The Score Version A Lender Uses Is Often Old

Scoring models are released in versions, and the release of a new one does not retire the old. Several generations run at once across the market.

Why a new model does not replace the old one

A lender's underwriting is calibrated to a specific model. Cutoffs, pricing tiers and approval rules are all set against the score distribution that model produces.

Switching versions means revalidating those rules, adjusting cutoffs and testing the effect on approval and loss rates, which is a substantial project.

Systems have to change too, along with regulatory documentation of how the model is used, so the incentive to move is weaker than the availability of a newer model suggests.

Where older versions persist longest

Lending secured by property tends to move slowest, because the models accepted are specified by the entities that purchase the loans rather than chosen by the lender.

Portfolios with long histories also resist change, since existing accounts were underwritten and are managed against a particular model's outputs.

Newer entrants and newer products adopt current models more readily, which means the version in use correlates with the age of the lending program.

What differs between generations

Successive versions change how particular items are weighted, including how certain collection accounts and paid derogatory entries are treated.

Some versions incorporate additional data types or trended fields that earlier generations do not read at all.

Because the treatment differs, the same file can produce meaningfully different scores across versions without any of them being incorrect.

Why consumer-facing scores add another layer

Scores shown by card issuers, banks and monitoring services are real scores, but they are the version and the model that provider chose to display.

They may come from a different bureau than the lender will pull, and they are calculated on that bureau's data at that moment.

The number therefore functions as a directional indicator rather than as a prediction of what an application will return.

What to ask instead of comparing numbers

Lenders are generally required to disclose the score used in a credit decision along with the reason codes attached to it, which identifies the model and the source.

Those reason codes are more informative than the number, because they state what the model found rather than where it landed.

Tracking one score consistently over time shows direction, which is the useful signal. Comparing scores from different sources measures the difference between models rather than any change in the file.