Google Flood Hub triggers pay $105 to 4,600 people in Kogi

Google Flood Hub AI river-flood forecasts accessed via API, with JBA Global Resilience hydrology analysis and a 20%-of-community-area trigger rule

Not assessed

Not assessed

Google (Flood Hub); trigger design by GiveDirectly with JBA Global Resilience

Not assessed

Not assessed

Not assessed

Not assessed

GiveDirectly and JBA used Flood Hub forecasts to identify communities where at least 20% of the area was expected to flood. GiveDirectly verified residents’ identities and bank details before sending payments.

Not assessed

Decisive for timing and geography: GiveDirectly told The New Humanitarian that satellite/AI flood modeling determined which communities were paid and when. Individual eligibility was decided by non-AI verification (ward residence, Smile ID checks, call-center calls). The investigation identifies that as the main source of exclusions. GiveDirectly's own page attributes the late payments to identity checks in two wards.

Decisive for timing and geography: GiveDirectly told The New Humanitarian that satellite/AI flood modeling determined which communities were paid and when. Individual eligibility was decided by non-AI verification (ward residence, Smile ID checks, call-center calls). The investigation identifies that as the main source of exclusions. GiveDirectly's own page attributes the late payments to identity checks in two wards.

Not assessed

GiveDirectly paid US$105 to more than 4,600 people in flood-prone communities, with 3,250 paid before the flood peak.

Reduced Risk

Not assessed

People. Organizations. Society.

Residents of 53 communities in Ibaji, Kogi State, Nigeria.

GiveDirectly reports more than 4,600 recipients. It acknowledged that it had used “household” and “individual” interchangeably.

Some residents were excluded because they lacked phone access or their details did not match government or bank records.

Not assessed

Yes

The New Humanitarian found that verification and phone-access barriers excluded some of the most vulnerable residents.

Not assessed

Not assessed

GiveDirectly surveyed recipients about weekly income and food shortages.

Not assessed

Weekly income and the share of recipients reporting food shortages.

recipients' own situation before the transfers (GiveDirectly follow-up survey; no control group)

GiveDirectly’s survey reported weekly income rising from ₦4,500 to ₦10,000 and food shortages falling from 97% to 7%.

Not assessed

The income and food figures are self-reported and not independently verified. About 1,400 recipients were paid on 24 October 2024, after floods peaked on 21 October. A later US$210 transfer has no stated recipient count.

Not assessed

Not assessed

anticipatory cash transfer per recipient before the flood peak (a second $210 transfer followed)

$105 (source equivalent: NGN177,866 per recipient). Reported equivalent; later naira payments may differ.

USD, identified from the source context. NGN, stated in the source. Nominal source amount; price year not separately stated.

August 29 to October 24, 2024, for the first transfer. Per recipient, first transfer.

Paid.

Recipient: Registered recipients in six wards of Kogi, Nigeria. Payer: GiveDirectly and its funders.

Recipient cash transfer. A transfer is not a net increase in societal resources or an estimate of avoided flood loss.

Not assessed

The amount is not a comparison with aid delivered without AI.

Not assessed

Not assessed

Developer: Google (Flood Hub); trigger design by GiveDirectly with JBA Global Resilience. Operator: GiveDirectly.

Benefit: Ibaji, Kogi State, Nigeria. Developer: United States. Operator: United States.

Not assessed

Not assessed

Not assessed

Not assessed

Forecast skill on the Niger and Benue floodplain. Phone access, charging and matching ID records among the poorest. Funds released and verification completed before the hazard.

2 follow-up entries. Outcome unresolved: 2.

27 June 2025. GiveDirectly's implementer report (published 27 June 2025, revised July 2026 to add the tradeoffs of remote phone-based targeting) says it is extending AI-triggered anticipatory cash with Google Research to new countries and other hazards.

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

AIBD-0095-source-0001–0003

Numbered references below.

Not assessed

Operator, Independent

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

People. Organizations. Society.

People: 2. Organizations: 1. Society: 2.

Recipients received cash, and GiveDirectly’s survey reports higher income and fewer food shortages. GiveDirectly also gained experience in sending payments after a forecast threshold was reached. It acknowledged that its recipient units were inconsistent. The investigation documents that some of the poorest residents were excluded.

Not assessed

Not assessed

Not assessed

Not assessed

Independently reported

Not assessed

Google-built hazard alert system underestimated the earthquake and failed to warn residents; Flood Hub triggers from the same developer decided who received anticipatory cash.

Google Android Earthquake Alerts System Reportedly Underestimated Kahramanmaraş Earthquake and Failed to Send Urgent Warnings to Millions

Humanitarian anticipatory action and cash transfers. NGO cash assistance.

Trigger and target anticipatory cash transfers to flood-prone communities before river flooding.

Date basis unassessed: 2024 (precision unassessed). Updated: 25 October 2026.

Not assessed

SHA-256: 98c5275b8297…

AI Incident Database: 1546.

Not assessed

Not assessed

9 revisions. Latest: 25 October 2026.