H Heuristics Digital Reports

H Heuristics · Digital Report № 2026-09 · September 2026

Scaling Low-Cost Adaptation Technologies Across Climate-Vulnerable Developing Economies

A diffusion framework for rapidly expanding resilience capabilities

Score the technology, not the country. Most of what looks like a delivery problem is a product-design problem — and product design is the cheaper lever.

AuthorHunter Hughes
InstitutionH Heuristics
Published12 September 2026
Report №2026-09
Reading time19 min

Abstract

Two explanations for the slow uptake of low-cost adaptation technologies are already in circulation. The first is cost, which is straightforwardly wrong for this class of goods: a reflective roof coating, a drought-tolerant seed, an SMS forecast and a cooking stove are cheap absolutely, cheap relative to the losses they avoid, and cheap relative to the hard infrastructure that dominates adaptation finance. The second is the delivery channel, which is much better and which an earlier report in this series argued at length. But a channel explanation locates the problem in the country, and it cannot account for the central variation: in the same countries, over the same mobile networks and through the same agent networks, pay-as-you-go solar reached millions of households while weather index insurance did not, and improved cookstoves were successfully delivered and then fell out of use. A channel cannot explain a technology that arrived and was abandoned.

This report offers a framework that scores the technology rather than the country. Five attributes do most of the work: divisibility, whether the good can be bought in increments matching an irregular income; trialability, whether it can be tried cheaply and reversed; observability, whether a non-adopter can see the benefit in an adopter; attributability, whether the adopter can tell the technology caused the benefit; and compatibility, whether it rides on existing practice or demands sustained behaviour change and maintenance. Scored across eight technologies, the resulting ranking corresponds closely to the observed adoption record. The report states plainly that this agreement is not evidence, since the framework was constructed with the outcomes known; its value is as a design checklist that can be applied to a product nobody has launched yet, and the proper test would be out of sample.

The framework's most consequential result comes from reading it by column. Adaptation goods are systematically weak on observability and attributability for a structural reason: their benefit is an avoided loss, which is a non-event. Mobile money delivers a transfer that arrives and solar delivers light tonight, but adaptation delivers a flood that did less damage than it would have — a counterfactual no household can observe and no neighbour can witness. Weather index insurance is the extreme case, since in most years nothing happens, which is the product working as designed, while basis risk means the payout may fail to arrive in precisely the year a loss occurs. The two best-documented failures bear this out. Hanna, Duflo and Greenstone tracked households in rural Odisha for up to four years after they received an improved cookstove: smoke inhalation fell initially, the effect had disappeared by the second year, and there were no changes in health outcomes or emissions, because stoves were used irregularly and not maintained. A review of twenty randomised evaluations finds farmers generally unwilling to pay market prices for index insurance.

Because attributes are properties of a product as designed and sold, they can be re-engineered, and that is cheaper and faster than the institution-building a channel-only diagnosis implies. Unbundle the cost to fix divisibility, as instalment collection did for off-grid solar, where the innovation was the payment mechanism and not the panel. Subsidise a first cycle to fix trialability. Bundle the invisible benefit with a visible one to fix observability — the stoves that achieve sustained use are sold on faster cooking and less fuel bought rather than on averted illness, which means the health benefit is the reason the programme exists and precisely the wrong thing to put in the pitch. Convert avoided loss into a payment that arrives to fix attributability. The framework then earns its keep a second time as a sorting rule: a good weak on divisibility and observability together is not a product at all, and early warning — indivisible, with a benefit that is a pure non-event, scoring 4 of 15 — should be publicly provided and measured by coverage rather than marketed for adoption.


Executive Summary

Cheap adaptation technologies with strong measured returns are distributed and then abandoned. The reason is usually looked for in the country. It is more often a property of the product.

