Customer Case Study · Credit
A GenAI co-pilot that reads the whole loan file, from twelve months of bank statements to the legal and technical reports, and drafts the credit appraisal memo. The underwriter still makes the decision, in minutes instead of hours.
Figures indicative; validated metrics available on request.
In affordable housing, many applicants are self-employed or paid partly in cash. Proving income means reading the bank statement line by line, and every file was done by hand.
Underwriters read 6 to 12 months of statements, ITRs, payslips and property papers, then typed the memo. About 3.5 hours a file.
Two underwriters could assess the same kirana owner's income differently. Cash deposits and one-off credits were treated case by case.
Missing documents surfaced only at credit review. Files went back to the branch, sometimes twice, adding days to turnaround.
A single bounced EMI or a pattern of round-figure cash deposits was easy to miss among four hundred transactions.
The co-pilot works inside the housing finance company's credit policy. Every number in the memo links back to the page it came from.
Select a step to see what happened.
We sat with credit teams in regional hubs and timed every step of the file.
Deliverable: Time-and-motion study and file-rework map.
Credit policy became the rulebook the co-pilot follows.
Deliverable: Policy rule-pack and CAM template signed off by Credit.
Claude does the reading and drafting; deterministic rules do the maths.
Deliverable: Co-pilot in a controlled environment, with an audit trail.
Underwriters ran the co-pilot in parallel before relying on it.
Deliverable: Live in pilot hubs, with underwriter feedback loop.
The same engine now supports more of the book.
Deliverable: Rollout plan and model-monitoring dashboard.
Choose an applicant and press Run co-pilot. It reads the file, classifies a year of transactions, assesses income, checks eligibility against policy and drafts the memo. Then take over: drag the loan on the eligibility chart, switch the transaction view or click a red flag.
Every page of every document, OCR'd and parsed
Each dot is one transaction; size is the amount. Switch to the AI view to sort them into what they are.
Inflows above the line, outflows below. The orange line is income as assessed under policy for this profile.
Each petal is a month of assessed income; the dashed ring is the median. Green is within 15% of it.
The shaded area is every loan and tenure the applicant qualifies for. Drag the dot, or use the sliders.
Click one to see the transactions behind it.
Updates live as you change the loan. Underwriter reviews and signs.
Applicants, transactions and policy limits are illustrative and do not represent any real customer. In production the co-pilot runs inside the housing finance company's environment on the actual loan file; figures indicative.
With the memo drafted and the missing-documents list raised on day one, files moved through credit faster and came back to branches less often.
Distribution of turnaround, affordable home-loan files, working days
First-time right vs sent back to branch, pilot hubs
Figures indicative; validated metrics available on request.
The biggest time sink was reading and typing, not deciding. Take that away and underwriters spend their time on the judgement calls.
Encoding the credit policy as rules made the co-pilot consistent, explainable and easy for risk to sign off.
Linking each figure and flag to its page is what earned the trust of approvers and auditors.