Case study · 2025

From a 9-day loan decision to under 4 hours

A non-banking financial company processing 1,200 loan applications a month took nine days to reach a decision, losing most applicants to faster competitors. We rebuilt origination end to end.

Client: Confidential (NBFC, South India)
Sector: Fintech & Banking
Duration: 8 months

Fintech & Banking

Confidential (NBFC, South India)

Laravel Python React PostgreSQL AWS Mumbai

<4 hrs

median decision time, from 9 days

71%

of applications auto-decided

3.4x

increase in completed applications

100%

of decisions individually auditable

The challenge

Where things stood

Applications arrived as physical documents at branches, were couriered to a central office, keyed into a legacy system, and manually assessed. Nine days from application to decision meant roughly 60% of applicants had accepted an offer elsewhere by the time a decision was ready. The paper trail also made audit reporting a manual reconstruction each quarter.

Our approach

What we did about it

We modelled the actual decision, not the stated process, and found that 71% of applications were decided on the same six data points. Those cases did not need a human at all; the remaining 29% deserved the underwriter's full attention.

Document extraction now reads bank statements, salary slips and identity documents automatically, with a confidence threshold that routes anything uncertain to a person rather than guessing. KYC verification runs against the standard bureau and government APIs.

A configurable rules engine handles the straightforward decisions, and every decision — automated or human — writes to an immutable audit log with the exact inputs and rule versions used.

What we delivered