Marina is the head bookkeeper at a small furniture manufacturer. The company has 23 employees, but all incoming bookkeeping runs through her: from supplier invoices to payroll. Every month, 40 to 70 invoices land in her inbox: boards, fittings, varnish, packaging, logistics. Some arrive as PDFs, some as scans, some as a photo of a delivery note taken on a supplier rep’s phone.
Six months ago, we connected an AI agent to that workflow. Here’s what came of it — no spin, just actual numbers.
How it worked before the AI agent
Every invoice had to be opened, read, and manually entered into the accounting system: supplier, invoice number and date, amount excluding VAT, VAT, total, due date. Then it had to be checked against the purchase order — were those boards actually ordered, and in that quantity? Marina tracked due dates separately in a spreadsheet and set herself reminders on her phone.
- Entering and checking invoices took 8–10 hours a week — more than a quarter of the work week.
- Twice a year, Marina missed a payment deadline (an invoice got buried among other emails) and the company paid a late fee.
- Once, a supplier accidentally sent the same invoice twice, and it got paid twice — the error only surfaced during the quarterly reconciliation.
What we set up
The agent was connected to the same inbox suppliers send invoices to. It recognizes the attachment (PDF, scan, photo), pulls out the supplier, invoice number and date, amounts and VAT, and the due date — then checks it against the database of open purchase orders. If everything matches, Marina gets a short card in Telegram and confirms it with one tap.
🔔 New invoice received
Supplier: Oakwood Supply Co.
Invoice #2214 dated Oct 9, 2026
Amount: $2,430 incl. tax (tax — $389)
Due date: Oct 19, 2026
Linked to Order #118 (16mm particleboard, 40 sheets)✅ Accept · 💬 I have a question · ❌ Cancel
After “Accept,” the agent enters the invoice into the accounting system itself, attaches the scanned original, and sets a reminder 3 days before the due date. Marina just reviews and confirms — no more typing numbers by hand.
How the agent catches duplicates and errors
Before adding an invoice to the system, the agent checks it against ones already logged: same supplier, same amount, same number within a short window — a reason to flag it.
🔁 This looks like a duplicate
Invoice #2214 from Oakwood Supply Co. was already uploaded on October 9. Is this a new invoice or a resend?Open original · This is a new invoice
Over two months, the agent caught three such repeats — all three turned out to be the same invoice accidentally resent by the supplier.
Reminders that actually work
⏰ Due in 3 days
Invoice #2214, Oakwood Supply Co., $2,430. Pay now?Paid · Remind me tomorrow
Not a single payment deadline has been missed in these six months.
What it delivered over six months
- Time spent on invoice processing: down from 8–10 hours a week to 2–3 — just review and the final call left.
- 0 missed payment deadlines, versus 2 the year before.
- 0 duplicate payments — the agent caught three potential duplicates before the money went out.
- Marina uses the freed-up time for what she never had time for before: analyzing supplier spending and prepping for quarterly reporting.
What the AI agent doesn’t do
This is worth saying honestly too. The agent doesn’t make payment decisions without Marina — it only proposes and waits for confirmation. It doesn’t understand tax nuance and doesn’t replace a conversation with an auditor. And if a scan is genuinely poor quality — a dark photo, blurred text — the agent honestly says “can’t read this” instead of guessing the numbers. In those cases, the invoice still gets entered by hand, same as before.
Is this a fit for your company
If fewer than 15–20 invoices pass through your books each month, the effect will likely be modest. But at 40 invoices or more, where entering and checking them eats up several hours a week, that’s exactly the kind of routine where an AI agent pays off fast. Demand for bookkeepers who can work with these tools is already showing up in job listings — we wrote about that in “Companies Are Already Hiring Dedicated AI Engineers”. And if you’d like to see another example like this one, here’s the case study of a shop owner whose AI agent freed up 10 hours a week for other work.
Want to find out if something like this would work in your bookkeeping? Message me on Telegram or WhatsApp — I’ll take a look at your actual invoice flow and tell you honestly whether it’d be worth it for your case.
Leave a Reply