20 November 2025
Sampling personal financing files without pretending the sample is the book
A well-drawn sample of application packs can show how income and purpose are being evidenced. It cannot certify a portfolio. The report should say so in plain language.
Personal financing books in Malaysia are large enough that nobody reads every pack. Sampling is the honest method. It is also the method most often oversold. A sample of thirty files from one origination month is a window. It is not a certificate that the other several thousand files are clean.
We design samples with the client, not for theatrical randomness. If the question is income construction on a new programme, we overweight that programme. If the question is overrides, we sample overrides, not the through-the-door population. Stratifying by branch or by introducer is useful when the operations story already points there. Stratifying because a textbook said so is not.
The report then has to carry the limit in the same tone as the findings. ‘In this sample, twelve income figures could not be rebuilt’ is a finding. ‘Therefore origination is weak’ is a leap. We will write the first sentence. We will not write the second unless the sample was designed, and sized, to support it — which, in a typical two-week engagement, it is not.
Clients sometimes ask for a percentage defect rate to put in a committee pack. We can count defects in the sample. We will label them as sample counts. If a statistical extrapolation is required, that is a different design with a different fee and a different timeline. Most credit-operations questions do not need it. They need examples that originators recognise.
Purpose codes deserve a special mention in personal financing. Files that say ‘education’ or ‘medical’ and then show disbursement patterns that look like refinancing of other consumer debt are not a scoring-engine problem. They are an application-narrative problem. A sample that only tests whether a purpose field was filled will miss them. A sample that reads the pack will not.