Building AI-based advisory services for smallholder farmers: Technical learnings from the AIEP Initiative
AIEP Initiative, Stewart Collis, Florence Kinyua, Vikram Kumar, Howard Lakougna, Christian Merz, Kirti Pandey, Christian Resch · 2026
Read Building AI-based advisory services for smallholder farmers: Technical learnings from the AIEP Initiative by AIEP Initiative, Stewart Collis, Florence Kinyua, Vikram Kumar, Howard Lakougna, Christian Merz, Kirti Pandey, Christian Resch (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-11537.txt.
Introduction
The AIEP Initiative team report lessons from AI farm advice for smallholders. Their 2026 preprint covers five advisory MVPs in two countries. The services ran in Kenya and in Bihar, India. Smallholder farmers often lack timely, local agronomy advice.
AI-based advisory tools aim to close that gap at scale. The team built minimum viable products and deployed them live. An 800-farmer study found high user satisfaction in the trial. The reported net promoter score sat near 60. Technical learnings from the builds are the paper's core. They cover what worked, what failed, and what scales. Such field lessons guide the next wave of farm AI.
Catalog record
| CREATOR | AIEP Initiative, Stewart Collis, Florence Kinyua, Vikram Kumar, Howard Lakougna, Christian Merz, Kirti Pandey, Christian Resch |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-11537
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Building AI-based advisory services for smallholder farmers: Technical learnings from the AIEP Initiative AIEP Initiative, Stewart Collis, Florence Kinyua, Vikram Kumar, Howard Lakougna, Christian Merz, Kirti Pandey, Christian Resch Abstract We report technical learnings from five AI-based agricultural advisory MVPs deployed in Kenya and Bihar, India, under the AIEP Initiative. A 800-farmer study found high user satisfaction (NPS ∼ \sim 60). All solutions implement a modular two-part architecture: (i) an interface component (IVR / WhatsApp / app) with ASR→MT→TTS for multilingual voice access; and (ii) a reasoning component combining LLMs capabilities with query orchestration, external data (weather/soil/markets), and RAG over curated agricultural corpora. We describe key challenges: (a) latency, especially for voice; reductions were achieved via in-country hosting and audio minimization, but consistent ¡5s remains challenging; (b) language coverage: low-resource ASR/MT integration and nonstandard scripts hinder end-to-end quality; and (c) corpus curation: access, validation, and maintenance are labor-intensive, as well as provide recommendations on how to develop similar systems. We discuss common enablers including (a) data sharing, (b) common corpora, (c) better language AI and (d) evaluation and benchmarking. We also present golden Q&A sets to evaluate LLM capabilities for smallholder agriculture. 1 Introduction: What is this about? Most of the more than 570 million farms in the world are less than 2 hectares and family run ( Lowder et al. 2016 ) , yet these smallholder farmers face diverse challenges, many of which have uncertainty and lack of information in common: What is the right strategy to contain an unknown disease? Which and how to use new seed varieties? How to behave in the face of changing weather patterns that defy traditional knowledge shaped by centuries-old reliable patterns? There are often answers to these questions, but they often lie beyond the reach of many smallholder farmers because traditional agricultural extension systems struggle to provide their service and support at scale, with ratios of extension agents to farmers below 1 to 750 in India ( Nandi and Swamikannu 2019 ) and 1 to 1000 in Kenya ( Republic of Kenya 2012 ) with large variations by region. Various digital solutions exist to close this gap, but many of those solutions struggle to provide the relevant, timely, and personalized information needed by smallholder farmers. At the same time, the effectiveness of video-mediated extension services like Digital Green’s have proven impacts on the uptake of knowledge and technology ( Abate et al. 2023 ; Van Campenhout et al. 2021 ) . SMS have been shown to have a modest impact at low cost ( Fabregas et al. 2025 ) . The Agricultural Information Exchange Platform (AIEP) Initiative, funded by the Gates Foundation and implemented by GIZ’s ’FAIR Forward – Artificial Intelligence for All’ Program with support from CLEAR Global, explored the potential of emerging artificial intelligence (AI) solutions to address this challenge and increased collaboration and exchange between partners working on these solutions to provide better digital services and specifically better agricultural advisory services. As part of the AIEP initiative, five cohorts made up of organizations with different skills ranging from technology and agriculture to product management and user research have developed minimum viable products (MVPs) of AI-based agricultural advisory systems that take advantage of the most recent technological innovation and available agricultural content in Kenya and the State of Bihar, India. Those MVPs aimed to improve existing digital solutions by providing easier access through voice and natural language interfaces, allowing for two-way communication and more relevant, timely, and personalized information, ultimately contributing to supporting smallholder farmers in overcoming the challenges to improve their yields and livelihoods. A first user study conducted by 60 decibels of over 800 smallholder farmers, 38% of which were women, provided first promising evidence on the potential of AI-based advisory. The net promoter score (NPS) 1 1 1 The net promoter score is calculated based on response to the question “How likely would you be to recommend […]?” on a scale from 0 to 10. The score is the share of ’detractors’ (responses of 6 and less) of the share of ’promoters’ (responses of 9 and greater; see ( Reichheld 2003 ) for details) as satisfaction measurement is 60 on average in cohorts which exceeds global (NPS 46), Kenyan (NPS 26) and Bihar (NPS 40) benchmarks of digital farmers services collected by 60 decibels. Importantly, women report a higher NPS than men. Qualitative interviews indicated that farmers often share the information received from AIEP solutions with their peers and encouraged them to also use the solution. Among the areas for improvement, the completeness and relevance of the information provided stands out. It is the main reason why farmers would not recommend solutions to others. In addition, 50% of those who reported that they had not applied the information provided reported that the availability of inputs or financial constraints kept them from taking action. This indicates that the information provided is not always sufficiently tailored to the context and that AI-based advisory in addition to information will need integration with other services that alleviate constraints. Unsurprising yet concerning is as well that farmers with lower educational levels reported a higher likelihood that information was not easy to understand (overall 33% of farmers), pointing to more work required to fulfill our original goals of providing solutions working for all smallholder farmers. This study provides first encouraging data points, but will need to be complemented by further studies in the future. Like most studies, this also showed some potential limitations. For example, educational attainment is higher than the average smallholder farmer, with 30% of the survey participants reporting a university, polytechnical, or other higher degree. Despite limitations, the 60db study suggests that AI-based advisory has strong potential. Farmers seem satisfied with the solutions, stressing their availability and quality of information. The next step which various partners of the AIEP initiative are preparing now is rigorous evaluations of the impact on yields and livelihood to show the material benefit of the solutions. This paper is one of two in which we aggregate insights and learnings that we and our partners have gathered in the last two years of building the MVPs. It aims to elucidate the technical foundations of AI-based agricultural advisory systems, drawing on the experiences and lessons learned from the AIEP project, while its companion paper describes and presents learnings about our innovation methodology (see AIEP Initiative et al. 2025 ). This paper explores the design and implementation of AI technologies in agriculture, discusses the challenges encountered, and provides recommendations for developing similar systems in other sectors. By sharing these insights, we hope to contribute to the broader discourse on leveraging AI for sustainable development. 2 Background on the AIEP initiative The AIEP initiative started in 2023 by selecting four cohorts out of more than 100 applications received in response to a call for proposals. Opportunity International, SafariCom Digifarm and Gooey.ai later joined as the fifth cohort. The key criteria for the selection of the eventual AIEP cohorts were technological innovation; end user and community orientation; gender-inclusivity; impact orientation; evidence; and user feedback; local representation; and long-term sustainability and clear ownership of the MVP. The collective MVP development started in late 2023 and continued until early 2025. During this