When early loan repayment merits a fraud referral
Two applications sit in the queue, and the new referral rule would send both to your team. Each client cleared a small store loan ahead of schedule, then returned within a few months asking for a much larger cash loan. On the screen, the sequence is neat and easy to recognise: early repayment, then a sharp jump in borrowing. You are about to make that pattern a reason to pause an application, not just a note in the account history. The cases look alike at the point where the rule has to make its call. What should count as the signal?
Compare the bust-out pattern with the same jump after on-time repayment
On this lender’s book, the early-repayment bust-out pattern had lower default than expected. It still deserves a direct comparison, because early closure was followed by another application from the lender within 90 days far more often than an on-time closure.
This is the same consumer lender introduced in our account of applications after a prior refusal, with its name and products changed. The book contains about 310,000 current applications, and every client in the comparison holds a current loan. Default is the lender’s flag for payment difficulty on that loan. Fraud and credit losses both appear under that flag.
Among clients who closed an earlier loan early, then applied within 90 days for at least five times as much, default was 7.1% against 8.4% expected. For clients whose only earlier loan closed on time, the same jump had 10.9% default against 8.6% expected, and the result held in both halves of the book. The cross-sell application rate was 26% after early closure and 9.4% after on-time closure.
The cross-sell increase is real, but the risk sits elsewhere. The default comparison puts the risk elsewhere. Referring the early-closure group on sight uses review capacity on a group that defaulted below expectation, while the same jump after an on-time closure had higher default.
Use the same large-jump comparison on both groups. Keep the expected rate beside the observed rate before deciding which pattern earns a referral.
Define early closure from the schedule and separate products
Define an early closure from the loan’s schedule, then split the results by product and refinancing. A loan counts as closed early when it was paid in full at least 90 days before its planned last due date and had no instalment 30 or more days late.
An individual instalment paid 30 or more days ahead of its due date is a different event. At least one such instalment appeared on 35% of earlier loans, so using an early instalment as the referral definition would pull a common payment habit into the flagged group. Use the full loan schedule and closure record.
Product changes the reading. Early closure was more common for cash loans than point-of-sale loans, and current-loan default went in different directions across those products. For clients with a closed earlier loan, default was 8.1% after any early closure and 7.9% with no early closure. The adjusted odds were 0.91 for point-of-sale loans and 1.16 for cash loans. The ranking changed only marginally when any early closure was included.
Fourteen per cent of early closures were refinanced by a new loan from the same lender within seven days. Those clients had 6.7% default. Combining refinanced and other closures, or cash and point-of-sale loans, makes a simpler rule. It obscures the different results and can send more clients to review, with no gain in risk ranking.
Set the definition from the schedule. Split early closures by product and by refinancing before comparing default rates.
Watch cash loans closed in their first quarter
On this book, the higher-risk referral signal was a cash loan closed in the first quarter of its term and left unrefinanced at the lender. It covered 7.5% of current applications and had 10.3% default against 8.4% expected. The result held in both random halves.
The comparisons explain why the product and timing splits matter. Early point-of-sale loans closed in their first quarter had 7.8% default, and cash loans closed later in their term had 8.6%. The higher rate is specific to first-quarter cash closures.
That narrower rule refers about 750 clients per 10,000 current applications, and about nine in ten referred clients repay. Ask why the cash loan was paid off, where the repayment money came from, and what the new loan is for. This book supplies no threshold for turning an answer into a fraud finding.
Keep the rule referral-only, and compare its performance with the early payment default analysis of the next loan.
Test whether store rankings predict later payment difficulty
Use an out-of-time store test to decide whether excess early closures deserve a store-level referral rule. Rank stores in an older period by early closures above the level predicted by their products, then measure outcomes on later loans.
The test covered 150 coherent stores. For later-loan lateness and lateness in the first 12 instalments, compare the outcomes with expectations adjusted for product and quarter. For current-loan default, use the expected rate based on outside-score third, current product, region rating and the client’s earlier payment record. Compare the store groups with random groups of the same size.
Stores ranked highest for excess closures showed later performance in line with the comparison groups. For the top 5% by excess, the adjusted ratio for current default was 0.98, and 48% of random store groups reached that result. Across fifths of stores ranked by excess, later-loan lateness also showed no increasing pattern. A raw share can look striking because it carries the products sold. A store risk ranking needs to separate that product mix from later performance.
The payment amounts offer a separate check on the cash-at-the-counter idea. In 99.7% of early closures, the closing payment matched the lender’s calculated balance. Round-thousand amounts appeared in 0.003% of those lump sums, compared with 0.5% of all payments. The payment record has no calendar dates across clients and no payment-channel details, so a same-day batch test across clients cannot be run from this book.
