Every growing organisation runs on spreadsheets, and every growing organisation has someone who wants to get rid of them. Usually that instinct is wrong, and the cost of acting on it is higher than the cost of the spreadsheets.
The problem is not the tool. It is that nobody can reconstruct how a number was produced.
Spreadsheets are genuinely good at things
They are fast. They are flexible. They fit work that has not been fully described yet, which describes most work in a growing organisation. A spreadsheet is often the correct answer for a analysis someone will run four times and then never again.
The trouble starts when a temporary answer becomes permanent infrastructure, which happens roughly eleven months after the spreadsheet is written, usually when the person who wrote it has moved role.
What "undocumented" actually costs
Ask three people what the monthly revenue figure is. If you get three different answers, none of them is necessarily wrong. They are each running a version of the logic that has accumulated in someoneu2019s head.
That is the failure mode. Not wrong data — unrepeatable data. The formula still exists, but the reasoning behind it does not. Why this column was excluded, why that month was adjusted, why the lookup ignores archived records. Those decisions were made once, in a hurry, and never written down.
The practical consequences:
- Nothing can be audited. If a regulator, funder or acquirer asks how a figure was reached, reconstructing the answer takes days and depends on someoneu2019s memory.
- The spreadsheet cannot be handed over. The person who maintains it becomes a bottleneck, and their holiday becomes a risk register entry.
- Automation is impossible. You cannot pipeline a workbook whose logic you cannot describe. Attempts to automate it usually just make the same undocumented assumptions run faster.
Three questions before you replace anything
Is it actually a problem, or does it just feel untidy? A workbook that two people use, that has never produced a disputed number, and whose owner is still in role, is fine. Replace it when it stops being true, not before.
What decision does it feed? If the answer is "the board pack, every month, for a number the finance director is personally accountable for", it deserves real infrastructure. If it feeds a Tuesday stand-up that five people attend, it probably does not.
Can you define the logic before you build? If nobody can state the rules the spreadsheet implements, a rebuild will simply hard-code the same ambiguity into a more expensive place. Document first. Then decide whether a platform is warranted at all.
The migration path that works
Start by writing down what the existing workbook does, in plain language, with the person who maintains it. This takes a day or two and is the single highest-value hour in the whole process.
Then split the workbook into three parts: the raw inputs, the transformation rules, and the output. Most of the value of a data platform is in the middle layer, and most of the risk lives there too.
Build the new thing alongside the old one for at least one full reporting cycle. Compare the outputs. Where they disagree, one of them is wrong, and finding out which is the entire point.
Then keep the spreadsheet. Do not delete it. Let people who are not yet confident in the new system check against it, and let it quietly become redundant rather than forbidden.
The goal is not a spreadsheet-free organisation. It is an organisation where any number can be traced back to its source by someone who was not in the room when it was defined.