Alternative credit scoring meets peer lending: the partnership model
MicroMoney, a firm building credit scores from alternative data, and Karma, a peer-to-peer lending platform, announced a partnership aimed at extending financial services to people without credit histories. The pairing is a template that recurs across emerging markets: one party claims to identify creditworthy borrowers who have no bank record, the other claims to fund them outside the banking system.
What alternative scoring actually uses
A conventional credit bureau score is a summary of repayment history. Where no bureau record exists — the situation of a large share of adults in developing economies — lenders substitute data the borrower generates anyway: mobile top-up patterns, handset metadata, contacts and call regularity, app usage, e-commerce transactions, utility and rent payments, geolocation stability.
Some of these signals carry genuine predictive power, especially the boring ones: how long a person has kept the same phone number, whether income arrives regularly, whether bills are paid on time. Others are proxies for social class dressed up as behaviour. The distinction matters, because a model trained on the second kind will systematically decline applicants who would have repaid, and no accuracy statistic reported by the vendor will reveal it.
What peer-to-peer funding adds, and what it risks
A lending platform matching individual funders with borrowers does two things at once: it raises capital outside deposit-taking institutions, and it transfers credit risk to people who are rarely equipped to price it. In lightly regulated markets, that has repeatedly ended the same way — a growth phase in which volumes rise, a downturn in which defaults arrive, and retail funders discovering that the platform's own balance sheet stands behind nothing.
The combination with alternative scoring sharpens the issue. A score derived from novel data has no history through a full economic cycle, so nobody — including its authors — knows how it behaves when unemployment rises. Models of this kind are typically calibrated on a benign period and deployed as though that period were permanent.
The questions that separate substance from announcement
Four of them apply to any partnership of this shape. What data does the model use, and did the borrower knowingly consent to its use for credit decisions? Can an applicant who is declined learn why and correct an error? Who holds the credit risk when a loan fails — the platform, the funder or a guarantor? And what does the loan cost the borrower once fees are annualised?
A tie-up that answers all four in public is doing something useful. One that offers only the phrase financial inclusion is describing a customer-acquisition channel.
Why the model persists anyway
Because the underlying problem is real. Hundreds of millions of adults are excluded from formal credit not by their behaviour but by the absence of any record of it, and the cost of assessing them individually exceeds what a small loan earns. Any technology that lowers that assessment cost is worth testing, and the honest version of such a partnership treats itself as a test: small volumes, measured outcomes, published loss rates, and a willingness to abandon signals that turn out to encode prejudice rather than risk.