FINDING 01

Score the technology, not the country

Within one country, on the same rails and the same incomes, pay-as-you-go solar spreads and weather index insurance does not; cookstoves are delivered and then stop being used. If the channel were the binding constraint, that could not happen. Five attributes of the product explain most of the variation.

FINDING 02

Adaptation's benefit is a non-event

Mobile money delivers a transfer that arrives; solar delivers light tonight. Adaptation delivers a flood that did less damage than it would have — a counterfactual nobody can see. That gives adaptation goods a structural penalty on the two attributes that drive word-of-mouth adoption.

FINDING 03

Product design is the cheap lever

Attributes are features of a product as designed and sold, and they can be re-engineered. Making a benefit observable, divisible or compatible with existing practice is far cheaper and faster than building the institutions that a channel-only diagnosis calls for.

An earlier report in this series argued that soft adaptation is diffusion-constrained rather than capital-constrained: the technologies are cheap, the returns are among the best in development, and what is missing is the delivery channel. That is right as far as it goes, and this report takes it as given. What it adds is that the diagnosis is incomplete in a way that matters for what to do.

Locating the problem in the channel locates it in the country — the missing agent network, the absent payment rail, the weak municipal institution. On that account two countries with the same channels should adopt the same technologies at the same rate. They do not, and neither do two technologies within the same country. Something about the goods themselves is doing work, and it can be named.

2.1bn
People still without clean cooking — the cheapest major adaptation and health technology available
Tracking SDG 7 (2025)
Year 2
When the measured smoke reduction from a distributed improved cookstove had disappeared entirely
Hanna, Duflo & Greenstone (2016)
20 RCTs
Randomised evaluations finding farmers generally unwilling to pay market prices for weather index insurance
J-PAL policy insight
+10–30%
Yield gains from digital climate advisory — a benefit the farmer can see in the harvest
Global Center on Adaptation

Put those four figures side by side and the pattern is hard to miss. The technology whose benefit arrives in the harvest diffuses. The technologies whose benefits are a respiratory illness avoided decades hence, or a payout in a year when disaster happens to strike, do not — despite being cheap, despite being distributed, and despite modelled returns that should make them irresistible.

Nobody can see the flood that did less damage than it would have. That is not a marketing failure; it is the product. The observability penalty

1. Not a Money Problem, and Not Only a Channel Problem

Two diagnoses of slow adaptation uptake are already in circulation. Both are partly right, and together they still leave the central variation unexplained.

The first diagnosis is cost. It is straightforwardly wrong for the class of technologies this report is about. A reflective roof coating, a drought-tolerant seed, a weather forecast delivered by SMS, a cooking stove: these are cheap in absolute terms, cheap relative to the losses they avoid, and cheap relative to the hard infrastructure that dominates adaptation finance. Whatever is stopping them, it is not the price of the unit.

The second diagnosis is the channel, and it is much better. Low-cost adaptation goods reach poor, remote, cash-constrained households only if something carries them the last mile: an agent network, a payment rail, an extension service, a municipal programme. Where those exist, cheap technology moves; where they do not, it stalls at pilot. This is the argument the earlier report made and it is well supported.

But it cannot be the whole story, for a simple reason. Hold the channel constant and the variation does not disappear. In the same countries, over the same mobile networks, through the same agent networks, pay-as-you-go solar reached millions of households while weather index insurance did not. In the same villages, cookstoves were successfully delivered and then fell out of use. A channel explanation cannot account for a technology that arrived and was abandoned.

What this report adds

If cost explains little and the channel explains some, the residual is in the goods. This report offers a framework that scores the technology rather than the country — five attributes that determine whether a cheap, useful thing spreads once it is available. The practical value is that these attributes are features of a product as designed and sold, which makes them a far cheaper lever than institution-building.


2. Five Attributes

Drawing on the diffusion-of-innovations tradition and on what the adoption evaluations actually find, five properties do most of the work.

