Build loan file checklists with ChatGPT, without customer data

Tools used

ChatGPTClaudeWhatsAppGoogle Sheets

Credit operations staff chase the same missing documents again and again: an unsigned form, an income proof older than the policy allows, a bank statement with the wrong period. This tutorial shows how a loan operations team in Mumbai can use ChatGPT to build and update document checklists for each product, and to write pending-document messages. It works only with product names, document types and anonymised file codes. No customer personal data goes into the AI tool.

What you need

• Your approved document list for each loan product, from your credit policy.
• A business ChatGPT plan, approved in writing by your compliance team. Get this before any loan-related work with AI.
• A Google Sheet called "File Status", using file codes such as NBFC-A-0142 instead of names.
• The current text of the RBI digital lending rules, read from the RBI's own circulars. This tutorial does not quote specific rule numbers, because those should be checked against the official text.

Step 1: Get compliance sign-off first

Before any file touches an AI tool, get written approval from your compliance head. The approval should name the tool, the account type and the data fields allowed. RBI's digital lending guidelines keep lenders accountable for their lending service providers, so the tools your team uses are also part of your compliance responsibility. Stablemoney's summary of the guidelines is a starting point, but read the circulars themselves.

Step 2: Write the master checklist without customer data

Use product names and document types only. Set up columns: Product, Document, Mandatory (Y/N), Valid for (days), Notes. An example row: Personal loan, Income proof, Y, 90, last three months' salary slips. Replace the example with your own policy values. Keep this sheet inside your company account.

Step 3: Draft the checklist from your policy

Paste your policy list, not a customer file.

Have your credit head review the output against the policy before it goes into the master sheet. Keep the prompt and the output together in the Policy versions tab, so the source of every checklist line can be traced later.

Step 4: Flag missing items from status codes only

Staff tick the documents received in the File Status sheet. Then copy only the file code, product and the lists of received and missing items. Never copy the documents themselves.

Step 5: Write the pending-documents message

Staff add the applicant's first name in WhatsApp by hand, so the name never goes into the AI tool.

Step 6: Keep an audit trail for policy changes

Keep a "Policy versions" tab in the sheet. Record each checklist version, the date it took effect, and who approved it. When the policy changes, compare the old and new lists instead of rewriting from memory. Put the version number in the file name when you share a checklist with branches, so nobody works from an old list by mistake.

Tips

• Keep one checklist per product, not one per branch. Branches should use the same policy.
• Replace every real identifier with a file code before pasting anything. If in doubt, leave it out. A file code should mean nothing to anyone outside your team.
• Review the checklists every quarter with your credit head. Policy changes faster than prompts do.
• Ask the model to say "check with compliance" when it is not sure. This keeps guesses visible.
• If a checklist line has no clear source in your policy, leave it out until compliance confirms it.

Keep it safe

Never paste Aadhaar numbers, PAN, bank account numbers, mobile numbers, addresses, salary figures or scanned documents into any consumer AI account. Use placeholder codes and document names only. Scanned documents stay in your loan management system. If your team wants AI to read scanned documents, that needs a separate review with compliance and a tool they have approved in writing.

Sources

More for NBFCs and finance: see what AI can do for your industry →