BODi

A task that ate ninety minutes a cycle now takes about three. The interesting part was never the speed. It was working out what a human was actually there to decide.

The problem

Recurring invoice and general-ledger review burned analyst time on records that were almost always correct. The cost was never the reviewing. It was that the handful of exceptions worth a decision sat inside a queue you had to read line by line to find.

What I owned

The tooling, not the policy. Thresholds, controls and sign-off stayed with FP&A and Accounting. I built to their definitions and they validated every output before anything was relied on.

Approach

• Defined a genuine exception with the FP&A and Accounting owners before writing anything.
• Rebuilt general-ledger cleansing so records arrived in one shape rather than several.
• Ran automated output against manual review until they agreed, then tightened the thresholds.

Result

Selected recurring tasks moved from roughly 90 minutes to one to three minutes. I sized approximately $784K in annualized operating expense reduction and presented it to the MD of Accounting.

What I would fix

The saving is a run-rate estimate from task frequency and loaded analyst cost, not realized P&L, and I would say so before anyone asked.

The thresholds are still set by hand. They should be learned from which flags accountants actually action. I would have built that feedback loop before optimizing anything else. Writing this up is what led me to an essay about where savings like this actually go, which is nowhere on the balance sheet.