Divisibility. Can the thing be bought in increments that match an irregular income? Lumpiness is close to fatal for a household without savings, and it has nothing to do with the total price. A solar system costing a month's income is unaffordable; the same system at a daily payment is not.

Trialability. Can a household try it cheaply and reverse the decision? A seed variety can be planted on one plot beside the usual one. A raised foundation cannot be tried at all; it is built or it is not.

Observability. Can someone who has not adopted see the benefit in someone who has? This is the mechanism by which adoption propagates without any marketing budget, and it is the single most powerful attribute in the set.

Attributability. Can the adopter tell that the technology caused the benefit? This is distinct from observability and is frequently confused with it. A good harvest is highly observable and may be attributed entirely to the rain.

Compatibility. Does the thing ride on existing practice, or does it require sustained behaviour change and maintenance? A technology that demands a new daily routine is competing against habit every day, indefinitely.

Table 1 — Eight technologies scored against the five attributes

Technology Divisible Trialable Observable Attributable Compatible
Pay-as-you-go solar33333
Mobile money (benchmark, not adaptation)33332
Digital climate advisory33223
Cool roofs and reflective paint22333
Drought-tolerant seed varieties33212
Improved cookstoves22111
Weather index insurance31001
Early-warning systems00112

Scores are analytical judgements on a 0–3 scale, not measurements. Read the rows for a profile and the columns for which attribute is scarce across the set.

Figure 1 — The resulting diffusion score

The sum of the five attribute scores from Table 1. The ordering corresponds closely to the observed adoption record — pay-as-you-go solar and mobile money spread rapidly, digital advisory is spreading, cookstoves and index insurance have not, and early warning does not diffuse through markets at all.

What the agreement with the record does and does not prove

The framework was built knowing which technologies diffused, so the fact that its scores reproduce that ordering is not evidence that it is correct. A fitted ranking always fits. What the framework offers is not prediction but articulation: it states, in a form that can be argued with and applied to a technology nobody has tried yet, what experienced practitioners already know implicitly. The proper test is out-of-sample — score a new product before it is launched, then see — and this report does not claim to have run it.


3. The Observability Penalty

Read Table 1 by column and one result stands out: adaptation goods are systematically weak on observability and attributability, and the reason is structural rather than incidental.

Almost every technology that has diffused rapidly among poor households in the last two decades delivers a benefit that happens. Mobile money delivers a transfer that arrives. Solar delivers light tonight. A phone delivers a call. The user experiences the benefit directly, immediately, and unambiguously — and so does the neighbour.

Adaptation is different in kind. Its benefit is an avoided loss, which is to say a non-event. The flood that did less damage than it would have. The harvest that did not fail. The illness that did not occur. None of these can be observed, because none of them is a thing that takes place; each is a comparison with a counterfactual that no household has access to.

Figure 2 — Two benefit structures

A benefit that happens against a benefit that does not Two chains. Pay-as-you-go solar produces light tonight, which a neighbour can see, which produces a signal to adopt. Flood-proofing or index insurance produces a flood that did less damage, which a neighbour cannot see, which produces no signal to adopt. A BENEFIT THAT HAPPENS A BENEFIT THAT DOES NOT HAPPEN PAY-AS-YOU-GO SOLAR a daily payment, a lamp that works Light, tonight immediate · unambiguous The neighbour sees it from across the yard, for free ADOPTION PROPAGATES no marketing budget required FLOOD-PROOFING, OR INSURANCE a cost now, against a future loss A flood that did less damage a counterfactual · often years away The neighbour sees nothing there is nothing to see NO SIGNAL TO ADOPT every household must be convinced separately

Schematic. The asymmetry is not about the size of the benefit — the avoided loss may be far larger — but about whether anyone can perceive it. Where there is no signal, each household has to be persuaded individually, which is the most expensive way to reach a hundred million people.

