Bureau-only scoring misses important BNPL risk signals. Bureau scores reflect credit history built over months and years, while BNPL exposures commonly last four to twelve weeks. The behavior associated with a BNPL default is therefore not identical to the behavior associated with a personal loan default, and a model designed for one may assess the other poorly.
For BNPL, the most relevant signal is often current account behavior: what the borrower is doing with their bank account in the four weeks before and during the BNPL tenure. Bureau data cannot provide that view in real time. Transaction data from banks connected through the AA framework can.
Why Bureau Scores Fall Short for Short-Tenure BNPL
A bureau score combines credit behavior across accounts over a substantial period. It captures historical repayment discipline, current outstanding obligations, credit utilization, and enquiry patterns. These are valid indicators for medium and long tenure products, where the central question is whether the borrower has consistently met obligations over time.
For BNPL, the question is more immediate: can the borrower currently meet four weekly payments of Rs. 2,500 each? A borrower with a strong bureau score who recently lost employment income and has a low current-month balance presents a different risk from the score alone. Conversely, a borrower with a thin bureau file, consistent salary credits over the past six months, and a stable end-of-month balance may be safer than the bureau record indicates.
BNPL defaults can follow liquidity shocks, including an unexpected expense, a delayed salary, or a sudden rise in other obligations. Transaction records may show these events through irregular balance declines, missing regular credits, or clusters of large outflows. Bureau data generally cannot show them until a default has happened and been reported.
Bank Transaction Signals for BNPL Risk
Several transaction-data signal groups are useful when assessing BNPL risk:
End-of-month balance pattern: Review the average end-of-month balance over the past three months and the minimum balance in the most recent month. Persistently low balances, where outflows consume most monthly income, leave less capacity to absorb an unexpected BNPL payment than a meaningful balance cushion does.
Recent income consistency: Check whether a regular salary credit or equivalent inflow arrived in each of the past three months, and whether the latest credit arrived when expected. A delayed salary in the latest month can indicate short-term liquidity pressure for a product with a one-month first repayment.
Existing debit obligation load: Measure recurring NACH/ECS debits against monthly inflow. When existing obligations take up a high share of income, less discretionary cashflow remains for a new BNPL payment.
Recent NACH returns: Recent NACH return events in transaction records can indicate payment stress that has not yet reached the bureau. That signal matters particularly for BNPL, where the amount may be small but repayment begins immediately.
Unusual activity in the past two weeks: Large withdrawals, a sudden balance decline, or changed UPI patterns shortly before application may point to a short-term liquidity event. These indicators are noisy and need context, but transaction data makes them observable at underwriting time.
Combining Bureau and Transaction Signals
The appropriate BNPL approach is not to discard bureau scores in favour of transaction signals. Each source describes a different risk horizon. Bureau data helps identify chronic credit-management problems, such as repeated defaults, excessive obligations across lenders, or recurring delinquency. Transaction data helps assess current liquidity and whether the borrower can service a short-term obligation now.
A combined framework can use bureau data to screen for chronic risk at population level, then use transaction data to refine the individual assessment. This is particularly relevant in the middle of the bureau score distribution, where the population is large and bureau-only prediction is weakest.
Strong bureau and transaction signals support low-risk approval. Poor bureau signals may support rejection regardless of current transactions. The additional value of transaction data appears in the middle: a borrower with an adequate bureau score may show current liquidity stress, while a borrower with a thin bureau file may show stable income and balance patterns. In either case, the bureau score alone is incomplete.
Implementation Considerations for BNPL Lenders
BNPL applications often occur at checkout, with the borrower expecting a near-instant decision. That creates tension with the AA consent-and-fetch process, because the borrower must complete consent before data becomes available.
One option is pre-consent during account registration. The borrower agrees to bank-data access before the first purchase, and the lender fetches and refreshes the data at defined intervals during the relationship. This moves consent friction to onboarding rather than the transaction moment.
Another option is bureau-only assessment for the instant transaction decision, followed by AA-sourced transaction data for a periodic portfolio refresh. Accounts can then be flagged when current transaction signals diverge materially from the bureau classification used at origination. This is not a real-time model, but it provides a systematic way to identify changed risk after origination.
The technology is secondary to having clear answers: when is transaction data collected, how often is it refreshed, and how does it affect actual credit decisions? Without those answers, a lender is discussing transaction data rather than using it.
A Note on Model Validation
A model using transaction signals requires validation on a representative outcome sample before operational use. AA transaction data is relatively new for many BNPL lenders, and these behavioral signals are grounded in what the data represents, not in a large validated dataset for any particular lender portfolio.
These signals are suitable starting points for hypothesis testing, not finished production specifications. Predictive weights, thresholds, and combinations for a particular BNPL product and borrower segment must be tested against that lender's own outcome data. Anything stronger would overstate the available evidence.