Field sales teams submit hundreds of hotel, meal, fuel and travel bills each month: thermal printouts, handwritten Hindi cash memos and phone photos. HMC built a reimbursement audit desk for the cement company's Finance team that reads each bill, checks it against the salesperson's entitlement and approved tour plan, and confirms the bill came from inside their geo-fenced beat.
Figures indicative; validated metrics available on request.
Sales officers and managers travel across their beats every week. Finance had to settle their claims quickly, fairly and within policy, with only paper and photos to go on.
Faded thermal slips, handwritten Hindi cash memos and blurry phone photos, all typed in by hand.
Caps differ by grade and city tier. Checking each bill against the right cap took time, so most were not.
A dinner bill 200 km from the approved beat looked the same as one next door.
Claims waited weeks; deductions arrived without reasons and salespeople pushed back.
We turned the travel policy into rules a computer can apply, taught it to read every bill format the field sends, and added location as a check.
Select a step to see what happened.
Three months of claims, bills and deductions reviewed with the finance team.
Deliverable: Leakage map and bill format library.
Every line of the travel policy written as a check.
Deliverable: Rulebook signed off by Finance and Sales.
From photo to decision in seconds.
Deliverable: Audit desk and claim register.
The desk recommends; Finance decides.
Deliverable: Pilot results and SOP.
Same rules for everyone.
Deliverable: Roll-out and monthly audit pack.
Seventeen sample bills from four salespeople for one October week. Read a bill, see how it is checked against entitlement and tour plan, and see where it was issued on the map. Upload your own bill with an OpenAI key.
Optional. Without a key, the demo reads the sample bills and pulls details out of pasted bill text with rules. Add your OpenAI key to read photos of any bill, printed or handwritten, Hindi included, and to draft deduction notes.
Pick a sample bill and press Read, or upload a photo of your own. Every field is pulled out with a confidence score, and every bill read lands in the table below, ready for CSV.
| # | Source | Salesperson | Vendor | City | Bill no | Date | Category | GSTIN | Items | Tax ₹ | Total ₹ | Decision |
|---|
Synthetic sample bills: vendors, GSTINs, people and amounts are illustrative. Uploaded bills are sent only to OpenAI with your key and are not stored.
Caps depend on grade and city tier; meals are capped per day; bills must match the approved tour plan. Open a bill to see all nine checks. Move a cap and every decision updates.
Synthetic policy and people. Checks: ① read ② arithmetic ③ GSTIN ④ entitlement cap ⑤ tour plan date ⑥ geo-fence ⑦ duplicate ⑧ submission window ⑨ non-reimbursable items.
Each fence is a circle around a town in the day's approved tour plan. Bill pins sit at the vendor's address; small dots are GPS check-ins from the field app. Green pins are inside the fence for that day, red pins are outside.
Vendor addresses are geocoded (Google Maps Places in production; a built-in gazetteer of UP towns in this demo). "Open in Google Maps" on each pin shows the exact spot.
The claim register shows what is payable now, what is held for an answer and what is disallowed. Draft the note each salesperson receives, from templates or with GenAI.
Choose a salesperson and draft the settlement note.
The desk recommends; Finance approves every claim before payment.
Finance moved from sampling one claim in ten to checking every bill, and settles within days instead of weeks.
Sankey: claim value through checking to outcome
Parallel coordinates: one line per claim
Figures indicative; validated metrics available on request.
Once caps, tiers and windows are rules, checking every bill costs the same as checking one.
A vendor address plus the day's tour plan answers "was this on the beat?" without accusing anyone.
Disputes drop when the salesperson sees the rule, the cap and the amount, line by line.