Loan decisions in minutes: AI credit scoring done responsibly
Illustrative case study · Client details and outcomes await verification. Explainable machine learning let a lender grow its book five-fold while cutting bad loans — with the regulator satisfied at every step.
What changed
The starting point
Mosi's loan officers manually appraised every application — ten days on average, with approval standards that varied by officer and branch. Growth was capped by hiring, and non-performing loans crept upward.
Leadership wanted AI scoring but feared regulatory pushback: any model would need to explain its decisions, avoid discriminatory bias, and leave a complete audit trail.
What we built
HLM built credit scorecards on repayment history and mobile-money statement data, deliberately choosing model architectures whose every decision can be explained in plain language — and testing them for bias across gender, age and region.
The models were wrapped in a digital origination workflow: applications arrive by app or branch, scores return in minutes with reason codes, and officers review only the borderline band. Every decision is logged for the regulator.
How the rollout ran
Phased, transparent, and designed to earn trust at every gate.
Weeks 1–4: Data & fairness design
Historical lending data prepared; fairness criteria and explainability requirements agreed with compliance.
Weeks 5–10: Model build & validation
Scorecards trained and stress-tested; bias audits documented; regulator briefing pack prepared.
Weeks 11–14: Champion–challenger pilot
Models ran alongside officers on live applications; thresholds tuned against actual repayment.
Weeks 15–18: Full deployment
Digital origination live in all branches; monitoring dashboard tracks drift, fairness and portfolio quality monthly.
Illustrative quotation · not a verified testimonial
“We expected a black box and got the opposite — every decision comes with reasons we can show an applicant or the regulator. It changed how the board thinks about AI.”Chief Risk Officer
More case studies
How AI cut unplanned downtime 22% at a major mine
Two years of ignored sensor data became a three-week early-warning system for equipment failure — and a new way of running maintenance.
Read Savanna Lodges Group*Five lodges, one system: a hospitality group's digital leap
A safari group replaced five paper booking books with one platform — and an AI concierge that sells rooms while managers sleep.
Read BrightPath Education Trust*From paper registers to 96% fee collection across eight schools
A school trust unified administration, fees and parent communication — and found the money that paper systems were losing.
ReadFacing a similar challenge?
Tell us where it hurts — we'll tell you honestly what's possible.