Lending, scoring, fraud, and payments
Credit scoring on alternative data, digital lending, fraud and anti-money-laundering screening, mobile money risk engines, and insurance pricing.
Research areas
We hold five areas and go deep in each, because depth in a domain is what allows a test set to resolve the populations that matter.
The five carry equal weight. They are where AI is already contributing to decisions at scale across the continent, and where a decision that goes wrong is difficult for the affected person to contest.
Diagnosis, triage, and screening
Computer-aided detection on X-ray and ultrasound, symptom checkers and triage tools, maternal and neonatal risk scoring, and diagnostic support in primary care.
Registration, targeting, and service delivery
Biometric identity and verification, social-protection targeting, benefit eligibility, claims processing, and tax and customs risk scoring.
Earth observation, yield, and early warning
Satellite-derived flood and drought mapping, early-warning triggers, parametric insurance indices, and crop yield forecasting.
Models used as de facto advisors
Large language models answering health, legal, and agricultural questions for people with no other source of advice, in English, Pidgin, and the major African languages.
Limitations
Depth in an area covers only the systems we have worked on within it, and none of these domains is uniform across the continent. Where a study calls for knowledge we do not hold, we bring in a named external specialist, disclosed in the finding, or we decline.