That conclusion is wrong, and the same mistake is being repeated across global health, including in areas like dialysis and medical equipment, reinforcing a Catch-22: The business models capable of meeting enormous clinical need are capital-intensive and operationally demanding, which means they typically stall out or get absorbed into grant-funded programs rather than growing into commercially viable operations. Investors and other funders read that as evidence that there’s no money in the system to pay for the services, and so the financing those businesses need to reach commercial traction never arrives. 

Once that conclusion takes hold, it becomes self-fulfilling. As a result, businesses that might have thrived while improving patient outcomes never get the chance to prove themselves.

Medical oxygen is where we can show that this Catch-22 can be broken — that the money is in the system, and the business models to unlock it do exist.

Why the catch-22 can be broken for medical oxygen

After COVID, the assumption across much of the global health community was that the challenge of providing medical oxygen had been largely solved. Billions were invested to expand production capacity — from oxygen concentrators to large-scale generating plants — and infrastructure that was genuinely inadequate a decade ago became meaningfully better. Yet the Lancet Global Health Commission on medical oxygen security found that 91% of people in sub-Saharan Africa still lack reliable access when they need it. While COVID-era investments significantly expanded oxygen production, these efforts stopped short of getting it to patients’ bedside with guaranteed reliability.

The bottleneck is no longer production capacity, but the Missing Middle: the absent layer of local businesses that handle logistics, maintenance, monitoring, training, and clinical support between the plant and the patient. As HealthPort founder Dr. Aishat Adeniji put it in a recent podcast hosted by Oxygen CoLab, even a hospital with 10 cylinders in stock can’t guarantee that a patient will be able to get oxygen at 2:00 a.m. without someone solving for “the last metre, not the last mile.” The evidence bears this out: Independent research led by Karolinska Institutet and Makerere University found that oxygen concentrators supported through services-based models were functional 95% of the time, against around 25% for equipment under standard procurement (in which equipment is owned and maintained by the facility) — even on the same wards. Meanwhile in Uganda, oxygen-as-a-service provider FREO2 maintained 100% oxygen uptime at partner facilities despite 215 power outages across these facilities in a three-month period. This suggests that the difference isn’t in the equipment: It’s in who is accountable for keeping it running. 

Building that accountability is the way to break the Catch-22. And where it has been built, the results speak for themselves.

Working business models with paying customers 

Over the past five years we’ve been lucky to work with a handful of local operators who have implemented services-based models for medical oxygen delivery. In this approach, a local operator takes performance responsibility for the full delivery chain (equipment, maintenance, distribution, clinical training, back-up supply, remote monitoring), and bills the facility post-service for what its patients use. The facility focuses on patient care, while the operator is accountable for uptime.

This model works for health facilities and their patients, but it can be financially punishing for operators without sufficient scale — especially when their access to affordable capital is limited. These businesses face high costs, covering customer acquisition, equipment and material purchases, and ongoing operating expenses, but post-service billing means they can work for weeks or months before facilities pay. For their customers, this is a feature, not a bug: Post-service billing is good for health facilities. They pay for what their patients use, after the fact, with no money upfront — a benefit that’s particularly important for facilities with thin budgets, which is where most of the public-sector need lies. But the working capital gap it creates limits scalability because, in many low- and middle-income (LMIC) markets, high commercial interest rates mean capital financing costs force operators to price for margin and require payment up front. This pushes pricing beyond what most health facilities can afford, and the addressable market shrinks to the private facilities that can absorb those terms — while excluding the public facility market that could offer a pathway to greater scale, along with the patients with the most unmet need. Our supplier mapping in Nigeria found that roughly 83% of oxygen SMEs serve private hospitals and clinics — roughly twice the share serving public primary care or maternity centers. These numbers provide a very clear signal of how the capital environment shapes who these businesses sell to.

HealthPort has been running a services-based model across hospitals in Nigeria for five years, with support from Oxygen CoLab and others that has allowed them to price for affordability, rather than survival, from day one. According to the company’s operating data, this has enabled HealthPort to deliver oxygen services at price reductions of up to 70% relative to prevailing market prices, and oxygen availability at its customer health facilities has moved from roughly 10% of clinical demand to 100%, including during surge periods. Retention across health facility customers is above 95%, and demand is outpacing capacity, with a waitlist for providers in as-yet-unserved regions. Meanwhile, HealthPort’s revenue from hospital customers has grown sixfold over three years, doubling in 2024 and tripling in 2025.

HealthPort shows what happens when the right capital lets businesses adapt their models to health facilities’ needs and deliver at prices those facilities can absorb: Latent demand becomes paying demand. Once supply became reliable and affordable, clinicians who had been rationing an unreliable supply started making clinical decisions they had previously been unable to make, and oxygen use at HealthPort’s customer facilities increased up to threefold. 

Businesses like FREO2 and HealthPort have paying customers and growing revenue — these are not programs being kept solvent by donor grants (Oxygen CoLab’s support to these companies ended early this year). And our mapping suggests that other oxygen businesses could follow, with the right market conditions and the right capital behind them.

What catalytic capital has to do differently 

However, the businesses that could fill the Missing Middle in medical oxygen don’t fit the capital structures that currently exist. Development finance institutions have transaction economics that push toward larger deals, and a business at pre-scale stage falls well below the threshold where per-deal due diligence costs are recoverable. SME lending requires financial documentation, governance structures and collateral profiles that early-stage health businesses haven’t had the time or support to build.

But solutions to this mismatch exist, and volume guarantees and blended-finance vehicles have de-risked suppliers and changed buyer behavior in health markets before. For example, MedAccess and Unitaid’s work with the East African Program on Oxygen Access combined volume commitments with demand-generation support from the Clinton Health Access Initiative to scale local oxygen production and distribution across Kenya and Tanzania, crowding in capital from local and international investors. 

However, these instruments work best in the environment they were designed for: centralized procurement and large multinational deals. Extending them to reach decentralized, last-mile SMEs will require three shifts in how catalytic capital is designed: 

• Pooled due diligence: Due diligence costs need to be spread across portfolios of SMEs in the same sector rather than borne on a per-deal basis; the economics of small-deal investing are otherwise unworkable. 
• Working capital built in: Operators facing working-capital gaps need bridging finance to be built into deals from the outset — not just equipment financing, which is what most oxygen-related capital has addressed. Trust-based philanthropic funders (Segal Family Foundation and Ripple Foundation among them) have shown what this flexibility can look like in practice, providing the kind of unrestricted and patient support that allows organizations like FREO2 to build the commercial track record that lets them take on larger and more structured catalytic capital, the kind they could not have absorbed at an earlier stage.
• Bundled support: Technical assistance addressing financial systems and governance should be embedded in the investment itself, available as part of the deal’s structure rather than required as eligibility criteria the business must satisfy beforehand.

We’re not alone in seeing the need for such a shift. Kaodili Udeh, Head of Regional (Africa) investments at MedAccess described the logic in our recent podcast: “Instead of supporting one large manufacturer at a time, we can support a portfolio of smaller manufacturers within the same sector. And what this does is it diversifies risk, but it also reduces transaction costs, because there are commonalities between the different SMEs, and so you can spread the due diligence costs around.”  

These instruments are catalytic precisely because they’re designed to be temporary — to build the track record and scale that make commercial and domestic finance possible.

Why Capital Alone isn’t Enough to Scale Oxygen Markets

But even well-designed capital won’t get oxygen markets to scale on its own, because it’s the conditions businesses operate in that determine whether the unit economics work.

In Nigeria, imported pharmaceuticals enter at 0% duty and oxygen therapy devices at 5%. Seamless steel cylinders (the primary container for distributing medical oxygen), on the other hand, carry an effective tariff of 60%. This isn’t a result of some deliberate design; it’s simply a residue of a system that categorized cylinders as industrial equipment rather than medical infrastructure. Even with the right capital and the right delivery model, a business can still find its unit economics unworkable inside that tariff structure.

The same challenge extends to regulation and market intelligence. Many regulatory frameworks written for industrial gas producers create market entry costs that medical oxygen SMEs can’t absorb. Nigeria’s National Agency for Food and Drug Administration and Control is responding directly to this need, evolving its capacity for the regulation of medical gases to facilitate the appropriate entry of new businesses into the market. In terms of market intelligence, this challenge manifests as a tendency among investors to support the same small group of well-established businesses, since they lack visibility into how smaller or younger enterprises are navigating these market conditions. In response, supplier mapping (the kind of exercise that identified roughly 14 credibly investable businesses out of over 100 we assessed in Nigeria) is the type of work that makes it possible for capital to reach the right companies rather than the most visible ones.

The key to success in these and other efforts to create an enabling environment for medical oxygen is coordination: Catalytic funders must deploy capital alongside policy reform and market intelligence, while governments base their procurement efforts on maximizing uptime rather than just acquiring equipment. And each of these activities must be treated as part of the same investment thesis rather than a separate lane of activity.

