Novissi emergency cash transfers targeted by a phone-and-satellite poverty model (Togo)

entry · p2026-09-06-293 · validated

the machine-learning approach reduces errors of exclusion by 4–21%

tech: learned-aid-targeting
passes the definition as: machine-learning models fit to survey-labelled mobile phone metadata and satellite imagery, which ranked subscribers by predicted poverty and thereby decided who was paid
from: 2020-09
region: other:TG
domain: access
side: benefit
people: 60,000 people (2020-09)
effect: 21 percent (2020-09) — fewer errors of exclusion (a 4 to 21 percent reduction) than the geographic targeting the Government of Togo would otherwise have used, so fewer of the poorest were left out of the subsistence payments
reversible: yes
who holds the switch: owner:Government of Togo
attribution: named
source kind: primary
the statistic: Aiken, Bellue, Karlan, Udry and Blumenstock, Nature 603, 864-870 (2022). The policy scenario the model actually decided was the September 2020 distribution of cash to 60,000 beneficiaries in Togo's 100 poorest cantons; the rural Novissi programme it belonged to distributed about US$4 million among 154,238 registrants, and GiveDirectly, the implementing partner, states 'In total, the program reached 138K people' (https://www.givedirectly.org/mobileaid-work-in-togo/). Against the government's feasible alternatives the model's exclusion error was 53 percent versus 59-78 percent; against a hypothetical full social registry, which does not exist in Togo, it is 9-35 percent worse. Payments were 12,250 FCFA (about US$22.50) a month for women and 10,500 FCFA for men, bi-weekly for three months, delivered by mobile money. In severity this is subsistence relief to people at the bottom of one of the world's poorest countries during a pandemic, so per person I would call it severe rather than ordinary; the model's own marginal contribution is the several thousand poorest people who were included who would otherwise have been missed. Togo's population in 2022 was 8,068,616 (World Bank).
source: https://www.nature.com/articles/s41586-022-04484-9
proposed by: researcher (opus), brief 2026-09-06, severe-benefits
received: 2026-09-06T17:40:15-04:00
harvest report: impacts/harvest/2026-09-06-severe-benefits.md

Rulings, in order

2026-09-06T17:40:16-04:00 · validatorpass — quote found on the fetched page (exact match; HTTP 200 text/html; charset=utf-8)
2026-09-06T17:40:46-04:00 · sessionseverity — Subsistence cash to 60,000 of the poorest people in Togo during the pandemic, chosen by the model: an essential made affordable, an income-level effect for a household — severe. The model's own margin is the several thousand who would have been excluded under geographic targeting; the count filed is the population it decided for (2020).

The proposal as it arrived (JSON)

{
 "id": "p2026-09-06-293",
 "received": "2026-09-06T17:40:15-04:00",
 "proposer": "researcher (opus), brief 2026-09-06, severe-benefits",
 "kind": "entry",
 "sha": "50691eabb1fb5df8",
 "data": {
  "kind": "entry",
  "tech": "learned-aid-targeting",
  "tech_name": "Novissi emergency cash transfers targeted by a phone-and-satellite poverty model (Togo)",
  "definition": "machine-learning models fit to survey-labelled mobile phone metadata and satellite imagery, which ranked subscribers by predicted poverty and thereby decided who was paid",
  "region": "other:TG",
  "domain": "access",
  "side": "benefit",
  "from": "2020-09",
  "people": {
   "n": 60000,
   "unit": "people",
   "date": "2020-09"
  },
  "effect": {
   "n": 21,
   "unit": "percent",
   "date": "2020-09",
   "what": "fewer errors of exclusion (a 4 to 21 percent reduction) than the geographic targeting the Government of Togo would otherwise have used, so fewer of the poorest were left out of the subsistence payments"
  },
  "reversible": true,
  "switch": "owner:Government of Togo",
  "statistic": "Aiken, Bellue, Karlan, Udry and Blumenstock, Nature 603, 864-870 (2022). The policy scenario the model actually decided was the September 2020 distribution of cash to 60,000 beneficiaries in Togo's 100 poorest cantons; the rural Novissi programme it belonged to distributed about US$4 million among 154,238 registrants, and GiveDirectly, the implementing partner, states 'In total, the program reached 138K people' (https://www.givedirectly.org/mobileaid-work-in-togo/). Against the government's feasible alternatives the model's exclusion error was 53 percent versus 59-78 percent; against a hypothetical full social registry, which does not exist in Togo, it is 9-35 percent worse. Payments were 12,250 FCFA (about US$22.50) a month for women and 10,500 FCFA for men, bi-weekly for three months, delivered by mobile money. In severity this is subsistence relief to people at the bottom of one of the world's poorest countries during a pandemic, so per person I would call it severe rather than ordinary; the model's own marginal contribution is the several thousand poorest people who were included who would otherwise have been missed. Togo's population in 2022 was 8,068,616 (World Bank).",
  "source": "https://www.nature.com/articles/s41586-022-04484-9",
  "quote": "the machine-learning approach reduces errors of exclusion by 4–21%",
  "attribution": "named",
  "source_kind": "primary"
 },
 "from_file": "impacts/harvest/2026-09-06-severe-benefits.md"
}