WhatsApp-native AI for 22K smallholder farmers in Zambia. 5-agent system, per-zone x per-risk-tier nudges, tool-augmented Claude chat in English / Bemba / Nyanja. SCU Spring 2026 Showdown project.
# GNA Agri-Intelligence — WhatsApp-Native AI for 22,000 Smallholder Farmers
> **A multi-agent AI system for Good Nature Agro (Zambia) that turns farmer-level data into per-zone, per-risk-tier WhatsApp nudges — and lets farmers ask any question, in English / Bemba / Nyanja, with answers grounded in agronomy, peer benchmarks, rainfall, and market prices.**
**22,597 farmers · 5 agroecological zones · 3 languages · 6 tool-augmented agents · Built on the SCU Spring 2026 Analytics Showdown findings**
---
## The Problem
Good Nature Agro (GNA) is a Zambian agribusiness working with **22,000+ smallholder farmers** under a contract-farming model: GNA provides inputs (seed, inoculant, fertilizer) on credit, the farmer plants and grows, GNA buys back the harvest. Two structural problems make this fragile:
1. **Information asymmetry.** GNA knows what works (per-zone planting windows, which inputs lift yield, which farmers are at risk). Farmers don't — most farmer support is reactive, via field-agent visits that don't scale.
2. **Repayment risk.** **23% of farmers don't fully buy back** their loan. Some are unlucky (drought); some are mis-managed (planted late, skipped inoculant). GNA's existing field network can't visit everyone, so the *highest-risk farmers often get the least attention*.
A linear sum: better information → better outcomes → better repayment → better unit economics → more farmers served.
> **Why now?** WhatsApp is the *de facto* messaging layer in rural Zambia. Multi-language LLMs handle Bemba and Nyanja well enough to be useful. Together, the cost of reaching a farmer with personalized, grounded advice has gone from "send a field agent" to "send a message" — and that's a different business model.
---
## Users & Jobs-to-be-Done
| User | Job-to-be-Done | Today's Workaround | Pain |
|------|----------------|--------------------|------|
| **High-Risk Farmer (4,948 of them)** | When something looks wrong with my crop, I want a fast, trustworthy answer in …