A decision point for catalytic funders of medical oxygen

Medical oxygen isn’t the only area where enormous clinical need coexists with businesses too capital-intensive to reach scale without financing designed to fit them. The same pattern traps a whole class of services. Biomedical equipment maintenance fits it exactly: 40-70% of medical devices in LMIC hospitals are broken or unused, and though local service businesses exist, the financing has never been structured to let them scale. In dialysis, to take one example, fewer than 2% of people with end-stage kidney disease in sub-Saharan Africa currently receive treatment, and the operators who deliver this service reliably are locked into out-of-pocket private markets because no one has built the financing that could enable them to reach government purchasers.

Oxygen is where the working models are in place, the evidence now exists, and governments in countries like Nigeria and Uganda have shown that they have the political will to act. If catalytic capital can be structured to fit in this sector, the same architecture will become available in other health-related sectors that are dealing with similar challenges.

The global health sector has been calling for a shift from aid to investment for years; designing the right type of capital is the last piece to executing that shift, and the success of this effort will determine whether this new investment-centric model is successful. That is the challenge — and the opportunity  — facing catalytic funders now.

Disclosure: Both authors are members of the Oxygen CoLab team and employed/contracted by Brink (part of Africa Practice), which supported HealthPort and FREO2 with grant funding and technical assistance, and worked directly with NAFDAC, Karolinska Institutet and Makerere University on the projects described in this article. Oxygen CoLab was funded by the UK’s Foreign Commonwealth and Development Office between 2020-2026.

Alex Losneanu is Innovation Director at Brink; Jason Houdek is an independent global health consultant focused on medical oxygen access and health-market development in sub-Saharan Africa.

Photo credit: Three Spots

Despite this drop in direct grants, official providers and multilateral development banks managed to mobilize a historic US $77 billion from the private sector in 2024. And broader multilateral development bank mechanisms have already pushed total private capital mobilization in developing economies past $100 billion, reaching $108.7 billion that same year. Yet this has not made everything easy for grant-seeking organizations. Private donors are far more demanding, often insisting on unique and measurable value propositions before capital is unlocked.

The emergence of precision-based capital

These changing funding realities are evident in how private capital is increasingly bypassing broad, open-ended thematic areas like public health or universal education, zeroing in instead on high-impact, technical interventions. Where government aid once funded the construction of a village clinic or the creation of a gender equity campaign, a private donor now funds the deployment of an AI app that detects tuberculosis just by listening to a cough on a mobile phone, or a digital passbook that makes government subsidies more easily accessible for marginalized girls.

Skeptics raise some valid concerns about this shift. Focusing strictly on niche projects might overlook the complex, interconnected realities of poverty. There is also a real danger that chasing easy-to-measure metrics will wipe out the deep, holistic community work that has always been the development sector’s forte. Besides, this shift is not seamless, as many NGOs and traditional field teams lack the specialized skills or inclination to adopt such corporate frameworks.

However, the fact remains that the overall pot of official development assistance is rapidly shrinking. This fundamental restructuring of global finance often makes the traditional, more comprehensive approach to development mathematically impossible. For organizations seeking to unlock private capital, targeting high-impact, outcome-driven interventions is not about abandoning intersectionality or ignoring the overlapping causes of social issues. Rather, it is about identifying strategic leverage points within a complex system to catalyze broader development.

Embracing the shift to private funding priorities

With their typical commitments to funding broad thematic areas, most development organizations — from large UN agencies to small community-level NGOs — tend to cover a wide range of activities. While this inevitably causes duplication, it is also seen as a prerequisite to ensuring substantive impact. For instance, while one UN agency might be the dedicated lead for reproductive health, several others will invariably include overlapping mandates like safe motherhood within their expansive portfolios.

These wide-spanning mandates of donors and fund-channeling agencies have, in turn, pushed implementers in the field to expand their own bandwidths as well. For decades, multilateral institutions’ request for proposal processes have been known to prioritize comprehensive, multi-sectoral approaches, inadvertently putting more specialized organizations at a disadvantage.

Now, as funding decisions become increasingly competitive and performance-driven, institutional donors and private investors are evaluating implementing NGOs less on the breadth of their mandates and more on the distinctiveness of the outcomes they can credibly deliver. Does this mean development organizations must narrow their scope of engagement? In many cases, yes — but not by compromising their missions. Rather, it means shedding activities that are not aligned with the outcomes they are uniquely positioned to deliver.

The methodology for achieving this can be found in key corporate strategies the development sector has long avoided. It requires understanding the principle of positioning: the ability to identify an unfulfilled, critical demand that maps directly to the organization’s offerings. And it demands a defined niche: the ability to demonstrate how those offerings fulfill the need better than any other entity within the larger ecosystem. While many in the sector have considered these concepts to be corporate jargon, incompatible with development-focused initiatives, it is high time we accept that taking some pages from the corporate playbook does not dilute a social mission: Instead, adopting these practices can enable the sector to stay relevant and survive.

Admittedly, this pivot is not without friction. The pressure of precision-based capital runs the risk of pushing NGOs away from their core missions, as they pursue more easily quantifiable metrics. Focusing on a niche project or goal that meets an unfulfilled societal demand and intersects with the organization’s core competency is an effective safeguard against this sort of mission drift. If an NGO can successfully execute this pivot, funders stop seeing it as just another interchangeable grant-seeker, and start seeing it as a proven expert in its field, whose specialized knowledge — backed by deep community ties, firsthand skills and a genuine dedication to the cause — make it a uniquely valuable partner.

Real-world examples of niche specialization

Consider the trajectory of Water.org, which perfectly illustrates how the corporate principles of strategic positioning and niche specialization function in tandem. In its early days, when it operated under the name WaterPartners International, the organization functioned much like a traditional NGO, relying on donor grants to directly fund and construct community wells — an increasingly unsustainable approach to scale through grant finance alone.

Establishing the positioning: The organization recognized the need to identify a specific, undisputed leadership area. So it pivoted from being a generic provider of rural water infrastructure to positioning itself as a pioneer in water microfinance. This required an honest recalibration of its portfolio; secondary, non-essential activities were shed so that all remaining community initiatives could be realigned to support this single, powerful identity.

Claiming the niche: To turn this positioning into a distinct reality that could attract private capital, it developed a highly specialized service called WaterCredit. Instead of asking donors to fund concrete and pipes, it asked philanthropic and corporate investors to guarantee microloans that empowered families to install their own water taps and toilets. This specialized financial instrument carved out the organization’s new niche: a distinct value proposition that turned vague philanthropic goals into a quantifiable social and financial return.

Critics often argue that precision programming of this type ignores the intersectional realities of development — that you cannot fix water access without addressing gender equity or climate change. However, claiming a niche does not mean denying this intersectionality; it means approaching complex systemic issues through a highly focused, measurable lens. WaterCredit drastically reduced the hours women spend fetching water, thereby tackling systemic gender inequities — but it did so through a targeted mechanism that private capital could actually underwrite. By turning water access into a data-rich financial asset like small loans from local microfinance institutions, with easy-to-track interest and repayment rates, it offered private capital a risk-return-impact equation that was clear, predictable and measurable at scale. Thanks to this shift from a broad mandate to a precise niche (along with the star power brought by co-founder Matt Damon), Water.org unlocked massive amounts of private capital, helping disburse US $8.2 billion in loans and reaching 92 million people.

Many others have also successfully adopted a strategic niche. For instance, Sanku was originally founded to sell a patented, specialized technology solution: a remotely monitored “Dosifier” that retrofits onto small-to-medium sized rural flour mills to automatically inject precise, safe amounts of micronutrients into flour as it is ground. When it later decided to pivot to a hybrid for-profit/non-profit model, rather than becoming a broad, multi-program NGO organizing an array of nutrition-focused projects, it built out its programming within this existing niche. It now provides millers with the tools, training, incentives and business models to fortify their food with lifesaving nutrients, creating a highly sustainable model that improves nutrition outcomes without raising consumer prices. Sanku’s specialized model is now reaching over 73 million people with fortified staples. It has set a new strategic target to reach 100 million people by 2028, and recently expanded with a nutrient premix factory in Ethiopia’s Kilinto Special Economic Zone.

New Story is another inspiring example. By moving from traditional construction to 3D-printed housing for low-income families, it created a high-tech niche that attracts private R&D-focused capital rather than relying on fluctuating aid budgets. Its innovations have reduced structural printing time to under 24 hours, compared to the weeks or months it can take to manually build a house. By treating housing as a scalable product instead of a manual project, the organization has helped house thousands across Latin America, and it built the world’s first community of 3D-printed homes in Mexico — at a price point that significantly increased the impact of every dollar spent. New Story was named one of Fast Company’s ‘World’s Most Innovative Companies’ for its work in de-risking R&D for housing finance and land development.

