Reference scorecards

NEXCLARA ships with three reference scorecards, one per credit family. They are the starting point of the institution's policy; they are validated by no institution and cannot be activated as they are.

How a scorecard rates an application

Each scorecard starts from a base score and adds, for each variable, the points of the band in which the application's value falls. The total score reads on a scale from 0 to 1,000, divided into outcome zones: approval, conditional approval, review, decline.

Three principles apply to every scorecard:

• no point is invented: if a variable is missing and the scorecard has no "not provided" band, the score is declared not computable and the application goes to review;
• missing data is never a zero: it stays missing, with its reason, all the way into the explanation;
• some data is restricted: it may count in the score, but is never presented to the applicant as a reason for the decision.

What the institution must set

A reference scorecard enters a policy as a draft, with empty parameters. Before any activation, the institution sets its own values, for example:

• the completeness threshold below which no automatic approval is possible;
• the product cap and the maximum term taken into account;
• the target debt-service ratio and the indicative monthly rate used to compute the maximum amount;
• depending on the scorecard, the age bounds of the exclusion rule, or the lists of favourable sectors and sectors under watch.

The values used by the platform's tests are synthetic: they come from no institution and are recommended to none. See Decision policy lifecycle.

What is still under construction

• No statistical model is used: the model's share is zero and no probability of default is produced. The score comes from the scorecard alone.
• No point adjustment is defined by the reference scorecards: the adjustment stage exists in the engine, but it is empty.
• Weights, bands and zones cannot be changed from the interface; only the institution's parameters are set on screen.
• Several readings of the specifications remain to be validated with a pilot institution (band boundaries, confidence levels, maximum amount calculation); they are recorded without being decided on the institution's behalf.