01The problem
An online shop that sells on four channels has its numbers scattered across three dashboards, and every morning someone copied them by hand into a sheet. In the starting case the total row added up only one of the two revenue columns: read on its own, each day looked right; the month-end totals did not. Rebuilt from the raw data, the same period changed sign.
02How it is used
- You ask for the update in one sentence, for example “update the cash flow”.
- It finds the missing days, up to yesterday, and takes the data directly from the platforms.
- It checks before writing. If the numbers don't add up, the update stops and flags the problem instead of filling the cells.
- The sheet stays as it was: same columns, same formulas, the same reading the client is used to.
03How it is built
- Sourcesthe raw data from the sales and advertising platforms: revenue, orders, spend
- Writingonly the cells for the missing days; it doesn't overwrite what it finds and doesn't touch the formulas
- Checksthe totals are reconciled before every row; the last day written is always yesterday
- Assistantinterprets the request and coordinates the work; the numbers are calculated by code
04Results
The rebuild brought to light over a third of the revenue that the sheet had never added up, over eight months checked against the source. Since then no number has been copied by hand.