Clear value propositions are not just for established organizations. Whether it’s a rural women’s group painting murals in their village to combat domestic violence, or a bikers’ club sprinkling seeds to grow forests, any entity seeking resources for development or social impact work must prove its worth, and show why its work is special. The larger the entity, the more challenging this becomes — especially when it involves rationalizing a wide array of ongoing projects and activities.

The making of niche-focused NGOs

Transitioning from broad operations to a highly specialized niche is a fundamental pivot in both thinking and action. Such a realignment cannot happen in a vacuum, executed solely by NGOs themselves: It requires international policymaking and fund-channeling agencies to facilitate the process by earmarking risk capital and updating their selection protocols to favor specialized core competencies over generic, multi-purpose proposals. It also requires funders to help upskill NGO leadership and program teams in financial modeling, risk profiling and outcome-based design, so they can successfully migrate from traditional grant writing to structuring underwriteable projects. Instead of funding another generic project cycle, global fund aggregators like the UN need to finance development organizations’ core realignment towards Impact Breakeven, allowing NGOs to streamline their programmatic portfolios without facing immediate financial collapse. To make this feasible, one strategic linchpin could be the introduction of what I call “Transition Capital” — i.e., funds earmarked to cover the hidden costs of expert consulting and market analysis while an organization restructures its way of planning and operating.

Fortunately, the architects of global development finance are already setting the stage for this shift. This is visible in initiatives like the World Bank’s Private Sector Investment Lab and the International Finance Corporation’s highly targeted outcome bonds, such as the reforestation-linked outcome bond in the Amazon region, a specific financial instrument where returns are tied directly to audited, measurable success.

But despite this progress, a glaring gap remains. At one end of the spectrum, top-tier multilateral fund mobilizers — organizations that structure financial frameworks to attract private capital rather than deploying their own funds — are speaking the language of risk-adjusted returns. However, at the regional and country level, many organizations still remain anchored in conventional planning, where risk profiling is treated as a peripheral compliance checkbox. To bridge this divide, national governments and the regional hubs and local offices of multilaterals must evolve beyond their roles as compliance managers, acting instead as the strategic mentors required to help local implementers navigate this paradigm shift.

A high-velocity blueprint for funding this shift is already operational in India through the Skill Impact Bond: It pays implementing NGOs for highly specific employment-related outcomes, forcing them to shift from generic training modules toward focused, measurable execution models in order to get paid. With this instrument, the National Skill Development Corporation (NSDC) and its partners act as risk capital providers, offering upfront capital to implementing organizations. Investors recoup their capital, plus a financial return, from outcome funders when a program meets its rigorous, audited milestones on the ground. To date, the Skill Impact Bond has trained over 34,000 youth (74% of whom are women), achieving a 76% job placement rate and 62% retention rate, significantly exceeding national benchmarks.

Fund aggregators like the NSDC are ideally placed to promote this new mindset among NGOs. Sitting strategically in the middle of the aid hierarchy, global development banks, multilaterals and national agencies form a vital channel that can speak the language of precision-based capital while retaining deep community-centric intuition. Moreover, they have the scope to smoothly integrate tools for competitive analysis and market gap assessments into conventional grant templates, theory-of-change matrices and logical frameworks.

The aid era rewarded breadth. The era of precision-based capital rewards distinctiveness. Development organizations that can clearly answer why they are better positioned to deliver a specific outcome will successfully navigate this transition. More importantly, they are the ones that will define the next generation of social impact.

DISCLAIMER: The views and recommendations expressed in this article are solely those of the author and do not necessarily reflect the official policy or position of any other organization or individual.

Rajat Ray is a Social Innovations Advisor with over 40 years of cross-sectoral experience spanning multilaterals, international civil society organizations and multinational advertising.

Photo credit: peshkov

The workhorse of this market is the entry-level solar energy kit, a category that accounts for more than 80% of the industry’s affiliated sales: a panel, a battery, and a charge controller that powers lights and charges phones, which ESMAP’s widely used Multi-Tier Framework for energy access classifies as Tier 1 access. Because few off-grid households can pay cash upfront, companies sell these systems through pay-as-you-go (PAYGo) financing. The household gradually pays off the asset through small payments made on fixed schedules, but the model is deliberately forgiving: A customer who runs short of cash loses service until the next top-up rather than losing the asset. Sector analysts have identified this repayment flexibility as a core feature of PAYGo.

Yet even with these financing alternatives, affordability remains an issue. World Bank research finds that only 22% of unelectrified households can afford the monthly payment for a Tier 1 solar kit purchased via PAYGo.

The energy access conversation in Africa is usually framed around deployment. Flagship efforts such as Mission 300, the World Bank and African Development Bank initiative to connect 300 million people by 2030, are defined by connection targets, cost reduction and rural reach. These are necessary goals. But they address a logistics problem, and logistics is not where the model strains.

For companies working at the last mile, the binding constraint is the time it takes for a household to repay an asset. PAYGo may remove the affordability barrier for customers, but for solar businesses, it introduces a new challenge: the problem of time. A solar home system a household cannot buy for cash can instead be paid off over 18, 24 or 36 months. The price for the consumer is spread out, but the cost of waiting is added to the provider’s balance sheet — and from there it spreads to the company’s investors.

In my years leading at Izili Group, a last-mile energy access business operating in markets from Nigeria and Senegal to Madagascar, I watched this time transfer shape everything: company performance, investor conversations and the quiet pressure that builds between the two. Much of the capital that supports energy access presents itself as patient, willing to wait for these PAYGo assets to be repaid. But in practice, it often behaves otherwise.

Understanding the Mismatched Timelines in PAYGo Solar

The industry’s performance data tells a story of steady maturation. Benchmarks from GOGLA’s PAYGo PERFORM initiative show that customers repay an average of 72% of the financed solar system’s value at 2x the contract term. That is not a lender losing 28 cents on every dollar: Contracts are priced with partial repayment in mind, and GOGLA’s monitor shows leading firms strengthening their unit economics at exactly these levels. What the benchmark reveals is pace: The industry measures recovery at twice the contract term because that is how long the cash takes to come back.

These numbers describe a model that works, but slowly. PAYGo portfolios mature over time, shaped by irregular incomes and lumpy household expenses — e.g., the harvest that comes in late and the school fees due each term — each of which pulls cash away from their PAYGo installments for weeks at a stretch. GOGLA’s latest cohort analysis found that “time itself is a risk factor,” as longer repayment horizons expose companies and customers alike to greater uncertainty.

But though value in this business accrues over years, not quarters, the capital that’s financing PAYGo solar companies rarely exhibits similar flexibility. Loans to PAYGo operators typically come with fixed repayment deadlines and fixed interest, often in dollars or euros, while revenues arrive in naira, shillings or other local currencies. That currency gap carries real risk: When a local currency weakens, the company’s debt grows overnight because the money it’s bringing in from customers is worth less, even if collections stay perfectly on schedule. And when a portfolio’s repayment curve stretches after a poor harvest, the debt schedule does not stretch alongside it. The mismatch is structural: capital with fixed demands, financing a business with flexible cash flows.

This mismatch is dangerous because of how the model consumes cash. A PAYGo company pays for hardware today and recovers the money over years, so every new customer widens the gap between cash going out and cash coming in. GOGLA’s monitor shows that the median firm still operates at a loss, with financing costs rising. For these companies, continuous access to capital is not fuel for growth. It is an operating requirement, like diesel for a generator.

The Impacts of Short-Term Capital on a Long-Term Business Model

That is why the recent pullback in funding to the sector has been so damaging. Total investment in off-grid solar companies fell 30% from 2023 to 2024, to roughly $300 million, and although funding stabilized at $315 million in 2025, the number of companies receiving any investment fell 41%, from 97 to 57. GOGLA’s own analysis draws the causal line explicitly: It attributes companies’ exits from the market directly to the funding contraction, and mentions that other surviving firms are experiencing financial distress that may require them to restructure before they can raise funding again. Meanwhile, some key players have pursued consolidation, including Ignite’s acquisition of ENGIE Energy Access, one of the sector’s largest operators, and Izili Group’s acquisition of Qotto. And all of this unfolded while recent customer cohorts were showing improving repayment performance, according to the same GOGLA monitor. When the capital stopped, companies with strengthening fundamentals did not simply slow down. They disappeared or were absorbed.

The deeper problem is one of matching the investor to the product. Energy access in Africa has largely been financed like venture capital; GOGLA notes that the sector’s startup funding moves with wider African venture capital trends. But the business behaves like infrastructure: It resembles water systems, rural roads or telecom towers far more than software. The assets are tangible. Revenues are modest but durable. Social returns arrive immediately, while financial returns build slowly over years. Nobody expects a toll road investment to achieve an exit in five years, yet solar portfolios serving the same populations are routinely held to that timeline.

