The pile of supplier invoices is one of the most universal processes there is: they arrive by email, as PDFs, each supplier with its own layout, and someone copies values into the accounting software. It is also one of the processes where AI proves value fastest — as long as the design respects one rule: the accountant approves, always.
Why this was hard before AI
Classic OCR read characters, not invoices. A new supplier template broke everything; an inline discount or mixed VAT required hand-made rules nobody wanted to maintain.
Current models read an invoice like a human: they find the total wherever it is, understand that "shipping" is a cost and tell an invoice from a credit note. The template stopped being the problem.
The loop with AI
The invoice hits the inbox → AI extracts supplier, lines, VAT and totals → classifies against the chart of accounts and that supplier’s history → prepares the entry → the accountant reviews and approves. A repeat invoice with the usual values takes seconds to approve; an anomaly stands out because it arrives highlighted.
The system learns from corrections: reclassify one item and the next similar invoice arrives proposed with the right account.
What can go wrong (and how to design for it)
Models make mistakes. That is why an entry never books itself: it arrives prepared. Totals are validated arithmetically — if the lines don’t add up to the total, the case is flagged. New supplier, out-of-pattern value or a tax ID that doesn’t match: mandatory review.
The right question is not "is the AI always right?" — it is "is the error detectable before it has consequences?". In a well-designed loop, it is.
The essentials
| AI reads invoices like a human — templates stopped being the problem | 01 | |
|---|---|---|
| Entries arrive prepared, never book themselves; the accountant always approves | 02 | |
| Arithmetic validation + exception rules make errors detectable | 03 | |
| The system learns from the accountant’s corrections | 04 |
Written by Pedro · Founder
Other notes
Why material loss never shows up where you look for it
In any operation that transforms raw material there is always a gap between what should have come out and what did.
What happens to an email order when AI enters the loop
Almost every company we talk to has a version of this process: an email arrives with an order — sometimes a PDF, someti…
AI in hospitality: the processes that no longer need to be manual
Hospitality runs on attention to the guest — which is exactly why it hurts to watch front-desk teams copy bookings from…