Weather index insurance is the extreme case, and instructively so. The household pays a premium and in most years nothing whatever happens, which is the product functioning exactly as intended. Worse, index products carry basis risk: the index may fail to trigger even when the household suffers a real loss. That severs attributability at precisely the moment it matters most, and it does so in the direction that destroys trust.

Figure 3 — Two attribute profiles

Both products are highly divisible — a premium and a daily solar payment are both small and regular — and they diverge almost completely on everything else. Scores as in Table 1.

It is worth noting that this is the same problem that another report in this series identifies at the level of the state, where nobody is photographed beside a flood that did not happen and prevention therefore loses the budget contest to response. The attribution failure operates identically at household scale. A politician cannot claim credit for an averted disaster, and a farmer cannot recommend one to a neighbour.


4. What Spread, and Why

The successes have a common feature, and it is not that they were cheaper or better funded.

Mobile money is the benchmark against which adaptation diffusion should be judged, because it moved from a Kenyan pilot to a mainstream financial system across much of Africa and South Asia in under two decades. By 2025 roughly one adult in ten in developing economies was using a mobile-money account to save, and the World Bank attributes a measurable surge in formal saving to phone-based accounts. India's Unified Payments Interface was processing on the order of 19.5 billion transactions a month by mid-2025.

Pay-as-you-go solar is the most instructive case in the set, because the technology was not the innovation. Photovoltaic panels and lithium batteries were available and falling in price for years without reaching off-grid households. What changed was the instalment collected over mobile money — a pure divisibility intervention, documented by IRENA as a distribution innovation rather than a technological one. Add an instantly observable benefit and the product scores full marks on all five attributes.

Digital climate advisory is the adaptation good that is actually diffusing, and the framework says why. Forecast and agronomic guidance delivered to a phone has been associated with yield gains of 10 to 30 per cent, with the Global Center on Adaptation publishing a roadmap for scaling it across Africa, and randomised evaluations of digital extension in India finding meaningful gains at very low cost per farmer.

A yield gain is the rare adaptation benefit that is a thing which happens. The harvest is bigger. It is visible in the field and in the sack, and although attribution is imperfect — a good year confounds it — the signal is strong enough to propagate.

Figure 4 — Sustained effect, as measured

The contrast that the framework predicts. Digital advisory yield gains per the Global Center on Adaptation; the cookstove result from Hanna, Duflo and Greenstone (2016), where the initial reduction in smoke inhalation had disappeared by the second year. The two measures are of different quantities and are placed together to compare persistence, not magnitude.


5. What Did Not, and Why

Two well-documented failures, both cheap, both distributed, both rigorously evaluated — and both failing on the same two attributes.

5.1 Improved cookstoves

The cookstove is the canonical adoption failure in development, and unusually it has a long-horizon randomised evaluation. Hanna, Duflo and Greenstone tracked households in rural Odisha for up to four years after they received a stove that laboratory testing showed would reduce indoor air pollution and fuel use. Smoke inhalation fell initially. The effect had disappeared by the second year, and there were no changes in health outcomes or in greenhouse-gas emissions. Households used the stoves irregularly and inappropriately, and did not maintain them (see also the working paper).

Diagnose that with Table 1 and it is unsurprising. Compatibility is the lowest score in the set: the stove changes how cooking is done and requires upkeep, so it competes against an established routine every single day. Observability and attributability are close to zero, because the promised benefit is a respiratory illness that does not occur, possibly decades later, to someone who will never know it was averted. The stove was cheap and it was delivered. It scored 7 of 15 and it behaved accordingly.

Figure 5 — The cheaper technology has the larger deficit

Roughly 2.1 billion people lack clean cooking against about 655 million without any electricity, per Tracking SDG 7 (2025). Electrification is the more capital-intensive problem by a wide margin, and it is the one closer to being solved.

