling; AI-based clinical decision support; automated diagnostics; remote patient monitoring; AI-assisted robotic surgery; clinical trial recruitment modeling and prediction; brain-computer interfaces supporting rehabilitation; new options for driving health outcomes, quality, safety, profitability, and patient engagement).

Since precision insights are fueled by excellent data, these business strategies are increasingly reflecting digital transformation as a priority. For example, data-driven companies take process automation seriously – it not only improves productivity and quality, but it also standardizes data collection. Building on this, market leaders are pursuing:

• Strategies and performance management plans that explicitly identify and measure how precision operations drive competitive differentiation and growth.

• Product and service offerings (e.g., product management roadmaps) that incorporate precision insights as key drivers of value creation with customers.

• Organizational development plans that identify a) key business process automation opportunities, and b) people development plans covering the breadth of skills needed for a data-driven operating posture.

• Data strategy and governance programs that address growing data volumes, depth, timeliness, quality, transparency, and actionability.

2. Technology & AI

Technology roadmaps are critical. Every health and life sciences organization needs a well-conceived enterprise architecture that can support a data-driven operating model and unique market value:

• Developing clear “as-is” and “to-be” enterprise architecture designs that align to the business strategy and unique precision characteristics of the organization

• Accommodating AI models that span both central computing architectures and autonomous edge-computing deployments (e.g., hardware chips optimized for federated AI computations and low-power consumption).

• Rapidly integrating new data sources and an increasing collection of interconnected devices that need to exchange data across networks and systems.

• Establishing and managing great “sources of truth” for the organization that provide confidence and trust in the organization’s data assets.

• Effectively deploying analytical insights into line-of-business settings, including no-code / low-code development solutions.

3. Leadership, Culture & AI

Leaders cultivate data-driven cultures. Despite talented people with good intentions, company cultures do not always promote data-driven work. For leaders looking to leverage AI and data as a differentiator, focus areas should include:

• Assessing existing barriers to data-driven approaches and insights within the organization’s current culture.

• Creating increased transparency and employee empowerment around data.

• Evoking discussion and collaboration across business, clinical and technology leaders around how analytical insights – designs, simulations, decision support recommendations – can best be incorporated into operations.

• Evaluating and monitoring the safety, efficacy, and appropriateness of AI-driven insights, especially when analytical models may be sourced from 3rd parties.

• Encouraging and rewarding innovation – even failures – to demonstrate the value being placed on discovering new paradigms for the organization.

Of course, “precision” is not the only trend driving medical innovation today. And technologies such as CRISPR are demonstrating how broad-based innovations can be applied with high precision as well. But AI’s role is clear: in the war against disease, the soldiers, their military units, and their smart weapons are all increasingly leveraging AI.

A Brief History of Untrusted Insights

Back in 2006-2007, Thomas Davenport wrote an HBR article and similarly titled book called Competing on Analytics (disclosure: he also generously wrote the foreward to my book).  His works captured an idea that had been gradually emerging since ERP, CRM, and business intelligence systems gained traction in the late 1990s and early 2000s: data and analytics were becoming competitive business competencies, and market leaders will be data-driven.  As Davenport shared, the early experiences of organizations on the data-driven path showed that trustworthy data and insights were a problem:

“Executives are aware of the [data quality] problem:
in a survey of the challenges organizations face in developing a business intelligence capability,
data quality was second only to budget constraints.”

– Davenport, T (2007).  Competing on Analytics,
p. 163, Harvard Business School Press.

Over 15 years later, organizations are still struggling to leverage trusted data to become data-driven.  And in many ways, the stakes are even greater today.  The volume of data we are generating continues to grow exponentially.  Regulatory policies related to data privacy, protection, and consumer rights have introduced additional obligations on organizations.  And the widespread availability of both data and software tools has dramatically lowered the barriers to entry for competitors, increasing the pressure on business leaders to “figure this out.”

As a result, many executives find themselves in the middle of a self-fulfilling prophecy: when leaders avoid data sciences investments because either the data is not trustworthy or the organization is not seen to be ready for better analytics, they guarantee the data remains untrusted and the organization remains unprepared.  As Robert Frost might say, the only way out is through.

Finding Value in Data-Driven Insights

The current fascination, fear, and hype associated with artificial intelligence have raised executive awareness about the opportunity for using data and analytics in new and powerful ways.  Leaders and their boards are faced with a fundamental question: what potential business value can be unlocked through investments in data sciences and analytics capabilities?

