Overview:
Roughly 1.8 billion people face significant flood risk. AI-based flood forecasts are now free and public in more than 150 countries, but prediction is not protection: a forecast helps a household only if the warning reaches it in time to act. In Bihar, India's poorest and most flood-prone state, forecasts cover more than 90 percent of households in our study area, yet only about a third receive any warning in a typical season. This brief summarizes a four-year randomized trial across 319 panchayats and 3.6 million people in North Bihar. It tested whether trained community messengers, relaying official forecasts by loudspeaker, flags, SMS, and WhatsApp, could close that gap.
Findings:
Pairing forecasts with community dissemination delivered more warnings, spurred earlier action, and protected households when severe floods struck, at very low cost.
- More warnings, earlier: Community messengers more than tripled the flood warnings households received, and warnings were about 30 percent more likely to arrive before floodwater reached the home.
- Trust and preparation: Households trusted alerts more and used the lead time to stock food, sandbag, and dig drainage channels, paying for these actions themselves.
- Protection when it mattered most: In severely flooded villages, households offset about four-fifths of flood-related welfare losses. They were less likely to have floodwater enter their homes or to go hungry, and they borrowed roughly $145 (or ₹12,632) less.
- Low cost, high returns: Delivery cost Delivery cost $0.26 (or ₹23) per household per season. Counting reduced borrowing alone, benefits exceeded costs at least 29 to 1.
Next Steps:
JEEViKA, the Government of Bihar’s Rural Livelihoods Mission, has ordered adoption of community delivery through its women’s self-help group network. Inclusion Economics is supporting the transition and testing whether results hold at scale, as well as whether timely warnings help small businesses in flood-prone towns.