For two nonprofits working at very different ends of the climate change spectrum, AI has changed what they can credibly offer the people they serve. SEEDS India, which works on building the resilience of people exposed to disasters and climate change impacts, and Farmers for Forests (F4F), which works in agroforestry, arrived at their respective AI solutions from different starting points and in pursuit of different aims. But both have seen a shift in the basic value proposition of their work: from broad and generalised interventions to something more precise and accountable.
Where AI enters the picture
SEEDS India works across the disaster-management cycle, from first response to rehabilitation and resilience-building. The organisation identified that traditional disaster relief planning was often reactive and overlooked communities at the highest risk due to a lack of granular, real-time data. It therefore developed an AI model that brings together satellite imagery, meteorological data, land-use maps, building footprints, roof types, vegetation cover, population density, and household surveys to calculate hyperlocal, multi-hazard vulnerability scores. The model flags hotspots for heat, floods, cyclones, and earthquakes, ranks households by risk, and generates impact forecasts.
Earlier, the organisation would treat an entire settlement as a single unit of risk. A neighbourhood exposed to extreme heat, for example, might receive a common advisory. In this case, the differentiation criteria would be broad, such as households with young children or elderly residents. Now, by deploying the AI model to identify factors such as roofing material and heat absorption, the organisation can identify which homes within a settlement are likely to face the highest heat gain and prioritise those households for support. In one case, the model helped identify high-risk homes within pockets of around 3,000 households, allowing its teams to prioritise those families for targeted advisories and one-on-one support.
As co-founder Manu Gupta puts it, “The difference is between a general newspaper health advisory and a doctor diagnosing an individual patient.”
This marks a shift in the nature of the service that the organisation offers: from broad advice to a diagnosis that can explain why a particular household is vulnerable and what it can do about it.
For F4F, the starting point was scale rather than precision. Agroforestry, the practice of integrating trees with crops, has been limited in India due to high initial costs. Agroforestry significantly increases farmer income while also removing large amounts of carbon from the air. The carbon market has the potential to finance large-scale agroforestry implementation, but traditional monitoring methods are manual, time-consuming, and based on small representative samples. F4F seeks to address this gap through its AI-enabled measurement, reporting, and verification (MRV) platform, which uses drone imagery to detect and classify trees, estimate height, canopy, and carbon stock with high accuracy, and quantify uncertainty for reliable carbon reporting. Where F4F once relied on manually collected field data to estimate tree survival, carbon sequestration, and biodiversity outcomes, it can now produce precise and auditable figures for each of these parameters.
What AI makes possible
AI has helped transform what both organisations can offer, allowing for more targeted, scalable, and accountable interventions. Four shifts in particular stand out.
1. Scale
The clearest impact of AI on both organisations has been on scalability. For SEEDS India, identifying the highest risk households within a large settlement means it can extend a tailored response to the households that need it most without needing proportionally more staff on the ground.
This also makes its ambition of reaching all high-risk districts in India by 2030 more conceivable. Bringing together varied data from 225 unique districts would be extremely challenging to accomplish manually, but AI can help manage this information at scale. The organisation, however, recognises that reaching this goal will require not just technology, but funders, governments, civil society organisations, and data sources to work together.
For F4F, the gain is arguably more tangible. A ground survey of a single plot, which involves counting trees by hand and measuring bark circumference and height with a tape, takes about four to five hours. The same plot can be surveyed by drone within 2–20 minutes, depending on the height at which the drone flies and the resolution required. Using AI makes it possible to scale the programme to millions of trees across millions of acres, much of it monitored over a six-month cycle over projects that run for 30–40 years. That difference in time is what makes monitoring at this scale possible without a significantly larger field team.
2. Granularity
AI has also changed what each organisation’s data can say. F4F’s earlier ground surveys produced a single figure for a sampled set of trees, with no way to see which specific trees had been counted or how the number was calculated. Drone imagery gives it tree-level data, allowing the organisation to answer questions from a farmer about why they were paid a particular amount or from a carbon buyer who asks how a sequestration figure was reached.
For SEEDS India, greater granularity has changed not only who receives an intervention but how the organisation diagnoses a problem. After landslides devasted Himachal Pradesh’s Mandi town in 2023, the organisation initially assumed from its observational data that the community’s core concern was a lack of homes. Once AI helped reconcile that data with community voices gathered directly from residents, the organisation found that the more significant gap was the absence of government support.
A similar shift happened in Cuddalore. By combining cyclone models with community-generated risk registers, SEEDS India found that heavy rainfall, rather than wind, was the more significant risk in some areas. This changed the recommended response from staying indoors and securing doors and windows to evacuation where necessary.
