How GiveDirectly Uses AI to Send Cash Before Disasters Strike

When a cyclone barrels toward the coast of Mozambique, the traditional humanitarian playbook goes something like this: the storm makes landfall, communities are devastated, agencies assess the damage, donors pledge funds, and weeks later aid trickles in. By then, people have already lost their homes, their livestock, and sometimes their lives. GiveDirectly is trying to flip that sequence entirely.

GiveDirectly is a nonprofit built on a simple idea: give cash directly to people living in extreme poverty and let them decide how to spend it. Since 2008, the organization has delivered over $800 million in direct cash transfers to some of the poorest households on earth. But its latest innovation is not about how much money it moves. It is about when.

The Case for Anticipatory Action

The concept is called anticipatory action. Instead of waiting for a crisis to unfold, GiveDirectly uses AI and satellite data to predict when and where disasters will hit and sends cash to vulnerable families before the event occurs. If you know a flood is coming in five days, a family that receives $200 today can move belongings to higher ground, stock up on food, and reinforce their shelter. The same $200 arriving three weeks after the flood buys far less.

AI triggers these payments when forecasts and climate models indicate a high probability of disaster. The system monitors satellite imagery, rainfall patterns, river gauge data, and atmospheric conditions. When thresholds are crossed, cash transfers are automatically initiated to pre-registered households in the projected impact zone, giving families a critical window to prepare.

Seeing Poverty from Space

One of the hardest problems in humanitarian aid is targeting. In regions where census data is outdated or nonexistent, how do you figure out who needs help the most? GiveDirectly's partnership with Google.org has produced a remarkable answer: you look from above.

ML models trained on high-resolution satellite imagery identify poverty indicators invisible at ground level but unmistakable from space. Roof materials correlate strongly with household wealth. Road quality indicates access to markets and services. Agricultural patterns reveal whether a family farms for subsistence or commerce. These models produce granular poverty maps that help GiveDirectly target aid to the households that need it most, cutting enrollment costs by more than half.

Why Cash Works

The evidence base for direct cash transfers is now one of the strongest in development economics. Randomized controlled trials across dozens of countries show that cash reduces poverty, improves health outcomes, increases school enrollment, and stimulates local economies. Cash is also among the most cost-effective forms of aid. GiveDirectly estimates that roughly 83 cents of every dollar donated reaches recipients directly.

When combined with anticipatory action, the impact multiplies. Research from pilot programs in Kenya and Bangladesh suggests a dollar of anticipatory cash can be worth four to seven dollars of post-disaster aid in terms of avoided losses. Families who receive cash before a flood lose fewer assets, recover faster, and are less likely to pull children out of school.

On the Ground

GiveDirectly currently operates in Kenya, Rwanda, Malawi, Mozambique, and several other countries. Its anticipatory programs have been tested against floods in Bangladesh, droughts in Kenya, and hurricanes in the Caribbean. In each case, AI models flag the risk, payments go out, and families use the money in precisely the ways you would hope.

In Mozambique, ahead of Cyclone Freddy in 2023, GiveDirectly sent mobile money transfers to thousands of families in the storm's projected path. Recipients used the cash to buy food, reinforce their homes, and evacuate to safer areas. Post-storm surveys found that recipient households suffered significantly less damage than comparable non-recipient households in the same communities.

A New Paradigm for Aid

Anticipatory cash is not without its difficulties. Forecasts are probabilistic, which means sometimes payments will go out and the disaster will not materialize. GiveDirectly argues this is acceptable because the cost of a false alarm is far lower than the cost of failing to act. Scaling the approach also requires robust mobile money infrastructure and continuous model refinement as climate change makes historical weather patterns less reliable.

But what GiveDirectly is doing represents something larger than a single program. The combination of machine learning, satellite imagery, climate forecasting, and mobile money creates a system that is faster, cheaper, and more dignified than traditional disaster relief. It treats people in poverty not as passive recipients but as decision-makers capable of protecting their own families, if given the resources and the time.

Over $800 million in cash transfers later, GiveDirectly has proven that trusting people works. Now, with AI, they are proving that trusting the data can save lives before the disaster even begins.