Low and Grow: lending to thin-file customers by starting small

What "Low and Grow" is (and the problem it solves)

Low and Grow —start small and grow— is a lending model built for customers with no history or with references that don't favor them. Instead of denying credit when in doubt, the institution approves a small initial amount and uses real payment behavior as the best available reference.

The goal is twofold: not closing the door on a potentially good customer (inclusion) while capping exposure when there's still no evidence (risk). It's especially useful for serving the base of the pyramid, unbanked or thin-file populations, and volatile contexts.

The four steps

The strategy rests on a simple but disciplined cycle:

• Approve small at first. With no history, you approve —not decline— but for a reduced amount: enough to start the relationship and observe.
• Observe payment behavior. The first loan replaces strict value judgment: how the customer pays generates the references that didn't exist before.
• Grow gradually. If they comply, terms improve and the amount increases on following loans, step by step.
• Calibrate speed, term and price. Raising the limit isn't enough: you must calculate how fast the amount grows, with what term and at what rate. Too big a jump breaks the strategy.

The golden rule: risk is sensitive to the amount

The principle that makes it all work is that risk is sensitive to the amount. In practice, Low and Grow is a precision game: the institution that wins is the one offering the highest amount the customer can handle without exceeding their capacity to pay.

The typical mistake is growing too much or too fast. If the amount exceeds what's appropriate, or the installment triples from one period to the next, "payment shock" appears: the customer reaches a point where they simply stop paying. Growing isn't raising the limit as fast as possible; it's raising it as fast as behavior and capacity to pay allow.

What data to start with when there's no bureau

With no traditional history, the initial assessment relies on alternative and, above all, deterministic data: information verifiable at the source rather than declared by the applicant. Real-time verified employment and income, phone line reputation, utility payments, transactional footprint.

The difference between real data and inferred or declared data is what sustains the strategy: the more deterministic the signal, the smaller the margin of error on the first amount. This doesn't eliminate bias —no data does—, but governed and traceable it lets you decide soundly where the bureau falls short and expand credit in an auditable way. Our guide on alternative data and open finance goes into which sources to use and how to orchestrate them.

How it runs in the decision engine

Low and Grow translates well to a decision engine because it is, at heart, a set of versioned rules: the initial amount per segment, the conditions to scale up, the caps at each step. Every policy version is recorded, auditable and reversible.

Three engine capabilities make the strategy operable and safe:

• Data on demand. The engine requests extra signals only when the profile warrants it, in milliseconds and without friction for the customer.
• Test on a slice of the portfolio. Before moving the growth curve for everyone, the change is tested on 1–2% of operations (champion/challenger): if it goes wrong, the impact is contained.
• Dynamic limits. Lines go up —and down— with payment evidence, without rebuilding the policy by hand each time.

When NOT to grow (and other warning signs)

Growing everyone equally is as risky as not growing at all. Some signals to pause or review:

• Imminent payment shock: the next step would push the installment far above what the customer has been paying.
• Irregular behavior: arrears, sustained minimum payments, or usage that doesn't match declared capacity.
• New customers and fraud: on the first operation it's wise to lean on anti-fraud signals —orchestrated in the same flow— to isolate anomalous cases before approving.