Projects / Fintech
Payments matched to invoices
Fintech Fintech client
✦ Result
94%
of payments matched with no one touching them
An agent that reads every incoming payment, finds the invoice it pays and books it, so finance stops matching by hand.
Before: 3 days to close each month's books
The problem
Two people spent the first days of every month matching bank payments to invoices. References were missing, amounts were split across invoices, and customers paid three invoices in one transfer.
What we built
An agent that reads each payment as it lands, pulls the likely invoices, and decides: one invoice, several, or partial. Clear cases are booked straight away. Anything uncertain lands in a review queue with the agent’s reasoning beside it.
Every decision is logged, so the auditor sees why each payment was matched.
Results
Most payments now book themselves the day they arrive. The team reviews the rest in minutes, and month-end close went from 3 days to 4 hours.
How it was built
- Bank feed API
- Accounting system API
- LLM extraction
- Rules engine
- Review queue
- Audit log
- Postgres
- Evals
- Week 1: sample a month of payments
- Weeks 2–4: matcher, rules and review queue
- Weeks 5–6: shadow mode next to the team
- Week 7: live, with daily reports
- 1 lead engineer
- 1 AI engineer
- Finance lead (client side)
FAQs
What happens when it isn't sure?
It doesn't guess. Low-confidence matches go to a review queue with its best suggestion, and one click books it.
Does it touch money?
No. It only reads payments and books matches. Moving money stays with your team.
How long until it pays for itself?
About 3 months for this client, measured against hours saved.
Got a process like this? 20 minutes, and you'll know if it's worth building.
Book a call