What OJK Expects From AI-Assisted Lending Decisions

Indonesian banks and multifinance institutions are required to be able to explain and defend every credit decision they make, regardless of whether a human or a model produced the underlying analysis. That doesn't change when AI gets involved; if anything, it raises the bar on documentation.

The standard isn't "use AI" or "don't use AI." It's "can you explain the decision"

For a lender working with OJK's risk management expectations, the practical question isn't whether AI assisted with a credit decision. It's whether the decision, and everything that fed into it, can be reconstructed and explained on request.

That means a useful AI underwriting tool needs to produce, alongside the score itself:

The specific documents and data points the score was based on
The reasoning connecting those inputs to the output
A log of who reviewed the score and what action they took
A way to flag and document exceptions or overrides

A score with no visible reasoning is a liability, not a shortcut, regardless of how accurate it is.

Where this shows up in practice

A few moments where explainability isn't optional:

Internal audit. Risk and compliance teams need to sample decisions and verify the underwriting criteria were applied consistently.
Regulatory examination. OJK can request the basis for specific credit decisions, particularly for non-performing loans.
Borrower disputes. A rejected or disputed borrower is entitled to understand why.

What this means for choosing an AI underwriting tool

The relevant question for evaluating any AI-assisted underwriting tool isn't just "how accurate is the score." It's "can I hand this score, with its reasoning, to an auditor or examiner without extra work." Tools that produce a number without the supporting trail push that documentation burden back onto the lender's own team, which defeats much of the point of using AI in the first place.