JUMO: Lending to People With No Credit History

The most consequential fintech question in Africa is whether phone behaviour predicts creditworthiness. This story covers the platform model, the operator partnerships, behavioural underwriting, unit economics on tiny loans, funding, regulation and consumer protection — part of the South Africa Company Stories hub.

Why is credit scoring so difficult in these markets?

Because most adults have no formal credit history. They have never had a bank loan, a credit card or a store account, so there is nothing for a bureau to report and nothing for a conventional model to score.

Income verification is equally hard. A large share of the workforce is informally employed or self-employed, with no payslip, no employment contract and irregular earnings that vary week to week.

Traditional lending therefore either excludes these customers or prices for the worst case, which produces interest rates so high that only the most desperate borrow — a selection effect that guarantees poor outcomes.

What does behavioural data actually show?

Patterns that correlate with stability and repayment: how regularly airtime is purchased and in what amounts, how frequently the mobile wallet is used, whether balances persist, how long the customer has held the number.

None of these measures income directly. They measure consistency, which turns out to predict repayment behaviour reasonably well — a customer with regular small transactions over two years behaves differently from one with sporadic activity.

The strongest signal is repayment of previous loans on the same platform, which is why lending typically starts with very small amounts and grows as a customer establishes a record — effectively building the credit file that did not exist.

Why partner with mobile operators?

Because they hold the data and the distribution. The operator knows the customer’s usage history, has a billing relationship and reaches them through a channel they already use daily.

Mobile money also solves disbursement and collection. A loan can be paid into a wallet in seconds and repaid from it automatically, without branches, cash handling or payment infrastructure of any kind.

For the operator the arrangement generates revenue from data it already holds, increases wallet usage and reduces churn, since a customer with a credit relationship is far less likely to change networks.

How do the unit economics work?

On very small loans over short periods — often equivalent to a few dollars over a few weeks — which means the absolute revenue per loan is tiny and the cost of any manual process would exceed it entirely.

Everything must therefore be automated: application, scoring, disbursement, reminders, collection and write-off. A single human interaction destroys the economics of a loan that size.

Profitability then depends on volume and on default rates. At these margins a small deterioration in repayment converts a profitable book into a loss-making one, which is why scoring accuracy is not a refinement but the entire business.

What are the consumer protection concerns?

Short-term small loans carry high effective annual rates by construction, because fixed costs are spread over a small principal and a short period. Expressed annually, the numbers look alarming even when the fee is modest in absolute terms.

The genuine risks are repeat borrowing and multiple simultaneous loans across providers, which can trap customers in a cycle where each loan repays the last — a documented problem in several digital credit markets.

Responsible design addresses this through affordability limits, cooling-off periods, cross-provider data sharing and clear disclosure in the language and format the customer actually receives. Regulators in several markets have moved to require it.

How is the lending funded?

Usually by a licensed bank partner that holds the loans on its balance sheet, with the platform providing origination, scoring and servicing in exchange for fees or a share of the economics.

That structure keeps the platform asset-light and places regulated activity where the licence sits, which simplifies compliance in markets where a technology company cannot lend directly.

It also constrains growth to the partner’s risk appetite, which is why platforms have pursued additional funding structures and, in some cases, sought their own licences in specific markets.

What does regulation look like across markets?

Highly varied. Some regulators have embraced digital credit with light-touch frameworks; others have imposed rate caps, licensing requirements and reporting obligations that changed the economics substantially.

Interest rate caps are the most consequential intervention. A cap set with conventional lending in mind can make very small short-term loans structurally unprofitable, which removes the product rather than making it cheaper.

Operating across many jurisdictions therefore requires a platform flexible enough to reconfigure products per market, which is a significant part of what such a company actually builds.

What about savings products?

They matter more than the headlines suggest. A customer with somewhere safe to keep small amounts is more financially resilient than one who can only borrow, and savings behaviour is itself a strong credit signal.

Commercially, savings provide funding and increase engagement, and the customer relationship becomes broader than a series of transactions each of which involves paying interest.

The challenge is that margins on very small savings balances are minimal, so the product is justified by the relationship and the data rather than by direct profitability.

What is the lesson?

That absence of credit history is an information problem, not evidence of poor creditworthiness. Millions of people repay reliably and simply have no mechanism to prove it.

The second lesson is about automation thresholds. At loan sizes this small, any manual process is fatal, which forces a level of end-to-end automation that most conventional lenders have never had to achieve.

The third concerns responsibility. A product that makes borrowing effortless for customers with thin margins of financial safety carries obligations that go beyond compliance, and the operators that ignored this created the regulatory backlash that followed.

How does the platform model differ from being a lender?

A platform provides the technology, scoring and servicing while a licensed partner funds and holds the loans, earning fees rather than interest and carrying limited balance sheet risk.

The advantage is capital efficiency and regulatory simplicity: the platform can operate in many markets without holding a lending licence or the capital that regulators require lenders to maintain.

The disadvantage is dependence and economics. Growth requires partners willing to expand their books, and the platform captures a smaller share of the value than a direct lender would — which is why several such businesses eventually seek licences of their own.

What happens when a market imposes rate caps?

Very small short-term loans usually become unprofitable and disappear, because fixed origination and servicing costs cannot be recovered within a capped annualized rate on a tiny principal.

Regulators frequently intend to reduce the cost of borrowing and instead remove the product, which pushes borrowers toward informal lenders charging considerably more with no consumer protection at all.

The more effective interventions have been affordability rules, mandatory data sharing between lenders and limits on repeat borrowing, which address over-indebtedness without eliminating access.

Why do partnerships with operators eventually strain?

Because the operator holds the customer relationship and the data, and once a lending product proves profitable the operator has an obvious incentive to build or buy the capability rather than share the economics.

Several mobile money operators across Africa have moved in exactly this direction, launching their own credit products or acquiring the technology, which compresses the space for independent platforms.

The defence is technical depth and multi-market scale — a scoring platform improved across many portfolios is genuinely hard for a single operator to replicate — but it is a defence that erodes as operators accumulate their own data.

What does responsible digital lending look like?

Affordability assessment that limits exposure relative to observed cash flow, hard caps on simultaneous loans, cooling-off periods between advances and clear disclosure of the total cost in absolute terms rather than only as a rate.

It also means declining customers the model would approve but who show signs of distress — escalating borrowing frequency, shrinking repayment margins, reliance on new loans to close old ones.

Commercially this reduces short-term volume and protects the portfolio and the licence, which is the same trade-off responsible lenders have always faced and which the speed of digital origination makes easier to ignore.

How does mobile money differ from banking?

Mobile money holds value in a wallet backed by funds in trust accounts at licensed banks, accessed through a phone and converted to and from cash through an agent network rather than branches or cards.

It reaches customers no branch network can serve economically, which is why adoption in several African markets far exceeds bank account penetration and why the wallet, not the account, is the relevant financial channel.

The limitation is product depth. A wallet handles payments and transfers well and requires partnership with licensed institutions to offer credit, savings and insurance — which is precisely the gap that lending platforms exist to fill.

What does the data actually predict?

Repayment probability over short horizons, which is a narrower claim than creditworthiness. Behavioural models are good at ranking customers over weeks and considerably weaker at forecasting behaviour a year out.

That suits the product, since the loans are short, and it constrains any extension into larger, longer facilities where income stability matters more than transaction consistency.

Model performance also degrades when conditions change, because the relationships were learned in one economic environment, which is why continuous recalibration and conservative limits during volatility are not optional.

Frequently Asked Questions