Fintech2025Confidential client

Invoice reconciliation agent

An agent that matches invoices against bank transactions every night, flags mismatches and drafts the follow-up emails for a person to approve.

Agentic AI
Intelligent Automation

Start a similar project All work

Client: Confidential client
Industry: Fintech
Year: 2025
Services: Agentic AI · Intelligent Automation

The challenge

Month-end close took days of manual matching across the ERP, bank exports and email.

The finance team matched invoices to bank transactions by hand at the end of every month. Exports came from the ERP and two banks in different formats, partial payments and bundled transfers had to be worked out line by line, and chasing missing payments meant writing emails one at a time. Close dragged on for days and the team had little time left for review.

What we built

An agent with tool access to the ERP and bank feeds, a review queue for edge cases and a full audit log.

A nightly reconciliation agent
It pulls open invoices from the ERP and new transactions from the bank feeds, then matches them on amount, reference, date and payer, including partial and combined payments.
A review queue for edge cases
Anything the agent is unsure about is held for a person, with the candidate matches and the reason it was flagged side by side.
Drafted follow-ups
For overdue or short-paid invoices the agent drafts the reminder email. Nothing is sent until someone approves it.
A full audit log
Every match, change and approval is recorded, so the auditors can see who decided what and why.
Guardrails
The agent can read the bank feeds and can only write to reconciliation fields in the ERP.

Outcome

The team starts each day with most transactions already matched and a short list of exceptions to review, instead of a month-end spreadsheet marathon.