When capital built for quick returns meets a business built for long-term engagement, the business adjusts in ways that damage both its economics and its mission. Companies tighten credit approval prematurely, shrinking the customer base that installment financing exists to serve. They cut field service teams to reduce costs, even though service quality is what keeps customers paying. They withdraw from harder markets first, writing off distribution networks and customer relationships that took years and real money to build, and forfeiting the scale on which the model’s economics depend. Each decision is rational in the short term. Yet each undermines the company’s fundamentals because, in this model, installments drive demand, service drives repayment, and scale drives average cost down.

Capital that understands time

Better-aligned capital already exists, and it is instructive to look at how it is structured. Results-based financing pays companies for verified connections rather than promised growth, and it has moved from pilot to policy: More than $900 million has been committed to the off-grid solar sector, over half of it in the past few years. The largest single example is the $300 million off-grid solar component of Nigeria’s World Bank-backed DARES program. In procurement, All On’s Demand Aggregation for Renewable Technology program pools equipment orders and provides working-capital finance for distributors, reducing equipment costs significantly due to its concessionary financing solution and bulk pricing negotiations with suppliers. CLASP’s Productive Use Financing Facility uses targeted subsidies and grants to make income-generating appliances such as solar water pumps and refrigerators affordable, helping to align the amounts a customer repays with the earnings the asset produces.

However, these programs, valuable as they are, mostly restructure international and donor money. The more consequential shift is in who is providing the capital. GOGLA’s 2025 investment data shows local currency transactions reaching a record 47% of sector investment, with domestic commercial banks in Nigeria, Tanzania and Madagascar financing off-grid solar for the first time.

In Nigeria, InfraCredit’s local currency guarantees have mobilized domestic pension funds and insurers into off-grid energy bonds: By the company’s account, roughly ₦12 billion has been deployed across five local developers, reaching more than 28,000 beneficiaries. The logic of these investors is structural, not sentimental. A Nigerian pension fund holds decades of obligations in naira, so it can hold a multi-year naira asset to maturity without being forced to sell or refinance at a fixed date. A dollar-denominated fund holding naira-based obligations lacks this flexibility, because it earns its return in dollars while its portfolio companies earn theirs in naira. So if the naira loses value, this inflates the debt these companies hold, as the very currency their customers pay in no longer holds enough value to cover these obligations, even when collections continue on schedule. Additionally, portfolio companies in dollar-denominated funds must exit on a set schedule, too often precisely when capital has dried up and refinancing is hardest — which is exactly what happened across the sector during the funding contraction in 2024. African institutional capital is not inherently more patient or more generous. But it is structurally matched to the PAYGo business model: It lends in the currency customers pay in, on timelines its own liabilities can hold to term.

A growing investor class calibrated to PAYGo solar’s time horizons does not lower the bar for operators. Energy access companies must keep earning trust with customers and investors through service quality, transparent reporting and customer protection. The sector has built its own discipline mechanism for this: The PAYGo PERFORM standards, developed by GOGLA, CGAP and the World Bank Group’s Lighting Global program, give companies and investors shared definitions for measuring repayment and customer ownership, so that a portfolio in Dakar can be compared credibly with one in Antananarivo. But accountability runs both ways. For companies to build trust in difficult markets, the capital behind them must continue to flow for long enough that their efforts translate into repayment, ownership and sustainable operations.

The lesson from my years in this sector is not that it lacks “patient capital.” It is that much of the capital that is labelled as “patient” is actually not — a gap that is now visible in funding data, in company failures, and in markets quietly abandoned. The next phase of energy access will be financed by capital designed for the timelines the PAYGo model demands: blended structures that absorb early volatility, results-based funding that rewards verified outcomes over promised speed, and domestic institutions whose liabilities match the sector’s horizons. Electricity access in Africa is a public good delivered through private enterprise. If we want it to endure, the capital behind it must be built to wait.

Kolawole Osinowo is a Senior Research Fellow at the FATE Institute and a Public Voices Fellow Tackling Poverty, a partnership of Acumen and The OpEd Project.

Photo credit: Yuliia Kaveshnikova

Between the first loan we at Villgro guaranteed and the company’s eventual IPO, there was a credit ladder that BharatRohan had to climb rung by rung: accounts receivable financing from a second lender, a non-convertible debenture from a third, a follow-on loan from Caspian Debt without any guarantee, a warehouse credit line, and finally a bank overdraft provided at the lowest interest rate in the entire journey. Each rung was reached by successfully repaying the previous one.

While grants and equity dominate the conversation in impact finance, our experience shows that debt is often the most practical tool for enterprises that need moderate amounts of capital, i.e., between US $50,000 and $300,000. Debt has high potential to bridge this missing middle of capital needs — too large for a grant, too small and early for most equity investors. At the same time, debt providers face real and perceived risks around the creditworthiness of the sector or enterprise. That’s where a guarantee from a trusted intermediary to repay the loan in case of default can make that debt accessible.

However, across the 12 enterprises we initially supported through this model of guarantee-backed debt, we learned that a guarantee alone is not a silver bullet; the harder work is solving the problems that keep enterprises stuck. From our work, we have identified four major problems that must be addressed — two of which involve the enterprises, with the other two related to the lenders. I’ll highlight those challenges below, sharing some key takeaways for the ESOs and other intermediaries that provide debt guarantees and other support to these businesses.

Problem 1: Customers are not confident in the enterprise’s product

Guaranteeing loans to enterprises can only take you so far. For many of these entrepreneurs, who are often selling technologies that would increase customers’ long-term revenue in exchange for a high upfront cost, the product itself may need additional financing support to give customers the confidence to invest in it.

Take the case of Raheja Solar. They developed solar drying technology to help smallholder farmers — primarily women in cooperatives — preserve produce, as processed goods can earn significantly higher margins. While the model was effective in the long-term, farmers faced a risk that made them less likely to buy the dryers: They did not know if their increased margins would be enough to justify the investment. Since they had to take a loan to buy the dryers, they needed to have confidence that a market would be available to generate enough revenue to pay off the loan.

To address this concern, we structured a two-sided financing intervention: One element consisted of a guarantee-backed loan to the farmers and cooperatives from Samunnati, a lender specializing in farmer-producer organizations, so they could afford to buy the dryers. This was further supplemented by a guarantee from Raheja Solar to buy all the dried produce sold by the farmers, which was paid for by an additional Samunnati loan guaranteed by Villgro.

The results, according to Raheja Solar, included 36 solar dryers installed at six farmer producer organizations, 240 farmers with access to the technology, and approximately $450 in additional annual income per user. Raheja’s demonstrated model, along with Villgro’s guarantees, attracted two other Indian lenders who gave additional end-user financing. The positive reviews from pilot customers helped drive sales of these solar dryers in new geographies.

Takeaway for intermediaries: When an enterprise’s sales are stalled, the instinct is to find more capital for marketing. But working directly with customers often reveals that the barrier is market uncertainty; they do not know if they can recoup their initial investment in the product. We must solve this market confidence problem before financing can flow.

Problem 2: Enterprises reach for equity when debt would move faster

Impact enterprises at a growth inflection point often default to raising equity. The reasoning is understandable: Debt feels risky to entrepreneurs amidst uneven revenue, and equity investors seem like a more attractive option for securing growth-stage capital without the burden of monthly payments. What these entrepreneurs don’t realize is that debt can be a good choice, especially when time is of the essence.

SNRas Systems provides a good example of the advantages of debt capital. They developed the BlueBox, a nano-recirculatory aquaculture system that the company estimates can increase fish egg hatching rates from 70% to 95%. The enterprise had retail contracts, a clear path to expansion, and a strategic acquisition opportunity that would help them reach new supply chains and market segments. The only thing missing was $60,000 in working capital, and they needed it within weeks, not months.

An equity round would have taken far longer than the opportunity allowed. So in 2023, we at Villgro guaranteed a commercial loan from Caspian. SNRas used the working capital to complete the acquisition, expand into live fish transportation, and open four new retail stores. This unlocked exceptional revenue growth, from $72,000 in 2023 to over $1 million by 2025, according to information the company shared with Villgro. The enterprise has since raised over $1.2 million in debt and equity from multiple institutions without the need for a guarantee.

Takeaway for intermediaries: Rather than wait months for an equity round and miss a vital market opportunity, debt is faster, allowing the enterprise to meet key deadlines and reach scaling milestones that make them more attractive when equity investment becomes the right fit. Intermediaries may need to clarify the advantages of debt capital to entrepreneurs, addressing their concerns, explaining why equity isn’t always the best option, and helping them build a capital stack that’s suited to their current and future needs.

Problem 3: The lender perceives risk the enterprise doesn’t actually carry

Enterprises that are genuinely creditworthy but operate in sectors that lenders have no framework to assess face another challenge for accessing debt: For lenders, low visibility means high risk. They perceive the operational and financial risk of these enterprises as high and mis-priced because of their own lack of exposure to the market. As a result, they reject loans to entrepreneurs working in sectors like aquaculture engineering or biocomposite materials, because no one at the lender has ever underwritten debt to those types of companies before.

