AIM for Scale | AI for Agriculture

Evidence map: existing work and possible additions

November 10, 2026

Selected AI use cases across the five problem areas, compared with existing AIM for Scale packages and related initiatives.

Selected AI use cases compared with preexisting AIM for Scale innovation packages activities and related external work, with links to evidence and scalability assessments.
Problem area	Existing AIM coverage	Related work elsewhere	Potential contribution of AI for Agriculture
1. Biophysical hazards and resource base	Weather forecasting, pest and input advice, and AI livestock management. Packages and scope Weather includes AI forecasting. Digital Advisory covers pest and input advice. Livestock includes AI for animal health and management. The Livestock package, pp. 9–10, specifies computer vision and food-safety controls for novel feed. The additions here concern particular prediction and detection tasks. Scope to confirm: the Soil Health page does not yet set out detailed package activities. Our soil-mapping evidence should inform that comparison.	Weather advisory, image-based pest detection and AI-assisted crop breeding. PxD helps deliver AI forecasts and supports pest-surveillance services. Projects and sources The Collaborative includes TomorrowNow’s weather advisory and CGIAR’s AI phenotyping and breeding work. Wadhwani’s CropAce uses computer vision for pest detection and management advice. The specific aflatoxin, grain-screening and poultry-spoilage applications in our review are not explicit in the Collaborative project catalog reviewed. PxD: weather forecasts and pest surveillance AI forecasts at scale: PxD’s 2025 monsoon registry entry reports delivery of partner-produced AI forecasts to about 38 million farmers through government systems. This is reported reach; the Odisha experiment compares SMS plus voice with SMS alone, rather than AI with conventional forecasts. The same dissemination setting appears in our Masiwal and Aitken records. It strengthens existing Weather coverage. Pest surveillance: PxD’s India Project Management Unit supports the National Pest Surveillance System with government, Wadhwani AI and other partners. Programme involvement is documented; effects on crop losses and costs need assessment.	Possible additions Aflatoxin-risk prediction AI function: use satellite and environmental data to predict contamination risk. Food safety. Evidence and scalability: external tests identified low-risk samples better than higher-risk samples. Operational accuracy and surveillance costs still need assessment. Gachoki et al. (2025). Poultry-spoilage detection AI function: classify spectral images and sensor readings to detect meat deterioration. Postharvest losses. Evidence and scalability: a review describes ML methods; performance, equipment costs and benefits in developing-country settings remain to be established. Jia et al. (2026). Specific extension Poultry welfare monitoring through sound AI function: classify chicken vocalizations to identify welfare conditions, extending livestock monitoring. Livestock health and welfare. Evidence and scalability: a low-power technical demonstration in one facility. Health, productivity and income effects, and performance on other farms, are not established. Srinivasagan et al. (2025).
2. Information and knowledge	Agronomic recommendations, pest alerts, market prices and AI-enabled advice through multiple channels. Packages and scope Digital Advisory already includes AI-generated recommendations, local content validation and learning about delivery. Personalized AI advice is an existing focus; particular diagnostic and forecasting functions can extend it.	AI advice through FarmerChat, AgriVaani and PxD’s PaddyAI. PxD also pairs leaf-color charts with digital support. Projects and sources FarmerChat supports advice through text, voice and images. AgriVaani serves farmers and extension workers. Their project reports describe delivery and use; those measures do not by themselves establish yield or income effects. FarmerChat also appears in our literature table. Its inclusion in another initiative is not an independent body of evidence. PaddyAI helps agronomists draft, translate and voice messages within an existing coffee advisory service. Its completed test with 28,796 farmers reports lower listening for AI-scripted messages; quality assurance and the costs of customization remain important. PxD: leaf-color charts and digital support Tool and setting: cotton and rice farmers in India use a physical chart to assess crop nitrogen needs, supported by training and voice reminders. PxD (2026). Evidence and scalability: the article describes randomized rollout and implementation lessons. Surveyed farmers reported 84% adoption in both scaling locations; this does not establish sustained use after the project. Local partners, adapted instructions, field monitoring and tool sharing shaped delivery. Detailed literature assessment pending. Possible AI contribution: while no AI use is reported, AI could reduce the cost of tailoring reminders by crop stage and language, and support follow-up questions. Additional benefits and costs would need testing against the existing digital service. Relevant subcategories: soil fertility, agronomic advice and extension capacity.	Existing coverage Tailored advice and extension messages AI function: answer agronomic questions and help extension services prepare local-language messages. Agronomic advice; extension capacity. Evidence and scalability: FarmerChat provides implementation evidence; PaddyAI tests engagement and comprehension. Farm benefits, sustained use and full service costs need further assessment. Our contribution can compare AI functions within existing advisory efforts. Specific extensions Crop-photo diagnosis AI function: recognize disease symptoms from phone images, including offline use. Diagnosis. Evidence and scalability: Nuru’s technical evaluation tests diagnostic performance; the continued-use study documents adoption barriers. Neither establishes improved farm income. Produce-price forecasting AI function: predict prices across markets from historical observations, extending market-information services. Price and market information. Evidence and scalability: Ma et al. (2018) is identified in the rationale. Detailed literature assessment pending, including forecast performance, costs and effects on farmers’ marketing decisions.
