Many aid programs still rely on outdated lists that miss people in crisis. Can anonymized phone data + AI do a better job at finding families in need? GiveDirectly, Center for Effective Global Action (CEGA), and the Government of Bangladesh asked a bold question: Could anonymized phone data + AI better identify families in poverty? The result: “MobileAid,” a pilot that used machine learning to predict economic vulnerability from metadata like text message use, recharging behavior, and mobility. 💸 Selected families in Bangladesh received ~$136 via mobile money -- fully remote, faster, and cheaper than traditional surveys. 📊 MobileAid outperformed community selection and rivaled traditional proxy-means methods on accuracy at lower cost and greater speed. Scaling it requires: 🏛 Predictable rules for digital targeting 📡 Reliable data-sharing protocols 🧑🏽‍🤝‍🧑🏿 Community engagement to build trust Used well, this approach can be a fast-track list of who to reach first after a disaster hits. 👉 Read the full blog on how AI targeted aid in Bangladesh: (fb)

GiveDirectly piloted an AI-powered flood response, predicting where floods would hit in Nigeria and sending early cash to those most at risk. The results are remarkable ⬇️ → We used machine learning and local knowledge to ID the flood-risk areas near two rivers in Kogi, Nigeria → We pre-enrolled 38,000 people in those areas using text-message self-enrollment → We used Google's FloodHub forecasts and local insights to trigger payments just before peak flooding (see image) → We delivered early cash to 4,600 households, 48–72 hours after flood triggers were met and before peak flooding The result: 💸 Incomes more than doubled 🍽️ Food insecurity dropped 90% ✅ 93% felt better prepared for future floods 🛰️ Our pre-targeted areas were the most impacted Now we’re ready to scale this model to more countries and more climate risks. Because when early cash arrives in time, people can prepare and recover with dignity. 🧵 Full story: (fb)