2,047 invoices daily, 94.3% untouched
A small chartered-accountancy firm in Gurgaon took in more than 2,000 vendor invoices a day in a shared Drive folder. Three junior accountants spent six hours each keying them into Tally. That data entry cost about 180,000 rupees a month in salary. The builder set three Claude Code sub-agents under a small Python orchestrator on a 15-minute cycle. One pulls the vendor name, tax ID, line items and totals from each PDF as structured data. One checks the tax math and matches the vendor master. One posts clean records to Tally and returns the voucher number. Moving the checking step to a smaller model cut the cost per invoice from 0.40 rupees to 0.06.
Start with one job only. Fill in the three lines. Keep the bot on that job.
The one job this bot does: ______________________________
What it may do on its own: ______________________________
What it must ask me first: ______________________________
Two lines from this case, side by side. No new figures.
A small chartered-accountancy firm in Gurgaon took in more than 2,000 vendor invoices a day in a shared Drive folder. Three junior accountants spent six hours each keying them into Tally. That data entry cost about 180,000 rupees a month in salary. The builder set three Claude Code sub-agents under a small Python orchestrator on a 15-minute cycle. One pulls the vendor name, tax ID, line items and totals from each PDF as structured data. One checks the tax math and matches the vendor master. One posts clean records to Tally and returns the voucher number. Moving the checking step to a smaller model cut the cost per invoice from 0.40 rupees to 0.06.
In week one the system averaged 2,047 invoices a day. It posted 94.3% with no human touch. The other 5.7% went to an exception queue, mostly scanned PDFs with poor OCR. Compute ran about 840 rupees a day across all three agents, against about 180,000 rupees a month of manual effort. The builder puts payback at about two weeks. A later prompt fix for handwritten delivery notes dropped the exception rate to 3.1%. The firm is not named, and the report is the builder's own.
The after line is the source's claim, not our measurement.
smalloffice.space · Get a case like this in your inbox each morning. Subscribe free
Read this case on your phone now, and keep it for later. Type your email. We send this case to that address, and you get the next cases each morning too.