Why bounce count is the wrong metric
Most bank statement review processes count bounces: zero bounces is good, one bounce is acceptable, two bounces is a soft flag, three or more is a hard flag. This approach produces two systematic errors. It over-rejects borrowers with technical bounces — a SEPA Direct Debit failure caused by a bank system error, immediately resubmitted and cleared the same day, is logged as a bounce but is not a credit risk event. And it under-detects borrowers with structurally dangerous bounce patterns — a borrower whose salary arrives on the 7th but whose SEPA Direct Debit debit runs on the 1st will have a bounce-and-clear pattern every month that counts as "one bounce per month" when it is actually a structural cash flow timing problem that will persist and worsen when an additional instalment is added.
The Bank Statement Analyst AI analyses bounces in four dimensions: frequency (how often), severity (how long between bounce and clearance), pattern (random or systematic), and trend (improving, stable, or worsening). The four dimensions together produce a bounce risk score that is far more predictive of future default than a simple count.
The bounce pattern taxonomy: what each pattern means
| Pattern Type | Description | Default Correlation | Bank Statement AI Interpretation | Credit Signal |
|---|---|---|---|---|
| Cascade bounce | Multiple EMIs bouncing in the same month — 3+ separate SEPA Direct Debit returns in one calendar month | High: 68% of cascade-bounce borrowers default within 12 months | Income has collapsed or the borrower has exceeded their debt capacity. Multiple instalment obligations failing simultaneously indicates systemic inability to service current debt, not a timing issue. | Decline trigger |
| Worsening frequency bounce | Bounces increasing in frequency over the statement period — 0 in months 1–4, 1 in month 5, 2 in months 6–8, 3 in months 9–12 | High: deteriorating bounce pattern predicts delinquency onset within 6 months | The borrower's debt load has been increasing over the period and their capacity to service it is declining. The worsening trend is more predictive than any snapshot point in the history. | High risk flag |
| Salary-timing structural bounce | SEPA Direct Debit debit bounces every month on day 1 and clears on day 5–8 when salary arrives | Medium: not inherently a default risk but will worsen with additional instalment | The borrower's salary arrives after the SEPA Direct Debit debit date. This is manageable at current debt level but any additional instalment increases the monthly cash flow gap. The structural timing mismatch must be identified — not just the bounce event. | Structural flag — SEPA Direct Debit date change required |
| Technical/bank-error bounce | Single bounce with same-day or next-day clearance, no prior bounce history, RTGS credit same day | Very low: bank system error, not borrower cash flow event | SEPA Direct Debit system errors, bank maintenance windows, and processing delays produce bounces that clear within hours. Distinguished from genuine bounces by: immediate clearance, no balance issue on the day, prior clean history. Not a credit risk signal. | Not a risk signal |
| Isolated bounce with clearance | One bounce event in 12 months, clearance within 5 days, no pattern repetition | Low: one-off event, not predictive | A single bounce cleared within 5 days against an otherwise clean 12-month history is a life event (travel, payment forgotten, brief liquidity squeeze) not a structural risk. The clean history around it is the context the bounce count metric ignores. | Context note — not a decline trigger |
| Improving bounce trend | Bounces present in months 1–6 of the statement, zero in months 7–12 | Very low to low: improving trend indicates financial stabilisation | The borrower had bounce events earlier in the 12-month period but has cleared their bounce history. An improving trend — if the later period is clean — should not be penalised at the same rate as a recent or current bounce pattern. | Positive trend — reduced risk |
A bounce calendar: reading 12 months of SEPA Direct Debit history
What the four-dimension bounce score produces
For the account above, the four-dimension analysis produces: Frequency score — 12 bounce events in 10 months (score: very high, 94th percentile of delinquency-preceding accounts). Severity score — clearance time increasing from 4 days to 21+ days over the period (score: high and worsening). Pattern score — systematic monthly bounce on the 1st with increasing severity, indicating a structural debt service problem not a timing issue (score: structural pattern detected). Trend score — clearly worsening (score: deteriorating, maximum penalty applied).
The combined bounce risk score of 88 out of 100 produces an automatic credit recommendation: decline this application. Adding a new instalment to an account that is already in a cascade bounce pattern does not create a borrower who will struggle to service the new loan — it creates a borrower who will default on all their existing loans faster than they would have without it.
The bounce is not the risk — the pattern is
A borrower with one bounce in 12 months, cleared in 48 hours, who earned €120,000s every month without fail, is not a default risk. A borrower with 12 bounces in 10 months, with clearance time increasing from 4 to 21 days, whose account balance was zero on the instalment date seven of those months, is a default risk regardless of any other indicator in their application. The Bank Statement Analyst AI reads the pattern across the full 12 months, scores each dimension independently, and produces a bounce risk assessment that distinguishes the life-event bounce from the structural deterioration that predicts default — because those are different situations, and credit decisions made on the basis of a bounce count treat them as the same.