The challenge
Every month, the finance team manually checked supplier statements against their ledger, line by line. The work was slow, repetitive and prone to error, in exactly the area where accuracy matters most.
Our approach
The key design decision was how much the agent should decide on its own. Fully automating everything would save the most time, but in finance a confident wrong answer is worse than a slow one. We agreed with the finance team that the agent would resolve clear matches automatically and send anything uncertain to a person.
The solution
- OCR using the Mistral AI API to extract line items from incoming statements
- An n8n agent that matches each line against the ledger and applies agreed rules
- Automatic resolution of high-confidence matches
- Discrepancies posted to Microsoft Teams with the context needed to act on them
The outcome
The finance team no longer checks every line. They review only the exceptions the agent flags, month-end is faster, and errors from manual checking have dropped.