How AI invoice processing works in Stockpot
5 min read
Every restaurant runs on invoices, and almost every restaurant processes them the same way: a stack builds up by the office keyboard, someone types the totals into a spreadsheet at the end of the week, and the line-level detail (the actual prices you paid for actual ingredients) never gets captured at all. That detail is exactly where margins are won and lost.
Stockpot's invoice pipeline was built to capture all of it with as close to zero typing as possible. Here's what actually happens.
1. Drop the files, as many as you like
You can photograph an invoice on your phone or drag a folder of PDFs onto the Invoices page. Each file is saved immediately as a draft: a safe, inert record that doesn't touch your stock or costs yet. Because extraction runs on the server, you can drop twenty invoices and close the laptop; they'll be read in the background.
2. AI reads every line
The extraction model reads the supplier name, invoice number, date, and every line item: description, quantity, unit, unit price, VAT, and totals. It handles photos of crumpled thermal paper as well as clean PDFs. Each extracted line is then matched against your existing ingredient list using fuzzy matching, so "San Marzano Peeled Tomatoes 6×2.5kg" finds your "Tomatoes, Canned" ingredient even though the words barely overlap.
3. You review, then approve
Nothing changes in your data until a person approves the draft. The review screen shows the original document side-by-side with the extracted lines, so checking takes seconds: confirm the supplier, glance down the matches, fix anything the AI got wrong, approve. If you dropped a batch, "Approve & next" walks you through the queue.
4. Approval cascades through your whole system
This is the part a spreadsheet can never do. One approval:
- Creates the supplier if it's new, and any ingredients you've never bought before
- Adds the purchased quantities to stock, converting pack sizes into your stock units
- Records every price into per-product price history, so trends build automatically
- Updates ingredient costs, which flow straight into your recipe costings
- Fires a price-spike alert if any line jumped past your configured threshold
The result: the moment an invoice is approved, your recipe costs reflect today's prices, your stock reflects the delivery, and any nasty price movement is flagged before it quietly compounds across a month of orders.
What this replaces
For a restaurant processing 15-30 invoices a week, line-level manual entry is hours of admin, which is why almost nobody does it, and why most operators only discover a price increase when the month's food cost percentage comes in high. Automating the capture isn't about saving typing time; it's about making line-level cost intelligence possible at all.
Stockpot's free trial includes AI invoice extraction: try it with a real invoice and see your first price history build itself.