An AI-driven credit platform serving individuals and businesses in Africa came to us with a clear goal: unlock access to finance through alternative data and automated decisioning. The company had already identified that traditional lending infrastructure could not support the speed, scale, or risk profile their market demanded. They needed a partner who could design and deliver a B2B SaaS decision engine that would serve both lenders and merchants, with real-time credit decisions and collections that actually worked.
From the start, the scope was defined around outcomes, not just features. The client wanted to scale disbursements without blowing up default rates. They wanted to enable merchants to offer buy now pay later without those merchants carrying credit risk. And they wanted a system that could run in production day in and day out, with clear audit trails and the ability to iterate on models and rules as the portfolio grew. We aligned on those outcomes and then designed the architecture, data pipelines, and integration points to support them.
The challenge
Lenders in the space faced persistently high default rates. Credit files were thin; many potential borrowers had no traditional bureau history. Manual underwriting could not scale, and it could not deliver the instant decisions that consumers and merchants had come to expect from digital finance. Collections were unreliable. When repayments depended on manual follow-up or legacy processes, delinquency and loss rates climbed. The result was a market where many lenders stayed small by choice: scaling meant taking on risk they could not model or manage.
Merchants wanted to offer buy now pay later (BNPL) to increase basket size and conversion. But most did not have the capital to hold consumer credit risk, and they did not have the risk tools to underwrite or collect. Without a partner that could provide instant decisions and guaranteed settlement, BNPL remained out of reach. The gap was not just technology; it was a full stack: data, decisioning, repayment rails, and servicing, all aligned so that lenders could fund loans, merchants could offer BNPL without carrying the risk, and borrowers could get transparent, fast access to credit.
Our approach
We designed and built a B2B SaaS decision engine (Vida) that uses alternative data: banking, payroll, credit bureau, and behavioral signals, plus anti-fraud checks. The engine does not rely on traditional bureau history alone. It connects to bank accounts linked to the borrower’s national ID (e.g. BVN) and to government payroll systems, enabling automatic deductions at source. That closed loop (data plus repayment rails) was critical. Decisions could be fast and risk-based because the system knew how repayment would be collected.
We deployed an AI-powered risk model tuned on historical loan performance. The model delivers real-time credit decisions and is calibrated to keep portfolio PAR 90 below 10%. That level of default performance is rare in markets where bureau coverage is thin; it was achieved by combining multiple data sources, strict anti-fraud rules, and repayment structures that aligned with how people actually get paid. We also added a forward-flow and marketplace model so that banks and investors could fund loan books while the platform handled origination and servicing. That allowed the client to scale volume without scaling balance sheet alone.
Integration with bank and payroll systems was non-negotiable. We built the APIs and workflows so that once a borrower consented, the system could verify income and pull repayments automatically. For salary-backed loans, integration with government payroll providers meant that repayments could be deducted at source. That reduced arrears and gave lenders and investors confidence that the structure was enforceable. For merchants, we enabled a flow where the platform took the credit risk and guaranteed settlement to the merchant, so the merchant could offer BNPL without holding receivables or building a collections team.
Results and metrics
The platform has disbursed over $12 million to more than 314,000 borrowers in partnership with financial institutions. Repeat borrowing rate reached 86%: borrowers who repaid came back. That signal matters as much as default rates. It indicates that the product, pricing, and experience are aligned with what the market wants. Net monthly revenue reached $100k with positive EBITDA and gross margins near 93%. Fifty-four merchants were onboarded for BNPL, with early pilots showing higher average order values and guaranteed settlement. The decision engine gave merchants and lenders instant yes/no decisions and collections they could rely on.
Key metrics we track: PAR 90 default rate below 10%; 86% repeat borrowing rate; $12M+ disbursed; 314k+ borrowers; 54 merchants live on BNPL. These numbers are not just outputs; they reflect a system that is production-grade, compliant with local regulation, and built to scale. The client can now add new lenders, new merchants, and new products (e.g. different loan types or tenors) on top of the same engine and repayment infrastructure.
What this means for you
If you are a lender, a fintech, or a platform that wants to offer credit or BNPL in a market where bureau data is thin, this case study shows that alternative data, automated decisioning, and closed-loop repayment rails can work together to keep default rates low and repeat usage high. The build required deep domain knowledge in lending, risk, and local payment systems. We brought that to the table so the client could focus on distribution and product, not reinventing the engine.
The decision engine gave us instant yes/no with collections we could rely on. We could finally offer BNPL without holding the risk ourselves.Product lead, merchant partner
Why the architecture mattered
The decision engine is not a black box. It combines rules (e.g. minimum income, maximum tenor, anti-fraud checks) with a trained model that scores applicants on historical performance. The model uses the same alternative data that feeds the rules: banking behavior, payroll stability, bureau (where available), and behavioral signals. We built the pipeline so that new data sources could be added and the model retrained on a schedule without taking the system offline. That flexibility has allowed the client to expand into new segments and geographies without a full re-architecture.
Servicing and collections are part of the same platform. When a loan is disbursed, the repayment mandate is already in place. The system triggers deductions on the due date; if a deduction fails, the workflow moves to the next layer (e.g. retry, then outreach, then escalation). Lenders and investors see clear reporting: disbursements, repayments, arrears, and write-offs. That transparency builds trust and makes it easier to add new funding partners. Merchants see settlement on the agreed schedule, so they can manage their own cash flow. Everyone in the chain has visibility and predictability.
Lessons for similar projects
Three lessons stand out. First, alternative data only works if you have repayment rails that match it. Salary-backed loans need payroll integration; bank-linked loans need reliable debit mandates. Building the decision engine without building the collection infrastructure would have left default rates high. Second, calibration takes time. We used historical performance to tune the model and the score cutoffs. The client had enough volume and history to do that; for a greenfield launch, we would plan a phased rollout with conservative limits until data accumulates. Third, regulatory and partner due diligence cannot be an afterthought. The client worked with licensed lenders and followed local digital lending guidelines. We built the system to support audit and compliance from day one.
If you are exploring a similar opportunity (lending, BNPL, or decision engines for underserved markets), we can walk through your context, your data, and your target metrics in a discovery call. We will tell you whether our approach fits and what a path to production could look like.