We Tried to Cut Every Tool for a Month and Learned Which Four We Actually Cannot

Every audit, including the quarterly one I run, produces a list of tools you are confident about and a list you are not, and the second list never resolves, because there is no way to argue yourself into knowing whether something is necessary, which is why leaving a tool you still like takes so long. So in May we ran an experiment: for one week each, we turned off a tool and worked without it, in order of my confidence that we could.

Fourteen tools, fourteen weeks, one at a time so that we could attribute the pain. I predicted the outcome for each in advance and sealed the predictions in a document, which is the only part of this that felt slightly ridiculous and is also the part that made it worth doing. I was wrong six times.

The four we could not cut, in order of how quickly it became obvious: version control hosting, which lasted four hours before we restored it and is not a serious entry on this list. The password manager, which lasted two days — the failure was not security, it was that shared credentials immediately reverted to being pasted into a chat thread, and I could feel the archive of that thread becoming a liability in real time. Our accounting tool, which lasted a week but produced eleven hours of manual reconstruction afterwards. And the design tool, which is the product we sell through and was never seriously in question.

The surprises were in the other direction. I predicted our project tracker would be cut-proof and we lasted the full week comfortably, because with two people a shared list in a document is genuinely sufficient. We did not cut it — the week showed the tool was worth its price for the history and the client visibility, not for the tracking itself — but it demoted it from essential to worthwhile, and that changed how much I was willing to pay at renewal, and whether to take the annual discount at all.

The scheduling tool went the other way. I had it down as a nicety. Without it, one week produced nine extra email round-trips to book four calls, and one double-booking that I had to apologise for. Nine round-trips is roughly forty minutes of my attention broken into pieces across the week, which is worse than forty contiguous minutes. It went from a candidate for cutting to something I would now pay double for.

Three tools we cut permanently as a direct result: a screenshot annotation tool whose job the operating system does now, a second analytics product we had been running alongside our main one for a comparison that ended in 2024, and a file transfer service we used four times a year and can replace with a shared folder. Combined, $47 a month.

The method has one flaw I want to state clearly, because it took me until week nine to see it. A week is long enough to detect friction and too short to detect decay. Turning off backups for a week produces no pain at all and tells you nothing, because the value of a backup is entirely in a low-probability event. Anything whose benefit is insurance cannot be tested this way, and I nearly cut something on the strength of an uneventful week before I noticed the category error.

For anything that is not insurance, though, I have not found a better method, and I would recommend the sealed prediction specifically. Being wrong six times out of fourteen is the actual result of this experiment. It means my confidence about our own stack — a stack of fourteen things, in a company of two, that I personally bought — was barely better than a coin flip. I would assume the same is true of yours until you have checked.