5.2 Weather index insurance

Index insurance was designed specifically to solve a real problem — the administrative impossibility of assessing millions of smallholder claims — by paying out against a measured index rather than a verified loss. It is elegant, it is cheap to administer, and take-up has been persistently disappointing without large and sustained subsidies. A review of twenty randomised evaluations finds that farmers are generally unwilling to pay market prices for it.

The reasons named in that literature — basis risk, distrust of insurers, and the sheer difficulty of understanding the product — map directly onto the framework. Basis risk is an attributability failure with the worst possible sign: it means the payout may fail to arrive in the year the household suffers a loss, which is the one year the product is being evaluated. Complexity is a trialability and compatibility failure. And the absence of a payout in a normal year is indistinguishable, from the household's side, from having wasted the money (Carter et al., Annual Review of Resource Economics).

The common structure of the two failures is the point. Neither is a cost problem. Neither is straightforwardly a channel problem — cookstoves were handed to households and index insurance was marketed through existing rural networks. Both fail because the benefit cannot be seen and cannot be attributed, and in the cookstove case because using the thing means changing what you do every evening.


6. Re-engineering the Attributes

The practical value of scoring the product is that a product can be changed. Each attribute has known design interventions attached to it.

Table 2 — Fixing a weak attribute

Weak attribute Design intervention Example
Divisibility Unbundle the cost: pay-as-you-go instalments, on-bill finance, group purchase, sachet pricing Solar home systems, which sold for years only after the payment was broken up
Trialability Subsidise the first cycle, or make the decision reversible: demonstration plots, free first season, return options Seed varieties trialled on a single plot beside the household's usual choice
Observability Bundle the invisible benefit with a visible one, or instrument the benefit so it can be seen A stove sold on faster cooking and less fuel bought, rather than on averted illness
Attributability Convert the avoided loss into a payment that arrives and can be pointed at Index insurance with a guaranteed minimum payout; rebates paid on verified installation
Compatibility Design around the existing practice instead of requiring a new one A stove that fits the pots and the fuel already in use, and needs no maintenance routine

The cookstove row is the one that has actually been learned in the field, and it is worth stating explicitly because it inverts the usual framing. The stoves that achieve sustained use are generally sold on an immediately private and observable benefit — the kitchen is less smoky tonight, the food cooks faster, less fuel has to be bought or gathered this week — rather than on the health return that motivates the funder. The health benefit is the reason the programme exists and is precisely the wrong thing to put in the pitch.

Two honest limits belong here. Redesign is not free: unbundling a payment requires a collection mechanism, instrumenting a benefit requires hardware, and guaranteeing a minimum insurance payout costs money that has to come from somewhere. And redesign is not always available. Where a benefit is irreducibly a non-event with no bundleable companion, no amount of product design will make it diffuse — which is the subject of the next section.

The health benefit is the reason the programme exists, and precisely the wrong thing to put in the pitch.

7. What No Product Design Can Fix

The framework's second use is to sort adaptation goods into those that can be made to diffuse and those that must simply be supplied.

Early warning is the clearest instance of a good that will never diffuse through a market, and the framework says why with some precision. It scores 4 of 15. Its benefit is a pure non-event — a disaster about which you were told, and which therefore did less harm than it would have — so observability and attributability are irreparably low. And it is indivisible in a way none of the other technologies is: a household cannot buy a tenth of a national forecasting system, a satellite feed or a dissemination network. There is no increment to sell.

That is not a design failure to be solved. It is a correct identification of a public good, and the appropriate conclusion is that early warning should be publicly provided and should never have been expected to spread by adoption. The relevant metric is coverage delivered, not take-up achieved: 119 countries — about 60 per cent — report a multi-hazard early-warning system, against the Early Warnings for All target of universal coverage by the end of 2027, with the gaps concentrated in least-developed countries and small island states (UNDRR).

The sorting rule

Score a technology. If it is weak on divisibility and on observability, it is not a product and no distribution programme will make it one — fund it as public provision and measure coverage. If it is weak on observability alone, it is a product with a design problem, and the fix in Table 2 is cheaper than the delivery programme that would otherwise be built around it. Programmes routinely make the opposite assignment: marketing the unmarketable and publicly distributing things that would sell if they were redesigned.


