Every finance team has a number it doesn't put on the website: how many days it takes to close the month. For half of finance teams, it's six business days or more; only 18% manage the three-day close the benchmarks celebrate. And the single biggest time sink is the least glamorous line on the checklist: cash reconciliation, at 20–50 hours per month for many teams.
The reason isn't mysterious. The close is a relay race run across systems that don't talk to each other, with Excel as the baton. 94% of finance teams still lean on spreadsheets for close activities, and half of them name it as a key reason the close runs slow.
Anatomy of the slow close
Watch a close happen and you see the same motion repeated in different costumes:
- Pull a report from the bank portal; pull the ledger from the ERP; match them line by line
- Chase the invoices that landed after cutoff, and the approvals that didn't land at all
- Match AP invoices to POs and receipts; the average AP team's exception rate is around 22%, and every exception is an email thread
- Roll intercompany balances, accruals, and prepaids forward in a workbook whose formulas one person understands
- Paste the final numbers into the reporting pack
Each step is simple. The cost is in the seams: logging into five systems, exporting, reformatting, matching, and re-keying. Ardent Partners' 2025 benchmarks put the average cost of processing one AP invoice at $9.40, against $2.78 for best-in-class teams, with the average invoice taking 9.2 days end to end. Multiply the gap across a year of invoices and the "boring" processes turn out to be the budget.
Automation's unfinished business
Finance has been automating longer than any back office, and the plateau is instructive. Gartner finds 59% of finance leaders using AI in 2025, barely up from 58% the year before, and only 28% of teams with AI investments reporting measurable financial impact. Most orgs have data capture; 73% still haven't fully automated their core AP workflow.
The stall has a shape. The clean, structured middle of the process got automated years ago. What's left is the ragged edge: the bank whose portal has no export for that account, the subsidiary on the old ERP, the vendor who invoices from a template that breaks the OCR, the reconciliation that requires knowing that this wire always settles under a different reference. That edge is exactly where scripted automation gives up, and where teams that do automate reconciliation properly report ~95% fewer errors and close times cut in half.
The last mile of the close isn't a data problem. It's a screens-and-judgment problem, and it's been waiting for automation that can handle both.
Closing the seams, not just the books
This is the work AltOps points agents at. Not the judgment; the seams. A senior accountant records one real pass at the bank rec, narrating the folklore as they go: this account's timing differences, that counterparty's reference quirks, when a variance is worth a note versus an investigation. The agent then runs the pull-match-flag loop across the same portals and the same ERP, on its own computer, every day if you like, so reconciliation stops being a month-end mountain and becomes a daily molehill.
The controller still controls. Every match above a threshold, every unusual variance, every judgment call arrives as a reviewed queue instead of a blank workbook. The six days don't compress because anyone worked faster; they compress because the copying, matching, and chasing that filled them stopped being human work at all.
Written by Madhavam Shahi, teaching agents to run the back office at AltOps.
