Loan underwriting agent with explainable credit decisioning.

The challenge

Retail-lending volume kept climbing while underwriter capacity stayed roughly flat, which stretched decisioning windows on personal and auto loans into days. RBI explainability requirements meant every rejection had to be defended in writing, so the team could not just throw a black-box model at the queue.

How we approached it

Ajuni embedded Lakshya into the Finacle 10.2 loan workflow with a transparent reasoning trail attached to every recommendation, and Riya handled reconciliation against the credit bureau pulls. The deployment ran inside HDFC's existing Azure India South tenant with ISO 27001 controls inherited from the bank's platform.

Outcomes in production

• 62% faster decisioning on personal and auto loans
• Per-run cost driven to ₹0.42 at production scale
• Reasoning trail attached to every recommendation
• RBI explainability standards cleared in design review
• Underwriter capacity freed for high-value casework

Stack & guardrails

Timeline

1. Week 0–2

   Model-risk scoping

   Reasoning-trail design reviewed with model-risk and compliance teams.

2. Week 3–4

   Finacle integration

   Lakshya wired into existing loan workflow with read-only credit pulls.

3. Week 5–7

   Shadow underwriting

   Agent recommendations ran parallel to human underwriters for 6 weeks.

4. Week 8–10

   Regulator demo + live

   Reasoning trail walked through with internal regulator liaison; full cutover.