3. Finance	Finance supports investment and adoption; a dedicated AI credit-assessment application is not specified in the reviewed activities. Packages and scope Livestock links advice and inputs with finance. Digital Advisory, p. 3, recognizes finance as a complement. The possible addition is the specific assessment function.	Satellite and farm information are already being used to support agricultural credit assessment. Projects and sources Google describes Terrastack combining land records, satellite observations, crop activity and market data to support credit and climate finance. This is related Google agricultural work; the account does not establish improved loan access or farm income.	Possible additions Satellite-informed credit and insurance AI function: estimate plot productivity to inform lending and connect credit with crop insurance. Liquidity and credit; insurance and risk sharing. Evidence and scalability: the rationale cites an AAEA paper by Kramer et al. (2025). The literature table contains a related 3ie report whose full methods and findings remain under review. The AI contribution and effects on borrowing, investment and risk require verification.
4. Markets and transaction costs	Buyer–seller connections, input quality, livestock traceability and quality assurance. Packages and scope Digital Advisory includes evidence on matching buyers and sellers and accessing quality inputs. Livestock, p. 8, addresses traceability, certification and quality assurance.	The reviewed Collaborative portfolio does not explicitly describe these grain-screening or market-matching applications. Comparison scope This comparison refers to the published Collaborative project catalog. It does not establish that these applications are absent from every member’s wider work.	Possible additions Grain-contamination screening AI function: classify spectral measurements to identify and sort contaminated maize kernels. Quality verification; food safety. Evidence and scalability: tested sorting reduced contamination in accepted grain streams. Equipment costs, rejected grain and sustained operation matter for economic viability. Stasiewicz et al. (2017). Specific extensions Matching and transaction-price recommendations AI function: match buyers and sellers using price, quantity and location, and suggest transaction prices. Access and matching; competition and bargaining. Evidence and scalability: Newman et al. (2018) is identified in the rationale; detailed assessment pending. Related marketplace evidence examines trade and integration, but does not isolate an additional AI effect. Participation by smaller farmers remains a concern.
5. Institutions and public delivery	Extension capacity, farmer registries, shared data, content validation and institutional delivery. Packages and scope Digital Advisory explicitly proposes investment in these systems. General improvements to public data and extension overlap with existing work; the applications below address particular administrative and early-warning functions.	Food-crisis forecasting, agricultural statistics and field mapping. PxD tests ML for advisory delivery and supports government AI services. Projects and sources The Collaborative includes WFP’s HungerMap Live and FAO’s agricultural mapping work. Google’s Agricultural Understanding models identify field boundaries and crop activity. Field boundaries describe visible plots; they do not establish legal ownership. Forecasting a crisis also does not establish that institutions can prevent its effects. PxD: delivery and government AI services Personalized call timing: PxD’s Ama Krushi experiment uses ML to predict when farmers will answer advisory calls. This AI delivery study is distinct from the broader Ama Krushi impact study in our literature table. Government integration: the India Project Management Unit supports Kisan e-Mitra, which answers scheme and payment queries, and Krishi Saathi, which assists agricultural call-centre advisers. PxD supports assessment, integration and scaling with government and technical partners; it is not the sole developer.	Specific extension ML-personalized timing of advisory calls AI function: predict a farmer’s likely response to calls at different times and schedule delivery accordingly. Extension capacity; agronomic advice. Evidence and scalability: the PxD registry describes approximately 1.3 million farmers, with about 900,000 in the analysis. Call pickup rose by 1.7 percentage points relative to random timing during a three-week evaluation. Sustained engagement, farm benefits and full costs remain to be assessed. Athey et al. (2024) provides the working paper. Detailed literature-table assessment pending. Possible additions Parcel mapping for land administration AI function: delineate visible plot boundaries from aerial imagery to support land administration. Land rights. Evidence and scalability: Crommelinck et al. (2019) is identified in the rationale. Detailed literature assessment pending. Local validation, administrative costs and recognition of rights require separate assessment. Food-crisis early warning AI function: combine prices, weather and conflict information to forecast food-crisis risks. Conflict and service continuity. Evidence and scalability: Andrée et al. (2020) is identified in the rationale. Detailed literature assessment pending, including predictive performance, operating costs and whether earlier warnings lead to effective action.

What existing work contributes, and where our review can expand

Delivery design: AIM’s published packages specify institutions, delivery channels and implementation targets. The AI review can connect particular functions to evidence on farmer outcomes, costs and sustained use within those arrangements.

Further literature coverage: AI-assisted crop breeding is more explicit in the Collaborative’s portfolio than in our current collection. CGIAR’s phenotyping work and The Sainsbury Laboratory’s disease-resistance research identify an area to review before adding it to this map.