You already know which one it is
There is a spreadsheet in your company that matters more than a system you pay for. One person maintains it. It has a tab nobody else understands. If that person left tomorrow, something would break within a fortnight and it would take a while to work out what.
Everyone reading this knows which spreadsheet it is. That recognition is the interesting part — not that it exists, but that its existence is entirely unremarkable.
It exists because it was right
The instinct is to treat it as a governance failure. It is not. Somebody had a question the systems could not answer and answered it themselves, quickly, at no cost, without a project.
That is good judgement. The spreadsheet is a precise, unflattering record of what your software does not do — usually joining two things that live in different systems. Margin by client. Capacity next quarter. Which retainers are over. Almost every load-bearing spreadsheet is a join somebody performed by hand because nothing else would.
Removing it is the wrong first move
The usual reaction, once it is discovered, is to ban it or to build a replacement. Both tend to fail — the ban because the underlying question does not go away, the replacement because it is built from a description of the spreadsheet rather than the need behind it.
A better first move is to read it. What does it join? What does the maintainer correct by hand every month? Those corrections are a list of defects in your systems, written by somebody who has been living with them.
The honest test
We take this seriously because it is the clearest measure of whether a shared record is real. If four products genuinely write to one record, the joins people do by hand should not be necessary.
So the test we apply to ourselves is not whether customers like the software. It is whether they still keep the spreadsheet. If they do, the record is not as shared as we claim, and the spreadsheet is telling us exactly where.
It usually is. That is not a comfortable thing to publish, but a company that claims otherwise has either solved something remarkable or has not asked.