Skip to content
All writing
October 26, 2026· 4 min read

Turning a two-week reconciliation grind into a one-day job

#RAG#AI Agents

Financial reconciliation is the kind of work nobody enjoys: matching records, chasing mismatches, doing it by hand under a deadline. The version I walked into took about two weeks every cycle. We got it down to roughly a day.

The approach

The trick wasn't one giant model doing everything. It was agents that use tools: an LLM that can call the right function, pull the right record, and check itself, with RAG feeding it the actual reference data so it isn't working from memory. Each step is small and checkable instead of one black-box guess.

Keeping it honest

Speed is easy; speed without mistakes is the hard part. In finance a confidently wrong answer is worse than a slow one. So the system was built to stay accurate and leave a clear trail of what it did — the humans went from doing the grind to reviewing the results, which is a much better use of their time.

End result: about 85% faster, with the audit consistency the finance team actually cared about.