Keep store-level referrals out of the rule set unless an out-of-time test shows later risk after the appropriate adjustments. For a same-day payment test, first obtain calendar dates for each payment.
Backtest referral volume and keep the decision with the reviewer
Backtest each candidate rule against its expected default rate and the review volume it creates. Per 10,000 current applications, the bust-out shape referred 107 clients and defaulted below expectation. The first-quarter cash-loan rule referred 750, with default above expectation. The same jump after an on-time closure referred 77 and also had default above expectation.
| Rule | Per 10,000 | Default | Expected | Verdict |
|---|---|---|---|---|
| Bust-out shape | 107 | 7.1% | 8.4% | wrong way |
| Bust-out, cash after point of sale | 91 | 6.9% | 8.6% | wrong way |
| Same jump, on-time only loan | 77 | 10.9% | 8.6% | survives |
| Cash paid off in first quarter | 750 | 10.3% | 8.4% | survives |
| Latest loan at a top 5% store | 75 | 10.1% | 8.8% | raw only, not by product |
| Any early closure | 4,100 | 8.1% | 8.1% | no signal |
A referral consumes staff time, while a decline can turn away a client who would repay. The comparison belongs in the backtest before the rule goes live. Use the fraud rule backtest to run each proposed referral definition over your historical applications, count referrals per 10,000, compare observed default with the group’s expected rate, and check whether the result holds across separate halves of the book. Use that comparison to select which cases reach human review.
The opening cases look alike until their outcomes are read beside the comparison group. Ari repaid, Noa defaulted, and Remi defaulted after making every scheduled payment before the same large jump. The first two cases show why the bust-out story attracts attention. Remi shows why an on-time closure belongs in the comparison.
Use the following procedure on the lender’s whole book. It separates the pattern that looks persuasive from the patterns that earned review on this book.
- Define early closure from the schedule: paid in full at least 90 days before the planned last due date, with no instalment 30 or more days late. Keep individual instalments paid early separate from loan closures.
- Separate closures refinanced by a new loan from your own lender within seven days.
- Split the remaining early closures by product, including cash and point-of-sale loans.
- Compare an early closure followed within 90 days by a loan at least five times larger with the same jump among clients whose only earlier loan closed on time. Read both observed rates against their expected rates.
- Watch cash loans closed in the first quarter of their term and left unrefinanced at your lender. Refer this pattern for review, with the reviewer making the decline decision.
- Rank stores by excess early closures in an older period. Measure later-loan lateness and first-12-instalment lateness against product-and-quarter expectations. Measure current default against its expected rate using outside-score third, current product, region rating and earlier payment record. Compare each result with random store groups of the same size.
- Almost no closing payment came in a round amount, so look at a store whose closing payments do.
- Backtest the candidate rules on historical applications. Compare referral volume and observed default with expected default, and check the result in both halves. Refer the supported patterns for human review. Ask why the loan was paid off, where the money came from, and what the new loan is for. Apply your ordinary credit criteria, using the full application evidence to decide whether to decline.
Questions about early repayment and fraud
Why do borrowers repay loans early, and which early repayments warrant a closer look?
Many borrowers repay because they no longer want the goods bought with the loan or want to avoid taking on too much debt. Investigate when early repayments cluster at one store, sales agent or payment channel, or when a payoff is followed by a much larger application. Check who made the payment, where the money came from and whether the pattern repeats across clients.
What should I investigate when early repayments spike at one store or sales agent?
Compare the store or agent with similar ones, accounting for loan volume and product mix. Check whether staff made repayments for clients or handled cash, confirm the transactions with borrowers, and look for links to sales targets or particular payment channels. A concentration at one location or agent is a reason to open an investigation into how those repayments were arranged.
How does a bust-out scheme use early repayment, and how can we catch it before funding a larger loan?
A borrower or coordinated group repays a small loan early to improve its borrowing record and qualify for a larger offer, then stops paying after that loan is funded. Before payout, review the repayment source, linked applications, current obligations and ability to repay. Escalate a pattern that points to coordinated borrowing or an unexplained payoff for investigation before releasing the larger loan.
How can I spot an early repayment made with someone else’s stolen or laundered money?
Check whether the payer and payment account belong to the borrower, especially when the payoff happens soon after the loan is issued. Ask the borrower to explain a third-party payment or a source of funds they cannot account for, then verify the explanation against available payment records. A payer the borrower cannot identify or explain warrants an investigation into the source of the money.
What information should we record for each early loan repayment?
Record the calendar date, amount, payment channel, payer’s identity and relationship to the borrower, and the account used. Keep the loan’s scheduled end date and actual closure date so investigators can distinguish a full early payoff from an early instalment. Note whether the payoff was refinanced by a new loan from your lender, since refinancing can change how the repayment pattern should be read.
