Scored book sample audit

A retrospective reading of applications that already received a score, to see whether the inputs the engine used were faithful to the underlying financial file.

Laptop and printed reports on a meeting table during a file review

Once a file has been scored, the temptation is to treat the engine output as the record. The record is still the application. This engagement draws a sample from a scored book — personal financing, hire purchase, credit cards, or SME working capital — and reconstructs whether the financial inputs that sat in the scoring interface match the paper and statements behind them.

We look for transcription drift, optimistic income construction, omitted commitments, and stale bureau extracts. We also note files where the pack is stronger than the coded inputs, which is a different operational problem but still a file problem.

This is not model validation. We do not test characteristic stability, Gini, or reject inference. We test whether the application the engine met was the application the customer and the originator actually assembled.

Time. Typically 10–20 working days

Fees. Quoted after scoping. Indicative range RM 18,000–RM 55,000 depending on sample size and products in scope. Fees are informational. We do not take payment on this site.

Who this is for

Portfolio quality, model monitoring counterparts, and lenders preparing for an internal or regulatory review of originated files.

In the report

Sample from an agreed origination window and product set
Input-to-exhibit tracing for income, commitments, product, and bureau fields as provided
Error and omission log with severity notes
Thematic findings (for example, bank-statement windows that do not cover the stated income period)
Report and a working session with credit operations

Outside this engagement

Scorecard or characteristic analysis
Collections or arrears file review unless separately scoped
Customer outreach

Request this reading See the method