3. Trust and auditability
The most significant shift for F4F has been in securing trust. Previously, there was no easy way to verify how field teams had arrived at a ground-survey dataset. Now, during due diligence with carbon buyers and funders, F4F can produce tree-level data drawn from geotagged drone imagery that they can independently verify. Farmers, carbon buyers, and philanthropic funders can each be shown exactly how a payment, sequestration claim, or survival-rate figure was calculated. This auditability has helped F4F access larger and international sources of carbon funding. The organisation is careful, however, not to attribute this change to technology alone: Stronger on-ground implementation combined with technological capability was what enabled it to build credibility.
For SEEDS India, trust is centred on the relevance of a message rather than the accuracy of a number. A household that receives an advisory built around the specific material of its roof is more likely to act on it than one that receives a citywide bulletin. This precision is also contributed to by community voices, which remains central to validating and contextualising what the data shows. The organisation’s view is that human judgement is valuable in grounding and contextualising scientific data, even though applying it too early can introduce assumptions and bias into the analysis.
4. Access
AI has also changed whom both organisations can reach. SEEDS India can now identify specific pockets within urban settlements, often lower income neighbourhoods, that face disproportionate risk and would otherwise be folded into a broader average.
For F4F, the barrier before AI was economic. Carbon brokers prefer farmers with large plots because fewer, bigger plots mean fewer sites to send monitoring teams to. This effectively locks out smallholder farmers from carbon finance altogether. By reducing the cost and time required for monitoring, F4F’s system creates a path for smallholder farmers to be included in projects that previously ran on economies of scale built around excluding them.
The challenges that persist
Neither organisation arrived at its current solution immediately. F4F’s journey began in 2020, and it took roughly two and a half years of experimentation before the organisation had something it could deploy at scale. It first attempted to tag trees with QR codes, which did not hold up on land that farmers were still actively cultivating. Satellite imagery did not work either as it cannot capture anything smaller than 30 square centimetres per pixel. This causes the images to miss saplings entirely during their first five years—the exact window when mortality risk is highest.
Drone imagery has its own constraints too, since monsoon conditions make it hard to distinguish a sapling from a shrub. Intercropping, where farmers plant a tree species like mango alongside a staple crop like wheat, can make young trees difficult to detect. F4F also had to build much of its detection model from scratch, since it found no existing models trained on tree species common to Indian agroforestry, such as mango, citrus, teak, neem, or bamboo.
SEEDS India faced a similar process of trial and error. Its early model produced errors that required extensive ground truthing, and the organisation continues to watch for hallucination in its outputs. Training the model also involved making difficult judgement calls about which factors should contribute to a home’s risk score.
These experiences point to a challenge beyond building a model: building the systems around it. Once a model performs well, months of work may remain in figuring out how data should be stored and managed, or how server time and cloud costs can be optimised. Both organisations note that this infrastructure can take longer to get right than the model itself, yet it is often the kind of unglamorous work funders are reluctant to support.
There is also the question of adoption. SEEDS India climate-risk model has been extensively tested and used internally, but encouraging other agencies and organisations to adopt it has proved difficult because of inertia within the wider system.
What nonprofits should take away
A theme both organisations return to is the idea that a solution does not need to be perfect to be useful. SEEDS India describes this explicitly as its ‘good enough’ principle—the understanding that in crisis settings, a workable, resource appropriate response often matters more than a fully optimised one.
F4F presents another important lesson: AI should be adopted only where it is genuinely core to an organisation’s operations. Technology introduced simply to appear current is more likely to remain stuck at the pilot stage.
There is also a deeper organisational question that emerges as AI becomes embedded in everyday work: How do nonprofits ensure that the technology does not begin to shape the organisation’s priorities rather than simply helping it pursue them?
SEEDS India has seen situations where proposals and other internal outputs are passed through multiple AI tools, making it difficult to distinguish the organisation’s own understanding of a community’s needs from information generated or drawn from elsewhere. It therefore emphasises the need for responsible use and a clear internal AI policy.
For nonprofits, the transition to AI is a question of organisational judgement: Where should AI be used, where should human judgement remain central, and how can organisations ensure that the voices and needs of the communities they serve are not diluted?
The experiences of SEEDS India and F4F suggest that AI’s greater potential lies in changing what nonprofits can credibly promise: more targeted support, more granular evidence, greater accountability, and access to people or opportunities that conventional approaches struggled to reach. But those gains depend on better data, stronger systems, field knowledge, human judgement, and a willingness to change how work gets done.
—
Know more
- Learn more about what it takes to effectively fund AI for social impact.
- Read the following case studies on SEEDS India and F4F on the Digital Toolbook for Social Impact platform to better understand how they leverage AI in their work.