Take the example of Spectrus Sustainable Solutions: By 2023, they had been producing biomaterials from bamboo fibers and agricultural waste for eight years. They had a consumer brand on Amazon, inbound business-to-business orders from large corporates, two consecutive years of profitability, and over 1,000 product SKUs. But none of that mapped to a standard credit assessment, which prioritizes metrics like interest owed on existing loans, and current assets/liabilities, making it difficult for the company to obtain debt capital. Thanks in part to a default guarantee provided by Villgro, a lender, Nabkisan, was willing to provide a $125,000 loan. But what changed the lender’s understanding of the sector and willingness to stay engaged was our ongoing relationship with both parties: Villgro’s regular check-ins with Spectrus, aimed at documenting its repayment performance and addressing other potential concerns, and our explicit discussions with Nabkisan about how the enterprise’s risk profile compared to the lender’s initial assumptions.

Spectrus repaid the loan ahead of schedule. A major commercial bank, HDFC, subsequently extended a $300,000 unsecured loan at market rate with no guarantee required. The best part was that we were not involved in this transaction; Spectrus got the loan with a repayment record that spoke for itself.

Takeaway for intermediaries: A guarantee changes lender risk exposure, but it does not automatically change lender understanding. That requires deliberate engagement — repayment updates, risk profile discussions and sector context — throughout the loan cycle. It’s important to help lenders update their calibration of the sector based on evidence.

Problem 4: The loan sizes are too small to make it worthwhile for lenders

Even when an enterprise operates in a known and lendable sector, they may still receive a “no” if the loan size is too small. For lenders, the economics of small-ticket loans are unfavorable, leading them to focus on larger tickets.

BharatRohan faced this issue with its first small working capital loan of $30,000. For Caspian Debt, the operational cost and effort of underwriting did not justify approval, even with a guarantee. So Villgro assembled a complete due diligence file and shared it, along with a curated pipeline of lendable clients, with Caspian. The information reduced their operational costs, for the BharatRohan loan and more broadly, making it more feasible to offer a loan of this size.

We deliberately selected Caspian Debt because their ticket size range could accommodate BharatRohan’s future needs. That thinking paid off; as mentioned earlier in this article, BharatRohan received a second, larger loan from Caspian without needing a guarantee.

Takeaway for intermediaries: For small-ticket loans, it’s helpful to reduce the lender’s operational burden directly with a complete, curated due diligence file and a pipeline of similar enterprises, to reduce operational costs and make the economics work. While our role as ESO is to support social enterprises, lenders may also require support, as they must see a risk-return proposition that fits their mandate.

Conclusion

Across these enterprises, Villgro provided the same financial instrument, a first-loss guarantee, each time. What changed was the problem the guarantee was designed to solve — along with everyone’s willingness to design an intervention around the problem rather than around the instrument. 

Guarantees are not new and have been used for risk reduction for many years. Combining this well-understood instrument with ways to address the specific problem faced by the social enterprise is the key to addressing gaps that guarantees or technical assistance alone cannot bridge.

Srinivas Ramanujam is the CEO of Villgro Innovations Foundation.

Photo credit: PRASANNAPiX

This approach is superficially compelling, and even sounds a bit like justice: democratising the knowledge that rich farmers pay consultants for, and delivering it for free to a smallholder in rural Kenya on a $40 smartphone.

But on the ground, it doesn’t work that way. Even if we had perfect data (we don’t), and even if every farmer had access (they don’t), some key questions remain: What does AI advising in agriculture actually solve, who will pay for it, and more importantly, why do we keep designing AI to make humans redundant instead of more powerful?

The Version of AI for Agriculture That Gets Demoed at Davos

There is a dominant mental model of AI for agriculture. It is built in San Francisco or Amsterdam or London. It is demoed at Davos or COP or the UN General Assembly. It is sleek, frictionless, and built on an assumption so quietly embedded that most people never notice it — namely, that the farmer is alone: alone with her phone, alone with her data, alone with the algorithm, a singular rational actor receiving personalised recommendations and optimising her decisions accordingly.

This assumption is not malicious; it is genuinely well-intentioned. However, it means AI models are designed to remove the human from the loop with fewer intermediaries. From my experience, humans cannot be automated out. Instead, AI models should be designed around making the human more powerful, not less necessary.

What We Learned from 5,000 Women Farmers and a WhatsApp Bot

In my work with Kuza, a certified B Corp social enterprise that supports 1.2 million farmers through a network of more than 6,000 youth advisors in seven African countries, we’ve learned a lot about how best to leverage both human and digital resources to support smallholders. Kuza’s model connects farmers to advisory services, inputs, finance and markets through human-trained young advisors (known as agripreneurs) and through our digital platform, the One Network. One lesson from our work stands out above all: A trusted human is the key, and technology infrastructure is the enabler. Remove the human and the technology fails to deliver.

We ran an AI advisory pilot, supported by the Gender & AI Livelihoods program, that worked with 5,000 women poultry farmers in Kenya. These were the kinds of farmers impact-focused organisations typically target: excluded, largely subsistence-level female smallholders. They work in groups, with around 50-200 birds per egg production cycle, buying feed at retail prices, and selling eggs and chickens with limited bargaining power into volatile local markets. These are the exact farmers AI for agriculture claims it will transform.

We gave them access to Kuza’s personalised, WhatsApp-based AI chatbot, Ask Nia, which provided real agricultural advice, available on demand, expressed colloquially in common local languages. Our first concern, adoption, turned out to be surprisingly easy. The women farmers used it, adoption was widespread, and data analytics showed that engagement was positive. They were resourceful about sharing devices within groups, maximising usage time by letting their fellow group members know when others were done using them.

Adoption, however, did not take the form many AI solutions assume. It wasn’t individual; it was social. In fact, uptake worked, in part, because we started with existing social groups, led by trusted agripreneurs. Use was embedded in these group dynamics. Recommendations from discussions on our AI chatbot sparked multiple in-person conversations among users, who questioned, interpreted and sometimes set this guidance aside, with input from the agripreneurs. These recommendations were one input among many, filtered through collective experience rather than followed at face value.

What did not change was where trust resided. It stayed with the agripreneur, the group champion with the most presence, credibility and accountability. The chatbot could explain topics like Newcastle disease (a highly contagious virus affecting birds) or good breeding practices, but when a farmer was faced with consequential decisions, like whether to switch feed suppliers, she turned to someone she knew, not a bot she queried.

Indeed, the question of input costs offers a prime example of the type of issues that AI can and cannot solve. Feed accounts for roughly 70% of poultry production costs in Kenya. That single fact affects farmers’ operations far more than any advisory ever could. Over 90% of the participating farmers in our pilot program cited feed cost as their biggest challenge. Even a working chatbot — and Ask Nia did work — cannot overcome challenges that emerge from this underlying cost structure. Better information itself is not enough, if it cannot translate into better outcomes, and the business model for farmers remains fragile.

This is where much of the current narrative around AI in agriculture falls short. It assumes that information is the primary constraint. But in most of the markets we work in, the main constraints are structural. Input costs remain high, market access is inconsistent, cold storage is lacking and infrastructure is unreliable. Farmers are operating within systems where even correct decisions based on accurate information do not guarantee viable returns.

Where AI Actually Helped: Making an Invisible Market Visible

While our AI chatbot could not directly solve the challenge that actually mattered most to the farmers, the cost of feed, it did something that was arguably more interesting, and that ultimately enabled collective solutions to this challenge. It made a market visible that had long been invisible.

Individually, each of the 5,000 poultry farmers was too small to matter to any serious input supplier: They represented tiny, fragmented, unreliable sources of demand that no feed company would restructure its logistics for. However, our data — aggregated, analysed and made accessible through AI — told a different story. These 5,000 women together managed tens of thousands of birds on coordinated production cycles. This represented predictable demand, genuine volume — and from a feed company’s perspective, a customer worth having. Without AI, few businesses would have the capacity or resources to collect and analyse this data or recognise this opportunity. But with it, a new local market came into view, enabling providers to offer more affordable, bundled services.

As implementation progressed, the frame of the conversation between farmers and Ask Nia shifted from questions like “How do I get better advice about feeding my 100 birds” to topics like “Can we negotiate bulk feed discounts as a group?” Many of the agripreneurs, who provide their advisory services to farmers for a fee as an entrepreneurial endeavor, started exploring new business opportunities, (e.g., launching a small feed mill enterprise to serve this new market). Interestingly, when we put the question of how to reduce the cost of feed to the group, the responses were immediate and creative. Some farmers shared how they were experimenting with black soldier fly larvae to convert food waste into protein-rich animal feed. One mentioned a neighboring farmer, outside of the pilot program, who was producing his own feed in bulk. Someone pointed out that the vegetative waste most people throw away, when sorted correctly, is already animal feed. A number of the circular economy approaches regularly discussed in conferences began operating in practice.

The resource that unlocked all of this wasn’t algorithmic. It was human. These women had always been part of the broader agricultural market, although individually they remained invisible to it. AI did not create the demand that sparked the emergence of more affordable inputs. It made that demand visible enough to incentivise providers to start acting on it.