Experience has shown that building data sciences capabilities in the absence of a clear relationship to its expected business value is an excellent way to waste money.  But when those investments are targeted – when the scope of work is focused on specific business questions, workflows, and insights that generate business value – the probability of success grows considerably.

Within life sciences and healthcare, data and analytics are commonly associated with several types of business value:

1. Better decisions. One of the most important benefits that data and analytics offer is raising the quality of decisions that need to be made.  In our industry, data-driven decisions run a spectrum of concerns including:

• Clinical / scientific questions: issues related to patients, treatments, and research.

• Financial questions: issues related to revenue, costs, pricing, and reimbursements.

• Administrative questions: issues related to customers, employees, projects, processes, facilities, and supplies.

• Behavioral questions: issues related to activities, preferences and tendencies.

2. Faster actions with lower costs. Many insights are associated with actions that need to be taken – change study criteria, add manufacturing capacity, conduct a deeper quality review.  When these actions are taken sooner, organizations experience productivity gains.  In addition, as many insights need to be repeatedly created by different organizational teams, efforts that produce reusable data and analytical assets – effective data curation capabilities, for example – allow all teams to reduce repetitive, error-prone, and time-consuming data and analysis tasks and their costs.  And finally, some data-driven actions can also be automated, streamlining and optimizing the use of resources for more complex tasks.

3. Lower risks. Data and analytics can be especially useful for identifying emerging risks.  Is a patient’s condition declining?  Is study enrollment dropping below targets?  Does the current commercial pipeline support the future revenue goals?  A strong data and analytics strategy enables leaders to detect risks that compromise the most important aspects of business operations and patient care.

4. Better products and customer experiences. Many organizations are discovering that a well-conceived data and analytics strategy can be used to create highly differentiated customer experiences.  Customers value on-demand access to current information (i.e., what is the status), transparency with business processes (i.e., what steps have been completed), and streamlined ways of engaging with their partners (i.e., where is all of the information).  By bringing disparate information together in ways that are easy to consume, organizations can create compelling products and services that drive higher customer lifetime value.

5. Stronger security and compliance. Well-developed data and analytical strategies also pay dividends in regulatory and security management.  When organizations have more disciplined approaches to managing data and insights, it becomes much easier to manage security (i.e., who sees what) and regulatory (e.g., how are controls managed) obligations.  In addition, data and analytics can also be used to detect patterns of electronic behaviors indicative of security or regulatory lapses (e.g., anomalous system access, excessive data retrieval).

So if these are the types of value commonly available, how do you identify which investments to prioritize?

Aligning Corporate and Insight Strategies

Historically, investment decisions regarding data and analytics were pursued relatively tactically primarily through operational reports (i.e., from line-of-business systems) and data extracts (e.g., for manipulation in Microsoft Excel).  Today, we know those approaches do not scale: there are simply too many systems, data, quality issues, access control obligations, reports, and data consumers to serve.  In addition, organizations need insights for more than simply good operational oversight – insights serve a more strategic function for the business.

Excellent business strategies provide unambiguous focus – a lens for continuously discerning the “important” from the “good”.  The same disciplines we use in strategic planning can be incredibly useful in helping organizations plan investment strategies in data and analytics.  We can use strategic planning artifacts – goals, strategic imperatives, and corporate capabilities – to characterize and prioritize the internal (i.e., employee-facing) and external (i.e., visible to customers and partners) insights that best support the corporate strategy and growth goals. Ideally, we want to focus on investments that accelerate value creation related to our strategy.  Questions we might want to explore include:

1. Based on our business model and market differentiation, what data and insights are most crucial in helping us make the best decisions about our organization?

2. What data and analytical capabilities would make us more valuable and attractive to investors in this market space?

3. What insights would help us know whether our corporate strategy is successful?

4. What time-sensitive metrics would better enable us to capture opportunities for operational improvements as we execute our strategy?

5. How could data and insights facilitate a totally different customer experience with our products and services?

6. How could we use data to detect and manage emerging risks for us and our customers?

7. How will competitors likely use data, analytics, and insights to win business?

8. How do we know if we are meeting our regulatory obligations regarding data privacy, security, and protections?

Grounded in the business value opportunities mentioned above, these questions (and others like them) serve to orient leadership thinking about data and analytics as both foundational business assets and key organizational competencies.  They also help to triage insights that contribute to building trusted, data-driven cultures.

So what role are data and analytics playing in your business strategy?