Reason Codes Say More Than The Score

A score reduces a file to a single number, which discards everything about why. The reason codes issued alongside it restore some of that information, and they are the more useful output.

Codes identify the largest deductions

A scorecard assigns points across several characteristics. Reason codes report the characteristics on which the file scored furthest below the maximum available.

They are calculated against the model's own scale rather than against other consumers, so a code can appear on a strong file simply because that factor was its weakest.

The ordering matters: the first code identifies where the largest number of points was lost, which is a more precise statement than any general description of the file.

The wording is standardised and terse

Codes are drawn from fixed lists, so the phrasing is generic: too many recent searches, balances too high in proportion to limits, insufficient history, recent adverse information.

That generality means a code names a category rather than an account. It does not identify which balance or which search produced the deduction.

Reading the file alongside the code is what turns it into something actionable, because the code says what kind of thing cost points and the file says which item it was.

Codes describe the model, not the lender

The codes accompany the score that produced them. A different model, built on different data with a different outcome definition, will surface different weaknesses.

A consumer-facing score and a lender's internal model can therefore disagree about what the main problem is, without either being incorrect.

Where a decline notice includes reasons, those reasons come from the lender's own process, including policy rules that no score would ever mention.

Some codes cannot be acted on

Codes referring to the length of history or the age of accounts describe facts that only time changes. They are reported because they cost points, not because they suggest an action.

Others describe the absence of something, such as limited experience of a type of credit, which is not a defect so much as a gap in what the model can observe.

Treating every code as an instruction leads to activity that adds searches and new accounts, which frequently makes the score worse in the short term.

Disclosure requirements differ by market

Some jurisdictions require lenders to give reasons for an adverse decision, and specify how those reasons must be expressed. Others require only that a decline is communicated.

Where consumer scores are sold or provided by agencies, the accompanying explanations are a commercial product feature rather than a regulated disclosure.

The obligations attached to automated decisions in particular have been changing in several markets, so what must be explained today may differ from what was required previously.