Month-end close went from a week and a half to a couple of days.

Their accounts team was three people, none of whom had been hired to type invoices all day. More than a thousand invoices a month came through a shared inbox, got keyed in by hand, and a meaningful share had a mistake in them. Closing the books took a week and a half. Audit prep dragged on for months. The finance chief told us he'd hired a controller specifically to fix this, and after six months in the role she was straight with him: the way it was set up wasn't sustainable, and they needed a different way of working — not another hire. That's where we came in.

1. Six months of invoices, every supplier type, sorted by who they were from, how they got paid, how reliably they matched a purchase order, and how many approvals they needed.
2. Mapping the approvals showed where reality had drifted from the written policy — which suppliers really triggered which approvers, not which ones were supposed to.
3. Early on we sat down with their outside auditor and agreed up front on exactly what evidence would keep the audit clean.

• The controller was in the room every week while we wrote the rules — she signed off on each one before it went live.
• Each person on the accounts team took the review queue for a day during the first stretch so they all understood the exceptions.
• We checked in with the outside auditor each quarter to keep the evidence lined up with what they'd actually want to see.

• An invoice lands in the inbox, the system reads it, checks it against the matching purchase order and receipt, and posts it to their accounting software.
• Anything that looks off is flagged with the context to decide quickly — similar past invoices, the supplier's history, a suggested category — and parked for an approver, who sees a single screen, not a PDF to dig through.
• The repeating month-end entries get prepared and queued; the controller reviews and posts them.
• At any point, the full trail behind any entry can be exported, so the auditor reviews the export instead of sampling people's email.