Automating reconciliation for a payments provider

The situation

What was going wrong

• 01

  Settlement files from four acquirers arrived in different formats and were matched manually in spreadsheets.

• 02

  Month-end close took eight working days and routinely slipped, delaying merchant payouts.

• 03

  Discrepancies were found weeks late, by which point the underlying cause was hard to trace.

• 04

  The process depended on two people, with no documentation and no cover during absence.

Our approach

What we did, and why

1. 01

   Mapped the real process, not the documented one

   We sat with the finance team for a week and recorded every file, rule, and judgement call actually being applied — including the undocumented exceptions that mattered most.

2. 02

   Built a normalising ingestion layer

   Each acquirer feed is parsed into one canonical transaction model, so downstream matching logic does not need to know where a file came from.

3. 03

   Automated matching with a human exception queue

   Confident matches settle automatically. Anything ambiguous is routed to a review queue with the evidence attached, so a person makes the call in seconds rather than hunting for context.

4. 04

   Made the audit trail a first-class feature

   Every match, override, and adjustment is recorded immutably with the user and reason, which turned the annual audit sample request into an export.

The outcome

What changed

Month-end close moved from eight days to under one. The exception queue surfaced a long-standing fee misconfiguration with one acquirer that had gone unnoticed in the manual process, and correcting it recovered more than the cost of the project. Two finance analysts moved off reconciliation and onto merchant analysis.

8 days → <1

Month-end close time

94%

Transactions auto-matched

100%

Adjustments with audit trail