Three systems, three answers, one number

The mess

Three systems, each correct by its own logic, each producing a different revenue figure for the same month. Finance counted on invoice date. The warehouse counted on dispatch. Sales counted on the day the order was signed. Nobody was wrong, and nobody could agree.

The monthly review had quietly reorganised itself around this. The first half was spent establishing whose number to use. By the time that was settled, the meeting was nearly over, and the actual decisions got the leftovers.

The method

The first week was not technical. It was writing down what “revenue in June” was supposed to mean, getting three department heads to agree on that sentence, and then getting it signed off. Everything after that was mechanical.

Once the definition existed, the pipeline could be built to enforce it: ingest all three sources, map each to the agreed definition, and reject anything that could not be mapped rather than quietly guessing. The rejections were the valuable part. They surfaced a category of orders that had been double-counted for years.

Validation went in at the same time. Records arriving with the wrong encoding, missing a customer key, or carrying an empty date stopped being silently coerced and started failing loudly, with the offending row named.

The dashboard

One report. The agreed revenue figure at the top, the three source figures underneath it, and an explicit reconciliation line showing the difference and its cause. The reconciliation line is the part clients keep mentioning afterwards — not because it is clever, but because when the numbers do diverge, nobody has to go hunting.

It is readable on a phone, because the finance director reads it on a train.