The Design Choice Nobody Names

Much of today’s AI debate is based on a long-standing design default: Technology creates value by reducing or eliminating human intervention wherever possible. From factory automation to self-checkout, ATMs and now generative AI, a recurring question has been: How can this task be performed with less human involvement? This is not true of every technology or every society, but it has become a dominant assumption in much of Silicon Valley’s approach to innovation.

Yet as Daron Acemoglu and Simon Johnson argue in “Power and Progress,” technology does not advance through an inevitable march of progress, but rather produces outcomes shaped by prevailing economic and social priorities. For decades, these outcomes have overwhelmingly favoured labour-replacing automation over human augmentation. The drive to remove humans from the loop is therefore not a universal law: It is a specific design choice.

Our work with poultry farmers shows that there’s another way. The agripreneurs we work with are not a cost to be engineered away. The group champion is not an inefficiency to be disintermediated. The trusted local advisor is not a legacy artefact waiting to be replaced by a chatbot. These people are the system. They carry the trust, the context and the relationships that make any intervention, technological or otherwise, actually work. Optimising agriculture in this context is not primarily an information problem. It is a trust problem, a market access problem, a collective action problem and a financing problem. And above all, it is a relationship problem: It needs the human connection, the social element.

This dynamic is already visible in how rural economies organise capital through informal savings and investment groups, where members regularly pool money and rotate access to the funds for business, education or emergencies. In Kenya, these “chamas” are widely used: They function without contracts or platforms, relying instead on trust, proximity and accountability between members.

Chamas reveal something fundamental about how decisions are actually made in these systems. Even where money, risk and productivity are involved, the binding constraint is not information but trust and coordination. New tools, whether financial or digital or AI, do not replace that layer; they are absorbed into it.

This is the lens through which we saw our AI pilot unfold. AI has a role in this sort of work, but the design question has to be changed: from “How do we use AI to replace the human intermediary” to “How do we use AI to make this human intermediary 10 times more powerful?”

AI can process the data that turns a group of 5,000 individual farmers into a viable, unified market for a feed supplier. It can flag when a flock’s feed conversion ratio is drifting, allowing farmers to intervene before it becomes a crisis. It can help an agripreneur serve 10 times more farmers without losing the personal quality of the relationship. And it can identify, at scale, which interventions are actually changing behaviour, and which are just being politely ignored. AI is a multiplier, but it needs someone to multiply.

A Different Design Brief for AI for Agriculture

The question every AI-for-agriculture team should ask before they write a single line of code for Africa is, “Are we designing to empower the human, or to remove them?”

If the answer is “remove” we need to be honest about it and own the choice, rather than portray it as neutral. If the answer is “empower,” these are some of our learnings about what actually works on the ground:

• Design for the group, not just the individual: African farming decisions are social. The unit of adoption is often the savings group, the cooperative or the family compound. An AI tool that works for one isolated farmer but breaks down in a group setting has missed this context entirely.
• Design to amplify the trusted human: The agripreneur, the extension worker and the group champion have legitimacy that no algorithm can purchase. Don’t cut them out — instead, give them better data, faster answers and tools that make them genuinely more capable. The goal should be to enable them to serve 2,000 farmers instead of 200, with no drop-off in the quality of their services.
• Design for structural constraints, not just informational ones: If the problem is feed costs, information is not the solution — demand aggregation, local processing and waste valorisation are. AI can help model and facilitate these, but someone has to go knock on the feed supplier’s door. That someone is a human.
• Design for data equity: Every interaction these farmers have generates data. That data should make their lives better, not just feed a platform’s AI model while they remain as marginalised as before. If AI is extracting value from excluded communities instead of adding it to them, we have reproduced the problem we claimed to solve.
• Design for financial sustainability: Ask who pays from the start. While the generic answer to questions about a new technology’s costs is that it will get cheaper over time as it scales, the proof of affordability is in the pudding, and depends on whether the farmer will pay for it. And farmers will only pay if they see value: increased incomes, reduced costs — something tangible in their bottom line. Too often, we expect farmers to adopt technology based on the promise of future improved productivity or incomes. They will not typically take that risk, as they are rational, practical decision makers.

The 5,000 women in our pilot didn’t need a better algorithm. They needed their collective scale to be visible to suppliers, and they needed the humans in their network to have the tools and data necessary to help them optimise their operations. AI helped with all of that *through* people, not instead of them.

This is the design choice that matters: Africa does not need AI that replaces the agripreneur. It needs AI that makes her unstoppable. That version is messier. It doesn’t demo as well. It requires you to understand the group dynamics of a local women’s cooperative, the trust dynamics of a rural agrodealer network, and the exact moment in a production cycle when a farmer most needs someone to pick up the phone.

The opportunity, therefore, is not to substitute these systems with AI alternatives, but to build on them. The future of AI in Africa will not be individual-first. It will be community-driven, youth-led, and AI-enabled, where technology strengthens the social fabric rather than replacing it.

Sheena Raikundalia is the Chief Growth Officer at Kuza.

Photo credit: LENblR

However, this approach is insufficient. With volatility in the recyclables market and almost half of all municipal solid waste having no value, according to our analysis at Waste Warriors Society, efforts to convert waste into value are unlikely to cover their own expenses, especially as new regulations like India’s Solid Waste Management Rules come into effect. Instead, user fees play a greater role in recovering collection and processing expenses.  

Collecting user fees from households is the exact course of action recommended by the Solid Waste Management Rules — but this is easier said than done. In low- and middle-income countries (LMICs) such as India, many residents are reluctant to pay to dispose of something they could simply burn or dump on the side of the street. A common refrain in these parts is: “We have very little waste, just a little plastic that we burn occasionally.” In these regions, changing people’s behaviour by convincing them to pay user fees and sort their waste properly remains a major challenge.

Waste Warriors Society has taken on this challenge in the Indian Himalayan region and lived to tell the tale — but not without battle scars. As a non-profit that prioritizes systems change, we build and manage waste collection systems with the aim of gradually handing them over to the community, to be run by local entrepreneurs or service providers. For the long-term sustainability of these systems, it is imperative to have a steady revenue stream from user fee payments. Through a combination of awareness campaigns, cleanup drives, stakeholder engagement and the provision of reliable services, our non-profit has managed to onboard and retain thousands of paying customers for our programs, involving the collection, segregation, processing and recycling of municipal solid waste at nine locations across the region. 

Customer churn, though, is inexorable and hard to understand. When we started out, we suffered our fair share of dropouts. Sometimes these happened across an entire village over only a few months. In other cases, we observed individual dropouts at locations where other households persisted, even without any change in the quality of service delivery or other differences between households. We realized that we needed to better understand the individual motivations for paying user fees, so we could prioritize our next steps to keep the balance sheet in the black.

To enhance our understanding of our customer base, we analyzed user fee records between April 2024 and October 2025, from 850 households across six villages near Dharamshala, in the mountainous state of Himachal Pradesh. We compared these payment records with survey responses obtained through interviews with both loyal and churned customers, aiming to learn the reasons behind their respective choices. Using the theory of planned behaviour framework to interpret our observations, we identified key takeaways for retaining paying customers, which can inform municipal solid waste management and other user fee-based models of social entrepreneurship.

Leaders and Community Members can Elevate or Diminish Participation

Across the six villages we studied, we found fairly similar customer attrition rates; where rates differed, a key explanation was the influence of leadership and community norms. As anticipated, a similar, small attrition rate was experienced in most wards over the 18-month period. One exception was in Barwala, a village with relatively lower income levels, where we charged 60% of our standard user fee. From six months to a year after onboarding, the attrition rate in Barwala was much more pronounced. 

One potential explanation for the steeper decline in customer engagement in Barwala was local opposition to our planned material recovery facility (MRF), which we proposed as a way to reduce collection and transportation costs for servicing this remote village. Local leaders mistook the planned facility for a dumpsite and actively discouraged households from participating in the waste management program. Despite this initial setback, Waste Warriors saw this as an opportunity to better explain our waste management services to the community. In response, we cancelled plans for the MRF, encouraged leaders to visit one of our existing well-maintained MRFs, and hired interns to lead community engagement campaigns. These campaigns spread awareness about the services, encouraging households to visit our MRFs at other locations and rejoin the waste management program. Slowly, these awareness campaigns led to a substantive recovery of enrolment in the waste management program in Barwala.

Figure 1. Effects of social norms on user fee payment trends. Percentage of users that paid user fees in each month from April 2024 to September 2025 in the Barwala Panchayat. Each line represents a ward.

In another village, Narwana Khas, a few miles east of Barwala, we noticed two wards where the participation rates were distinct from the five other wards in the village, showing somewhat more willingness to pay (see Figure 2 below). We learned that these wards are populated primarily by the Gaddi community, a close-knit ethnic group traditionally involved in nomadic grazing. Our interpretation is that some key community members participated in our services, leading to strong uptake through the rest of this community. As these observations demonstrate, social norms can influence participation in waste management programs top-down through leaders or by spreading through homogeneous communities.

