This Open-Source AI Model Is Helping Farmers in Sub-Saharan Africa Double Crop Yields

CropWise AI, trained on 15 years of African agricultural data, gives smallholder farmers personalized planting, irrigation, and pest management advice via basic SMS.

An open-source AI model called CropWise is producing significant agricultural improvements for smallholder farmers across Sub-Saharan Africa, delivering personalized planting, irrigation, and pest management recommendations through basic SMS text messages accessible on any mobile phone. The project represents one of the most tangible examples of AI technology addressing fundamental challenges of food security and poverty. Smallholder farmers in Sub-Saharan Africa — those cultivating plots of less than two hectares — produce approximately 80% of the region's food supply.

Yet they operate with minimal access to the data-driven agricultural guidance that has driven productivity gains in developed nations. Climate change has introduced increasing unpredictability in rainfall patterns, while evolving pest pressures and soil degradation compound existing challenges. Agricultural extension services are understaffed across most of the region.

CropWise AI was developed by a consortium of universities including the University of Nairobi, Makerere University in Uganda, and the University of Ghana, in partnership with the Bill and Melinda Gates Foundation. The model was trained on 15 years of agricultural data collected from 8 Sub-Saharan African countries, encompassing weather station records, soil surveys, crop field trial results, pest monitoring data, and yield reports from farmer cooperatives. The system's design reflects deep understanding of its user context.

Farmers send a basic SMS message containing three pieces of information: their location (even an approximate village name is sufficient for geolocation matching), the crop they are growing, and a description of their soil. CropWise responds with personalized recommendations including optimal planting dates calibrated to predicted local rainfall patterns, irrigation timing and quantity, pest identification with treatment options using locally available inputs, and harvest timing for optimal yield. A critical technical and design decision was building the entire system around 2G SMS infrastructure.

Approximately 60% of Sub-Saharan Africa's population lacks reliable internet access, and smartphone penetration among smallholder farming communities remains low. CropWise requires no smartphone, no internet connection, and no application installation. After an initial calibration exchange that establishes a farmer's location and typical crops, the system can queue responses during connectivity gaps and deliver them when signal becomes available.

The training data's regional specificity is a key differentiator. CropWise understands the specific crop varieties cultivated in the region — finger millet, sorghum, cassava, and local maize varieties alongside more widely studied crops. It accounts for region-specific pest pressures such as fall armyworm, which has devastated maize production across the continent, and understands microclimate variations that can shift optimal planting dates by several weeks between villages separated by as little as 50 kilometers.

Pilot program results have been substantial. Across 12,000 participating farmers in Kenya, Tanzania, and Ghana, the program measured an average 47% increase in crop yields compared to matched control groups farming without CropWise guidance. Certain crop and region combinations showed even higher gains.

For families operating at or near subsistence level, this improvement represents the difference between food insecurity and the ability to produce surplus for market sale. The project faces significant challenges in scaling. Reaching the most remote farming communities remains difficult — some areas have barely functional cell networks, and physical distance from any form of support infrastructure is measured in days of travel.

Trust-building presents a cultural challenge: farming practices are deeply rooted in generational knowledge, and recommendations from an AI system require community validation. The CropWise team has addressed this by partnering with local agricultural extension workers and farmer cooperatives, introducing the tool through trusted community relationships. Integrating traditional farming knowledge into the model represents an ongoing development priority.

Local farmers often possess granular understanding of their specific land and conditions — knowledge that may not appear in formal agricultural datasets. Crop

Wise has implemented feedback mechanisms allowing farmers to report results and observations via SMS, which are incorporated into model updates. This bidirectional approach aims to complement rather than override indigenous agricultural knowledge.

The consortium plans to expand from 12,000 to 100,000 farmers by the end of 2026, extending from the three pilot countries into Rwanda, Uganda, Malawi, Mozambique, and Senegal. The model's open-source license ensures that other organizations, NGOs, and government agricultural departments can adapt and deploy it without licensing costs — a deliberate choice to maximize impact over commercial return.

---

This Open-Source AI Model Is Helping Farmers in Sub-Saharan Africa Double Crop Yields

Related Reading