The tool nobody owns
It is not on the systems list, nobody approved it, three people would stop working without it, and it is the best requirements document in the building.
Contents
Every place I have worked has had two or three files that hold it together and appear on no inventory of anything.
A workbook that maps one system’s codes onto another’s, because the two were never going to agree and somebody built the bridge by hand. A tracker that is the only place the current state of a thing is written down. A sheet with a tab per entity that four people open before they open anything official. No owner in the organisational sense, no change control, no backup except that it has been emailed around enough times that a version exists in several inboxes. If you asked the business what it runs on, none of them would be named.
They grow where the official path is slower, not worse
I used to read these as a hygiene problem. Sloppiness, or a symptom of somebody not being trained on the real system.
That is not what selects for them. The official system is frequently better in every dimension I would care about on paper. It is controlled, it is backed up, it has permissions, somebody supports it. It also takes four days and a request form, and the file takes ten minutes and takes them now. Speed against the pace of the actual work is the entire selection pressure, and it is not close, and it does not care about any of the other properties.
Which means the file is not a failure. It is a measurement. It marks the exact place, to within a few hours, where the supported path stopped keeping up with what somebody was being asked to do that day.
The best requirements document nobody commissioned
Once I started reading them that way, they became the first thing I look at.
Think about what one of these files actually is. It is a specification of what people needed and could not get, written by the people who needed it, in their own vocabulary, tested every working day under real conditions, and revised immediately whenever it failed. Nothing formal comes close to that. A requirements exercise costs a fortune and produces a document describing what people say they do. The unofficial file describes what they do, because if it were wrong the work would have stopped.
The columns are the interesting part. Somebody added a column for a reason. The column that is filled in by hand every time is a field the real system does not have. The tab that everyone ignores is a step somebody thought mattered and nobody does. You can read a genuine process off the sheet in an hour, and the sheet cannot lie to you, because it has to work.
The consequence I did not see for years is what this does to replacement projects. A new system arrives, the file is not in scope because it is not a system, and the project goes live and appears to succeed. Then it fails slowly, months later, because everyone quietly kept their copy and the two states drift apart until somebody notices they have been reporting off the old one. It never fails at go-live. It fails when the copy becomes the truth again.
What I did about it, which did not work
At one point I did the responsible thing. I took one of these and formalised it. Proper location, a named owner, a documented change process, review before modification.
It stopped being useful within a couple of months, and it stopped for exactly the reason it had been useful. Its whole value was that any of the four people who used it could change it in ten seconds, at the moment they discovered it was wrong, without asking anybody. Put a change process in front of that and you have built a small slow version of the official system, which is the thing that caused the file to exist in the first place. I had not preserved it. I had converted it into the problem.
I do not have a clean answer to that and I distrust anyone who offers one.
And the risk is not imaginary
The other correction is harder, because it cuts against everything above.
These files genuinely are a single point of failure. One person, one machine, no backup, no documentation, and an assumption embedded in a formula that nobody else has ever read. That is a real exposure and it is not softened by the file being an excellent record of requirements. Both things are true at once and the people who talk about shadow tools tend to hold only one of them.
Some of them are also not requirements at all. A share of these files are one person’s preference that hardened into infrastructure because nobody pushed back, and they encode a step that has not been necessary for years. Telling that apart from a genuine gap requires asking why the file exists, in a way that sounds like an accusation no matter how carefully you phrase it, which is probably why so few people ask.
What I do now is smaller than a method. When I arrive somewhere new I stopped asking what systems they run, because the answer to that is a list and the list is not where the work is. I ask what they open first in the morning.