Smartphone-based Extension: AI-driven Farm Advisory for Smallholders

Thesis Abstract

Smallholder farmers in developing economies face persistent information gaps on best practices, input optimization, and timely advisories, which constrain productivity and resilience amid climate variability. This study addresses the problem of limited access to context-specific agricultural guidance by evaluating a smartphone-based extension platform that leverages artificial intelligence to deliver tailored farm advisory to smallholders. The aim is to assess the impact of an AI-driven farm advisory system on decision quality, adoption of best practices, and productivity outcomes among smallholder farmers. The specific objectives are (1) to develop and deploy a multimodal AI-driven advisory app integrating crop-specific recommendations, weather- and market-informed alerts, and interactive chat support; (2) to examine user acceptance, perceived usefulness, and engagement patterns among enrolled farmers; (3) to evaluate the effect of advisory service on yield, input-use efficiency, and cost-benefit performance; (4) to identify enabling and constraining factors for scale-up within local extension systems; and (5) to propose a theoretical and practical framework for AI-enabled agricultural extension. A mixed-methods research design is employed. The population comprises smallholder farmers (n=600) across three agro-ecological zones in a mid-size country, with a stratified random sample of 240 participants selected for in-depth analysis. Data collection combines quantitative instruments—structured surveys, platform analytics, and agronomic measurements (yield, input use, input costs)—with qualitative methods, including semi-structured interviews (n=40) and focus group discussions (n=6) to capture experiential insights. Instruments include a validated Technology Acceptance Model (TAM) questionnaire, a Knowledge, Attitude, and Practice (KAP) survey adapted for AI advisory, and a farmer-reported yield log corroborated with field measurements. Data collection spans two cropping seasons to account for temporal variability. Validity and reliability procedures include pilot testing (n=30), Cronbach’s alpha assessment for scale reliability, and triangulation across survey responses, platform logs, and agronomic records. Data analysis proceeds with a hierarchical linear modeling (HLM) approach to assess the impact of advisory exposure on yield and input efficiency while controlling for farm size, soil quality, and climate variables. Regression analyses identify predictors of adoption and productive performance, and propensity score matching mitigates selection bias between users and non-users. Thematic analysis, guided by the Theory of Planned Behavior and the Diffusion of Innovations framework, interprets qualitative data to explain adoption drivers and barriers. A conceptual model illustrating AI-driven advisory flow—from data ingestion (weather, soil, crop phenology) to personalized recommendations and farmer feedback—will be developed and tested. Expected findings include (i) higher adoption of recommended agronomic practices among app users, (ii) statistically significant improvements in yield (mean increase of 12–18%), and reductions in input wastage (10–15%), (iii) enhanced decision confidence and timely actions as evidenced by TAM and HBM constructs, and (iv) identification of critical enabling factors (digital literacy, network reliability, and stakeholder collaboration) and constraints (data costs, language diversity, and trust in AI). The study contributes to knowledge by integrating AI-driven advisory within a rigorous extension theory framework, advancing understanding of how digital decision-support tools influence smallholder productivity, risk management, and livelihood resilience. It also offers a practical implementation blueprint for public-private partnerships, capacity building for extension staff, and policy considerations for scalable digital agriculture interventions. The main conclusion is that AI-powered farm advisory platforms can meaningfully augment traditional extension services when designed with local languages, offline capabilities, transparent explainability, and active farmer involvement. Recommendations emphasize user-centered interface design, affordable data plans, capacity building for extension workers, iterative model updates using farmer feedback, and policy incentives to support data governance and inclusive digital access.