Every lender has had the meeting, the one no forecast seems to prevent. A pool underperforms. Someone asks what changed. The answer takes three weeks to assemble, because it lives in four systems owned by four functions. The credit file shows the score. It does not show who overrode it, or what the platform could not capture at the time. 

A delinquency rate reports the outcome and cannot explain it. By the time a loss surfaces, technology, operations and a vendor have each shaped the deal. Most credit risk management software records none of that. The rate tells you what happened. It cannot tell you the mechanism.

The aggregate number is calm, and that is the problem 

In the first quarter of 2026, auto loan balances moved into serious delinquency at an annualized 2.97%. A year earlier the figure was 2.94% (New York Fed household debt and credit report, 2026). Three basis points on a $1.69 trillion book. Read on its own, that is stability. 

Read against the rest of the table, it is ambiguous. The same flow across all household debt rose from 2.45% to 2.83%. Early delinquency eased for credit cards, from 8.7% to 8.6%. Mortgages eased from 3.9% to 3.8%. Meanwhile the 60-month new-car rate at commercial banks fell 51 basis points to 7.53% (Federal Reserve G.19 consumer credit, 2026). 

Auto held flat while parts of that picture improved and parts deteriorated. A flat rate is equally consistent with a safer book, a riskier book and a reshuffle underneath. It cannot tell them apart, because it counts outcomes rather than causes. To find the mechanism, follow one deal. 

Follow one deal and credit risk management software runs out of record 

  • A borrower applies. The origination system captures score, income and term. It does not capture the equity rolled in from a prior contract, because that field was never built. Two deals with the same score can carry very different loss severity on that difference alone. Technology risk has already narrowed what anyone can measure 18 months from now. 
  • The deal fails an automated rule. An underwriter approves it anyway and types the reason into a free-text note. This is operational risk. The exception is real but the reason is unsearchable, so that population can never be isolated or back-tested. 
  • Funding needs a document the core cannot produce, so a settlement tool bolted on two years ago produces it. That is vendor risk. The tool sits on its own release cycle, governed by procurement on a renewal calendar rather than by risk on a register. The 2023 *Interagency Guidance on Third-Party Relationships* asks banking organizations to tier third parties by criticality across the relationship life cycle. Few tier the one holding their contracts. 
  • The deal defaults. The credit file shows a score and a term. Three decisions that shaped the outcome left no record anywhere credit can read. 

Four fields decide whether you can reconstruct a loss 

None of this needs new analytics. It needs rolled-in equity at origination, contract term and payment-to-income recorded where the credit team can query them, along with a coded reason for every override. 

With those, loss given default becomes segmentable. The overridden population becomes a cohort you can track. Platform criticality becomes a line on the risk register with an owner instead of a renewal date. Without them, every post-mortem restarts from the credit file, which is the one document that cannot answer the question. We work through all four in our automotive finance whitepapers. 

credit risk management

Figure: Four risks shape one deal. Three of them leave no record. 

What to take from this 

  • A flat delinquency rate fits a safer book or a riskier one, so alone it settles nothing. 
  • A credit loss is shaped by technology, operational and vendor decisions long before it becomes a credit statistic. 
  • Manual overrides are unpriced credit decisions, and a free-text reason makes them permanently unmeasurable. 

Frequently asked questions 

What should credit risk management software capture that older systems do not? 

Rolled-in equity at origination, contract term, payment-to-income and a coded override reason, all in queryable form. Without them, loss given default cannot be segmented however good the model layer above it is. We cover the wider shift in auto finance trends. 

Why does vendor risk belong in a credit discussion? 

Because platform capability sets the ceiling on credit measurement. The 2023 interagency guidance already treats critical third parties as risk-managed relationships, not purchasing decisions. 

Risk-aware means explainable, not conservative 

The instinct when a portfolio looks uncertain is to tighten the credit box. That trades volume for a feeling of control and leaves the problem untouched, because a narrower box still produces losses nobody can explain. 

Capture the four fields, code the override reasons, put the platform on the risk register. The next post-mortem takes an afternoon rather than three weeks. Risk-aware lending is not lending less. It is being able to say why. 

In Transcend Finance, our automotive finance platform, those four fields sit in origination and credit analysis with override reasons coded rather than typed. If your last portfolio review raised more questions than it answered, we are happy to talk it through.

Related blogs