8. An Agenda for Faster Diffusion

Four moves, in the order in which they are cheap.

8.1 Score the technology before funding the programme

8.2 Redesign for observability first

8.3 Buy the rails once and ride them

8.4 Publicly provide what cannot diffuse

8.5 Conclusion

The standard explanations of slow adaptation uptake locate the problem in the money or in the country. The money explanation is simply wrong for the low-cost technologies that make up most of the achievable near-term gain. The country explanation is much better and still incomplete, because it cannot account for the fact that two technologies delivered through the same channel, to the same households, at the same price, spread at wildly different rates — or for a technology that was successfully delivered and then abandoned.

What the difference turns on is largely a set of properties of the goods themselves, and the most important of them is whether anyone can see the benefit. Adaptation is structurally disadvantaged here, because its benefit is an avoided loss: a flood that did less damage, a harvest that did not fail, an illness that did not arrive. These are the most valuable non-events in development and they are invisible, which means every household has to be convinced separately — the most expensive way there is to reach two billion people.

The encouraging part of the diagnosis is that product attributes are cheap to change relative to institutions. Unbundling a payment, bundling a visible co-benefit, guaranteeing a minimum payout, designing around an existing routine: these are weeks of product work, not decades of state-building. And where redesign genuinely cannot help — where the good is indivisible and the benefit is a pure non-event — the framework earns its keep a second time, by identifying what should stop being marketed and start being provided.


References

Every quantitative claim above is attributed inline. The principal sources are collected here.


Metadata

Keywords
technology diffusionadaptation technologyadoptionobservabilityimproved cookstovesindex insurancepay-as-you-go solardigital climate advisorymobile moneyearly warning systemsproduct designlast-mile deliveryclean cookingresilience capabilities
Topics
Climate Adaptation Technology Emerging Markets Public Health
JEL classification
O33, Q16, Q54, O13, D83 — technological change, diffusion processes; agricultural technology and R&D; climate and natural disasters; agriculture and natural resources in development; search, learning and information
Data and method
This report combines the diffusion-of-innovations tradition with the empirical adoption literature and institutional sources on energy access and early warning, including Hanna, Duflo and Greenstone's four-year randomised evaluation of improved cookstoves in the American Economic Journal: Economic Policy; J-PAL's synthesis of twenty randomised evaluations of weather index insurance and the Annual Review of Resource Economics reassessment of index insurance; the Global Center on Adaptation on digital climate advisory; VoxDev on digital agricultural extension in India; IRENA on pay-as-you-go models; World Bank Global Findex reporting on mobile money; Tracking SDG 7 (2025); and UNDRR reporting on early-warning coverage. The cookstove and index-insurance evidence was verified by search in September 2026. Every quantitative claim is attributed inline. Table 1's attribute scores and the Figure 1 totals derived from them are analytical judgements on a 0–3 scale rather than measurements, and the report states explicitly that because the framework was built with adoption outcomes already known, the correspondence between its ranking and the observed record is not evidence for the framework — it is offered as a design checklist rather than a predictive model. Figure 2 is a conceptual schematic. Figure 4 places two measures of different quantities side by side to compare persistence rather than magnitude, and says so. The report is analytical rather than predictive.
Report
H Heuristics Digital Report № 2026-09 · Published 12 September 2026
Licence
CC BY-NC-ND 4.0
Cite as
Hunter Hughes (2026). Scaling Low-Cost Adaptation Technologies Across Climate-Vulnerable Developing Economies: A diffusion framework for rapidly expanding resilience capabilities. H Heuristics Digital Report 2026-09. https://digitalreports.hheuristics.com/reports/scaling-low-cost-adaptation-diffusion/
↑ Back to top