Figure 2. Effects of group norms on user fee payment trends. Percentage of users that paid user fees in each month from April 2024 to September 2025 in the Narwana Khas Panchayat. Each line represents a ward.

Satisfaction with Collection Services is Key to Retaining Customers

While community-level norms and leadership explained some differences between wards, we also noticed variability between neighbouring households. To understand these differences, we compared survey responses from retained customers (>90% of months paid after onboarding) with those from churned customers (<75% of months paid). 

Retained customers tended to be more satisfied with the service provider’s reliability and communications. More specifically, these customers felt they could submit complaints to service providers and receive prompt redressal. Some even developed a strong enough bond that they began offering tea to waste collectors during their runs. These findings show that reliable and high-quality service, personal relationships and confidence in grievance redressal are important to customers and contribute to customer retention. 

Figure 3. Attitudes towards the service managers. Percentage of users among Sustained Participants, Intermittent Participants and Discontinued Households that responded “Yes”, “Sometimes” or “No” to whether the “Supervisor promptly resolves complaints.”

Insufficient Participation and Lack of Visible Cleanliness Outcomes Cause Customer Churn

Another challenge we noticed was a collective action problem. A common complaint raised by churned customers in unstructured interviews was that their immediate surroundings remained littered with waste, despite their participation in the collection program. Since they felt their fee contribution did not noticeably improve cleanliness outcomes in their neighbourhood, they stopped participating in waste collection. Some participants said they would not pay user fees until their neighbourhood was cleared of legacy waste, and care was taken to ensure that it would not become a dumping ground again. As one respondent, a retired army veteran, said, “I have made it clear to the person who comes to collect user fees: Until you stop people from dumping their waste in the field next to my house, I will not pay. It causes mosquitoes and a foul smell.”

Yet unfortunately, even when the program ran successfully, a few non-participants continued littering. Residents complained that, without full participation from the community, open dumping by non-participants in the program would continue. Multiple churned customers relayed some variant of: “We used to support your services, but one person can’t make a difference. Ask everyone to join, then we will also join.” 

On the plus side, many former users remained open to returning to the program if community uptake were greater: We asked five churned customers if they would resume paying user fees if everybody in their neighbourhood participated, and they all responded yes. 

Personal, Social and Outcome Perceptions Drive Household Participation

Development practitioners often attribute the lack of onboarding into environment-friendly programs to a lack of awareness. Business models for sustainability initiatives are typically designed around value propositions, and these programs may view complex behavioural changes through the simplistic lens of customer uptake, leading them to focus solely on service delivery or advertising. In reality, individuals are making choices based on multiple factors: Therefore, practitioners must adopt a wider lens that addresses the structural and social elements of this complex behaviour.

The theory of planned behaviour framework suggests that behaviours are determined by intentions, which in turn depend on attitudes, subjective norms and perceived behavioural control. Our analysis suggests that all three of these components influence customers’ likelihood to pay user fees for waste management services.

• Attitudes: Retained customers’ attitudes toward waste collection services tend to be positive, with high satisfaction around the grievance redressal process.

• Subjective Norms: Overall payment rates in villages drop when local leaders, whose opinions drive social norms, discourage participation. At the same time, community engagement campaigns that encourage people to take care of their environment can influence norms and improve participation. Similarly, social norms around whether the collective neighbourhood would participate in waste collection influenced individual participation rates, and shaped customers’ opinions of whether their efforts would be worthwhile. 

• Perceived Behavioural Control: When individuals feel a behaviour change is easy and can influence outcomes, they are more likely to make that change. As mentioned above, in our interviews, churned customers frequently cited the lack of improvement in neighbourhood cleanliness as a reason for not participating.

Strategic Interventions for Sustaining Community Participation

Analysis from this 18-month study has informed our action plan to continue iterating on our systems. We plan to focus on the following elements to increase community participation, reduce churn and improve user fee collection: 

• Maintain service quality to keep customers: When waste collection becomes irregular (e.g., in inclement weather or during the transition period when operations are handed over to local entrepreneurs or service providers), user fee payments often drop. Maintaining service quality is vital to retaining customers.

• Leverage social norms to improve participation: Social norms operate through multiple routes. Engaging with community leaders can have positive multiplier effects. Awareness campaigns appealing to morality can also help. Ultimately though, people tend to swim with the tide. When the majority is seen to be participating, it reinforces sustained participation and prevents churn. 

• Ensure visible cleanliness outcomes: Waste collection service providers typically restrict themselves to collecting and processing waste. However, understanding customer expectations regarding outcomes from the service is important to preventing customer churn. In this case, that would mean allocating resources towards cleaning and monitoring neighbourhoods.

Leveraging Behavioural Science for a Holistic Approach

Governments and businesses often talk about the importance of increasing the ease of doing business and removing regulatory barriers to encourage investment and promote entrepreneurship. However, in sectors like waste management, where operations depend on consistent user fee payments, the more immediate priority is to create the right conditions for individuals to participate. Predictable revenues make for more attractive investments.

Humans, as complex beings, make choices through a combination of personal and social beliefs. For that reason, we have focused on engaging people’s attitudes, subjective norms and perceived behavioural control to sustain behaviour change. Providing easy-to-use services and maintaining quality are central to our approach. Yet high-quality service can only go so far when the initial conditions (i.e., legacy waste) and the need for continued monitoring go unaddressed. Raising community awareness of programs and addressing concerns can facilitate uptake and retention. But even if they do everything right, well-intentioned programs may fail if they do not engage key members of the community or showcase membership to induce further participation. Hopefully, industry practice and behavioural science can come together to allow consistent progress in this sector.

Towards that end, we are currently collaborating with academics and behavioural science practitioners to design pilots where we integrate our learnings from this analysis. These programs could involve cleaning up legacy waste and setting up neighbourhood monitoring programs to prevent littering, among other potential approaches. Additionally, these pilots will aim to engage community leaders early in their outreach campaigns and post visible markers of participation, to help promote changes in social norms around waste management.

Integrating such approaches with a continued focus on service quality could yield improvements in user fee payments across households and make waste management financially sustainable in rural communities — not only in India, but across other LMICs.

Vishwanath Varma is a researcher, Archana Masih works in communications, Diksha Rana is a Research Consultant, and Ashish Kapil leads the Monitoring & Evaluation initiatives at Waste Warriors Society.

Photo: A resident emptying their trash during collection. Credit: Waste Warriors Society

Source: India’s AI Impact Startups, Kalpa Impact

In the run-up to the India AI Impact Summit hosted this February in Delhi, Kalpa Impact conducted an analysis of 97 startups and 13 non-profits deploying AI for social good at population-scale. We sought to understand who is building these tools, where they are being deployed, and most importantly, if there is evidence of impact.

Our analysis culminated in the launch of a report on India’s AI Impact Startups, published jointly by Kalpa Impact and India AI.

A few key insights emerged. For instance, of the surveyed startups and non-profits, a majority were building in HealthTech, AI Infrastructure, Climate Tech, EdTech and AgriTech. A further breakdown can be found in the chart on the right.

We also found that global tech giants like Facebook, Amazon and Google do not dominate the impact AI space in India. Instead, the ecosystem remains decentralized, with startups focusing on real-world opportunities to improve citizen lives.

We’ll share other learnings from our analysis in the article below, highlighting three key ingredients that can enable the successful deployment of impact AI in India and other emerging markets.

The First Ingredient: Localization

The first ingredient of impact AI — localization — refers to solutions which consider the entire ecosystem, are collaborative, and have a clear focus on addressing root issues.

India’s population is estimated to have exceeded 1.4 billion across 28 states and eight union territories, each with its own languages, customs and development challenges. From our analysis, we found that successful technology deployments build for these realities — for instance, by meeting the varied needs across both developed urban communities and rural and remote areas with poor internet infrastructure and zero to no internet connectivity. A one-size-fits-all approach falls flat in India’s diverse landscape.

Let’s take a look at AgriTech. India’s farms represent a strong opportunity to leverage AI for decision-making by integrating weather data, soil insights and market signals to empower farmers with guidance and unlock more efficient, resilient agricultural systems. Solutions like DeHaat, which provides AI-enabled insights and market access platforms for smallholder farmers, are revolutionizing the sector. For example, DeHaat delivers personalized recommendations on hyperlocal weather forecasting, soil analytics, produce quality grading and linkages to institutional buyers through its multilingual DeHaat Farmer app, combining digital intelligence with last-mile presence. The free app supports a reported 1.8 million farmers across 12 states, and its impact has been recognized by NASSCOM, Forbes and Niti Aayog.

