The recon that runs itself (almost)
200 invoices, one bookkeeper, nine days to close. Now the easy 92% match themselves and only the weird ones reach a human.
The bookkeeper wasn’t slow. The work was just boring — and boring work is where humans burn out and miss things.
The play wasn’t to replace her. It was to filter her queue.
The system reads each invoice, looks for a bank line that matches on amount + date window + vendor name (fuzzy), and:
- Confidence ≥ 92% → auto-match, log it, move on.
- Confidence 60–92% → suggested match, one-click approve.
- Confidence below 60% → drop it in the human queue with the top three guesses attached.
She still owns the books. She just stops doing the parts a junior couldn’t do anyway.
What changed
- Books closed by day 3 instead of day 9.
- Her queue went from 200 items to ~16 weird ones — the kind that need her judgment.
- She found two duplicate vendor entries the system flagged that nobody had noticed in 18 months.
That last one is the quiet win. Filtering noise also surfaces signal.
§ Related play
live RECONCILIATION · SAVES 14h / month
AP Auto-Recon (with a human queue)
Bookkeeper hand-matches 200+ invoices to bank lines every month. Books close on day 9. The work is boring, error-prone, and burns out the person doing it.
