Why Bureau-Only Underwriting Is Quietly Failing Thin-File Borrowers

There is a comfortable assumption in a lot of NBFC credit teams: if the bureau score is decent and the bureau history is clean, the loan is probably fine. For prime and near-prime borrowers, that assumption holds up reasonably well. For the segment most Indian NBFCs and MFIs actually serve, it is starting to fail in ways that do not show up until the book has already been built.

The thin-file problem is not going away

A large share of first-time and small-ticket borrowers in semi-urban and rural India have either no bureau history or a history so thin that the score is effectively noise. A single closed tractor loan from four years ago does not tell you much about repayment behaviour on a working-capital loan today.

What we see across Finpage implementations is that lenders in this segment fall into two camps. The first treats the bureau as the primary gate and everything else as supporting paperwork. The second treats the bureau as one input among several and builds the actual decision around cash-flow evidence, group and household data, and behavioural signals from the field.

The second camp is growing their books. The first camp is either rejecting good borrowers or, worse, approving bad ones because the bureau said nothing alarming.

What "alternative data" means in practice

The phrase gets overused, so let me be specific about what actually moves the needle for small-ticket lending:

• Bank statement cash-flow analysis, even six months of it, tells you more about repayment capacity than a bureau score for a borrower with no formal credit.
• Repeat-cycle behaviour within your own book. A borrower on their third cycle with you who has never been more than three days late is a better risk than most bureau-scored strangers. Surprisingly many lenders do not systematically use this.
• Field verification outcomes captured as structured data, not free text. A tele-verification or house-visit result that is recorded as a score, not a note, can feed a model. A note cannot.
• Household and group-level exposure, particularly in the JLG model, where one member's stress is a leading indicator for the others.

The systems obstacle

The reason more lenders do not do this is rarely lack of will. It is that their origination system captures the bureau pull as a structured field and everything else as attachments and remarks. You cannot build a decisioning rule on a PDF.

This is a design choice we made deliberately in Finpage: verification outcomes, cash-flow summaries, and cycle history are first-class data fields, so a credit policy can actually reference them. It sounds obvious. It is not how most legacy systems in this market were built.

If you are running a bureau-first credit policy on a thin-file book, it is worth asking a simple question: what percentage of your rejections had no bureau history at all? If that number is large, you are probably leaving good business on the table, and if your delinquencies are concentrated in bureau-approved loans, the bureau is not doing the job you think it is.