AI Advisory, Risk Communication, and Investment under Uncertainty: Evidence from Nigerian Smallholders
How to cite
Oyinbo, Oyakhilomen; Chamberlin, Jordan; Gebrekidan, Bisrat. 2026. "AI Advisory, Risk Communication, and Investment under Uncertainty: Evidence from Nigerian Smallholders." AEA RCT Registry. AEARCTR-0018919. https://doi.org/10.1257/rct.18919-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Smallholder farmers in rainfed systems in developing regions face high levels of risk and uncertainty in their production and investment outcomes. Yet, such stochasticity of farm-level investment returns is rarely well-defined, and even more rarely communicated to farmers in existing extension systems in SSA, especially amid increasing production and market risks. Technical advances in AI-enabled analytics, large amounts of remotely sensed data, and other large, geographically distributed datasets on market and production characteristics observed over time and space are now being explored to tailor return distributions to a farmer's specific plot, crop, and input choices on demand. However, little is known about the effectiveness of AI-enabled risk parametrization and communication in farm-level investment and modalities for priming farmers to inquire about returns distributions. Using a four-arm cluster randomized controlled trial across 240 villages and 1680 maize-producing households in northern Nigeria, we evaluate whether AI-enabled advisory systems that communicate the stochasticity of fertilizer investment returns can improve smallholder farmer decision-making compared with advisories that provide agronomic and price information without stochastic guidance.
Intervention
Our interventions include an AI-enabled advsisory chatbot that can be self-administered by farmers via Telegram on Android phones, a short passive video explainer of risk concepts and chatbot capabilities, and an incentivized fertilizer investment game. All study farmers can access the AI-enabled advisory tool in the local language and have the option of English language
Experimental design
The experimental design is a cluster randomized controlled trial with four treatment arms (T0, T1, T2, and T3) across 240 maize-producing villages in northern Nigeria. Farmers randomly assigned to the four treatment arms are exposed to the following:
T0 (Control): AI chatbot that provides recommended fertilizer package (rate, type, timing, and basic “how-to”), rainfall forecasts, agronomic and investment returns guidance; advice is deterministic and point-estimate based; no stochastic production or price content.
T1 (Risk-Aware Advisory): T0 + explicit stochasticity in production and market outcomes, scenario-based returns, and risk-adjusted profitability guidance.
T2 (Risk-Aware + Passive Explainer): T1 + short passive video explaining risk concepts and how the chatbot can answer probabilistic questions. Holds total dosage and exposure to risk content roughly constant relative to T3.
T3 (Risk-Aware + Risk-Inquiry Priming): T1 + interactive incentivized investment game in which farmers choose fertilizer investments under uncertainty, observe realized payoffs, and are nudged toward asking the chatbot about return likelihoods and distributions
Design details
Cited by
No citation data available for this registration yet.