Why I stopped trusting my own Excel reports
The reports were not wrong when I built them. They became wrong later, quietly, in ways the file itself gave no sign of.
How it goes wrong
A spreadsheet is a snapshot of a decision you made about the data, frozen at the moment you made it. The data keeps moving. The decision does not.
- A range that stopped at row 8,400 when the catalogue grew past it
- A lookup silently returning the first of two duplicate SKUs
- A hard-coded exchange rate someone entered during a specific week in 2023
- A filter left applied, so a copied column carried only the visible rows
Every one of these produces a number that looks entirely plausible. That is the actual problem — none of them fail loudly.
A report that breaks is a nuisance. A report that shifts by four percent and keeps rendering is a much worse outcome, because it gets believed.
What helped
Not abandoning Excel. Moving the parts that must be right out of it — extraction into SQL, shaping into Power Query — and leaving the spreadsheet as the presentation layer it is genuinely good at being.
Then a short list of checks that run every refresh: row counts against the source, totals reconciled to the ERP, a null scan on join keys, and a flag on any SKU appearing more than once where it should be unique. Cheap to build, and they catch the failures that would otherwise be discovered in a meeting.