The Second Ingredient: Building for Edge and Offline Use

The second ingredient of successful population-scale impact AI interventions involves building for edge use and completely offline usage. In our sample, only 8% of startups prioritized edge AI, which processes data locally on the device rather than sending it to the cloud. This approach to offline AI is crucial for product adoption in Indian villages with poor to no internet connectivity, and in remote areas that often lack relevant internet infrastructure altogether. Edge AI has the added benefit of being significantly cheaper, and reinforces user control over privacy and security features. 

Netradyne, a fleet management system which operates across commercial vehicles from semi-trucks to heavy diesel fleets, applies this principle by moving processing from the cloud into the vehicle. Its fleet safety system uses on-device AI to monitor driving time and identify risk signals like fatigue or lane deviation, triggering real-time audio alerts for driver self-correction. 

Similarly, Pinky Promise, a HealthTech AI app, provides advice on women’s sexual and reproductive health for rural communities with low internet access. Built by a team of gynaecologists and other doctors, Pinky Promise has been recognized for its ability to provide specialized medical advice to women in areas where meeting a gynaecologist often means travelling to big cities, spending out of pocket and combatting the stigma associated with seeking out these services. Since its launch in 2022, Pinky Promise has reached 200,000 women with its services.

The Third Ingredient: Voice-first Access

In India, AI for social impact features a vital third element: voice-first AI access. For those who sit at the intersection of literacy and digital divides, voice-first AI design allows users to interact in a way that feels accessible, convenient and low-touch. For many developers in India, voice-first AI is at the point where it’s considered an infrastructure layer — i.e., a core product component across multiple use cases. In our report, 28% of early-stage startups utilized voice-first AI approaches.

Several organizations and government initiatives in India (e.g. Sarvam and Bhashini) are building voice AI models for their specific use cases and publishing them as open source for others to use. By doing so, these multilingual AI models become shared infrastructure, allowing interoperability and mitigating vendor lock-in.

One emerging use case for voice-first AI involves India’s justice system, which is constrained by long wait times, legal complexity, a dearth of linguistic information, and knowledge asymmetry. In response, JusticeTech applications like Jugalbandi’s free voice-first multilingual AI assistant help citizens understand and access government schemes, public services and official information. Jugalbandi works across 170+ government schemes and has been recognized by Microsoft CEO Satya Nadella and studied by Harvard Business Education.

The Keys to Scaling Impact AI Across Emerging Markets

The Indian AI-for-impact ecosystem is growing rapidly, and while the longevity and efficacy of each of these startups and non-profits remain to be seen, one thing is clear: There are proven approaches to promoting greater AI use and maximizing AI’s ability to make lives better across geographies.

We have found these three ingredients — localization, edge and offline AI, and voice-first AI — to be key for scaling impact AI in India. But these findings are not only relevant to the Indian market. Prioritizing these elements can help startups across emerging economies transcend regional, language and economic barriers to ultimately overcome systemic exclusion.

India’s approach offers a useful playbook that can be applied in other countries where access and infrastructure remain uneven. In these countries, the future of AI will be shaped as much by how systems are built and deployed as by the technology itself.

Sushant Kumar is the Founder and CEO of Kalpa Impact; Ananya Mukherjee is a senior consultant with Kalpa Impact.

Photo credit: SR Mahakhud

But as the twice-delayed compliance deadline looms at the end of 2026, a potential flaw threatens to undermine the legislation. 

To meet strict EUDR data requirements, major commodity buyers are turning to remote sensing and satellite AI to prove that their cocoa, coffee, palm oil, soya, timber, cattle and rubber are deforestation-free.

But what happens when an algorithm misreads the landscape? More importantly, what happens to the vulnerable small farmers who are abruptly cut out of the market due to this sort of error?

If the industry relies solely on automated compliance without human input and context, we risk creating a massive “leakage” problem, where deforestation-linked goods are simply rerouted and sold to countries with no anti-deforestation laws. Overreliance on algorithms may ultimately clean up European corporate markets and spreadsheets while quietly pushing smallholders and deforestation itself into the shadows of the grey market. That isn’t the intended aim of EUDR, but it’s a very real possibility. 

The Limitations of Automated Compliance

The Earth Observation sector is in the middle of a golden age, as satellites and data analytics tools enable canopy and other landscape changes to be mapped from space with astonishing precision. For buyers hoping to avoid sizeable EUDR non-compliance fines, this capability offers a viable solution: They can now conduct risk mitigation without ever having to set foot in the plantation. 

However, satellite algorithms have a key limitation: they can identify changes in forest cover but cannot determine intent or causality, nor can they account for complex realities on the ground. 

In my work at MosaiX and the Earthqualizer Foundation, I bridge the gap between digital innovation and local realities every day. By helping global fast-moving consumer goods brands and supply chain partners navigate EUDR compliance using tools like satellite imagery and land plot mapping, I’ve seen firsthand the limitations of these technologies. When satellite data is processed blindly without human insight and local context, it oversimplifies complex geographies. For a small farmer wrongly accused of deforestation who lacks the resources to prove their innocence, a false positive can be devastating.

The Illusion of Compliance and the Threat of Leakage

Under the EUDR, the law is unyielding: Any plot of land deforested after the December 31, 2020 cut-off date is permanently barred from the EU market. And while the law states that companies must mitigate “non-negligible” deforestation risk before a product can enter Europe, it leaves the actual response entirely up to the discretion of the corporate buyer. Faced with a red pixel indicating tree-cover loss and no response protocol issued by EUDR, the path of least resistance for a corporate buyer is often to immediately exclude the supplier entirely rather than investigate. This creates the illusion of compliance, but it is a failure in the context of sustainable development. 

When small farmers are frozen out of premium, regulated markets, they do not simply pack up their tools and stop farming. Survival dictates that they find another buyer. So instead, they are often driven into “leakage markets” — i.e., regions or buyers with lower environmental standards, less scrutiny and lower prices. In these grey markets, deforestation continues while the farmers risk being pushed into poverty.

We cannot achieve environmental sustainability by sacrificing social equity. We need to design systems that keep smallholders and small suppliers included in sustainable markets and allow for their re-entry, rather than erecting algorithmic walls they struggle to climb.

Three Ways to Prevent Supply Chain Exclusion in the EUDR Era

To ensure regulations like the EUDR achieve their goals without over-reliance on satellite technology, we must fundamentally shift how supply chain data is managed and applied. Companies need to move beyond simple automated compliance and exclusion and prioritize social equity alongside environmental protection. Here are three steps companies aiming to comply with both the letter and the spirit of the law can take to bridge the gap between orbital tech and human geography. 

1. Anchor monitoring systems to accurate, up-to-date baselines: Companies must ensure that deforestation alerts are generated from verified, current land-use baselines rather than outdated or generic datasets. Weak baseline maps lead to overwhelming false positives, as they struggle to distinguish between a protected natural forest and an active agricultural plot. This often results in misdirected resources and unnecessary supplier friction, ultimately penalizing those at the very bottom of the supply chain. Monitoring systems must also enable precise supplier attribution, with traceability that connects the full supply chain directly to a specific plot of land. Without this granular traceability, automated alerts drive inaccurate accountability — resulting in blanket bans on entire regions or cooperatives, rather than targeted corrective action.

2. Implement an alert response protocol grounded in human-verified data: While the EUDR mandates zero tolerance for deforestation, corporate buyers shouldn’t treat an algorithmic red flag as an automatic suspension. Detection should trigger a conversation and a field verification, not an immediate ban. Automated screening must activate a clear response protocol for high-risk tree-cover loss alerts. This protocol should include on-the-ground verification to establish the true context of the clearing — determining whether it represents an actual land-use conversion, a natural event, or third-party encroachment along disputed land boundaries. Ultimately, automated alerts flag risk; field verification determines accountability.

To execute this verification fairly and interpret satellite data accurately, companies must invest in localized partnerships. Collaborating with local implementation experts and community organizations provides the nuanced knowledge required to investigate alerts thoroughly, ensuring that strict regulatory compliance doesn’t come at the cost of unjust supplier exclusion.

3. Shift from supply chain exclusion to Recovery and Re-entry Programmes: Perhaps the most significant flaw in the current EUDR compliance landscape is the lack of a route to redemption. Fearing massive fines for non-compliance, risk-averse corporate buyers are reacting with blanket, permanent expulsions of the smallholders themselves. Instead, companies should look to collaborate and pioneer Recovery and Re-Entry Programmes. By providing a structured pathway for suppliers to acknowledge responsibility, restore affected areas and safely regain market access, these programmes transform a rigid compliance risk into a genuine opportunity for landscape-level rehabilitation.

Conclusion 

The EUDR is a landmark piece of legislation, and satellites and AI have given us indispensable visibility into our supply chains. But at the end of the day, they are only tools. 

Laws and satellites do not save forests; people do. Getting compliance data from space is an incredible first step, but it fails without accurate baselines and verifiable, on-the-ground action. 

Priscillia Moulin is Director of Strategy for MosaiX’s Europe branch and Senior Advisor to both Inovasi Digital and Earthqualizer.

Photo credit: Matthew de Lange