ICAI USE CASE PRESENTED ON 07-10-2026            ProcessMyBill (AI Reimbursement Agent)

Employee reimbursement still runs on forms, portals and emails. Employees spend 10+ minutes per claim and rarely know what is claimable; approvers sign off without context; finance keys in every bill, checks policy from memory and chases missing documents for days; and the accountant re-keys the entry. The result: duplicates and over-cap items get paid, GST input credit is lost, and the audit trail is weak.

2. Solution Overview

ProcessMyBill takes a claim from a WhatsApp chat to a posted journal entry in Zoho Books. It is not "AI deciding". AI does one job: it reads each bill into a fixed schema (vendor, GSTIN, invoice no., date, tax, amount, line items). Python rules then decide - every GST and policy check is deterministic code, and every company number comes from the company's own policy table. People make the final call: the employee confirms, the approver approves, and finance validates before anything is posted.

AI reads. Python rules check. People decide.

3. The Process

1. Send bills: the employee sends photos or PDFs on WhatsApp, even during the trip. Bills build up under one claim ID.
2. Read and check instantly: each bill is read in about 5 seconds and checked at once; the employee gets a reply only when something needs attention (a cap, a duplicate, a blocked item).
3. Submit: the employee gives the purpose, answers the open questions in one pass and can correct any amount, vendor or date. Any correction is shown to finance as "was / now". A claim of 5-6 bills takes about 3 minutes, clarifications included.
4. Finance review (only if flagged): claims with an open flag land in the finance queue first; clean claims go straight to the approver.
5. Approval: the reporting manager gets an email with signed links - approve, partially approve (line by line, with a reason) or reject - from the phone. Reminder at 3 days, escalation at 7; never auto-approved.
6. Validate and post: finance sees the bill image beside the fields, every check result and the proposed journal (GL, input GST, payable). Approve & post creates the Zoho Books journal with bills attached; it never posts twice. The employee is told the payment date (16th / month-end).

4. Questions Asked at Submission - Examples

5. Generic Checks (same for every company, written once in code)

6. Company-Specific Checks (from the company's policy table - 23 rules, each citing its clause)

Every check returns one outcome - Pass, Fix, Explain, Cap, Educate, Refuse, Silent (finance only) or Review - with the reason and clause. Changing a limit means editing the policy table, not the code.

7. Technology and Deployment

The server is already deployed; WhatsApp and approval links reach it over a secure HTTPS endpoint. A company is onboarded with one masters workbook (employees, grades, approvers, expense heads to GL, policy rules) and its policy text. Only the bill image goes to the AI; masters and data stay on the server. AI reading costs well under ₹1 per bill.

8. Key Metrics

9. Target Users & Benefits

1. Employee: about 3 minutes on WhatsApp instead of forms and days of follow-ups; knows status and payment date
2. Finance: no data entry or chasing; reviews only flagged claims; posting is one click
3. Approver: decides from the phone with the bill, the flags and the policy in front of them
4. Company: leakage stopped before payment; GST credit protected; every decision on an audit trail
5. Built by a CA: a non-programmer built it with Claude Code in about six days, module by module with tests first