Bad Debt Agent Takes Over Close Prep for a P&L Number That Lived in a Spreadsheet That Crashes

The Starting Point

The bad debt reserve estimates how much of a company’s receivable it will never collect. It lands on the P&L. At this operator one accountant builds it by hand in Excel on the first day of close, for every property at once, and the rest of the close waits behind it.

July’s export was 27,407 rows. In one recorded session Excel choked on a filter, the accountant had to force-quit, autosave had not fired, and he started the cleanup over from the raw file.

Then, back on the clock, he hit a seventy dollar variance and set it aside to investigate later. Then a retail pivot that was not picking everything up. Set that aside too. Both were the right call under time pressure. Both mean a number moved toward the ledger with its check unfinished.

What We Built

The agent does the cleanup and rebuilds the workbook, then runs ten checks and cannot skip any of them. Where the numbers have to match exactly, they match to the cent or the run stops rather than hand over a figure that looks right. Where a small difference is allowed, it reports the amount and where it came from instead of leaving someone to go hunting. Anything it cannot resolve gets named: a property missing from the schedule, a reserve capped because it came out larger than the money owed. Every close ends with a log of what passed and what needs a person.

The rules are theirs: which charge codes roll to which allowance account, which lease stages trigger a reserve, the ten dollar tolerance, and a retail table their field team refreshes each quarter. No vendor could have supplied those, which is why the same build for a different operator produces a different agent.

Nothing posts without a person approving it. The difference is that the person is now approving a checked number.

10 checks. Completed and logged on every run.

To the cent. Tie-out tolerance, or the run stops.

What We Are Watching

The retail reserve is still human judgment, refreshed quarterly with the field team, and it should stay that way. The open question is where the line sits between a rule worth encoding and a judgment worth protecting.