The Cancel Test: The Fastest Way to Find Out What an AI Tool Is Worth

There are two ways to find out what a subscription is worth to you. One is to build a careful evaluation with test tasks and timers — the framework I use before paying. The other is faster, free, and works on tools you already pay for: turn it off and see what breaks.

I call it the cancel test. It’s the difference between asking “do I value this?” — a question humans answer with stories — and asking “did I miss it on Tuesday?” — a question that answers itself.

The method

Pick one subscription you’re unsure about. Not the one you use hourly; the one you think you’d defend but can’t remember last using. Every stack has one — usually the second chat assistant or the third writing tool (see What an AI stack actually costs).
Downgrade to the free tier — don’t delete the account. This isn’t about leaving; it’s a measurement. Keep your data, keep your history, drop the paid features. If the tool has no free tier, cancel at the end of the billing period and run the test in the paid time you’ve already bought.
Set a two-week timer and change nothing else on purpose. No replacement tool, no workflow redesign. The point is to observe the hole, not to fill it before you’ve seen its shape.
Log the friction, at the moment it happens. One line, in whatever notes app is already open: date, what you were doing, what you did instead, roughly what it cost you. “Tuesday — wanted better draft of client email — used my other assistant — fine, honestly.” That last clause is the data.

Reading the results

Two weeks later, the log reads in one of three ways:

Empty, or nearly. You just found $240/year. The subscription was a story you were telling yourself — “I’m the kind of person who has the good tools.” Cancel without ceremony. You can always come back; these companies will not stop selling to you.
Full of entries that end in “…and the workaround was fine.” Also a cancellation, and the more interesting one: the tool was real but marginal, and marginal at $20/month loses to free. This is the overlap tax made visible.
Entries with real cost — time, quality, or a small dread of a task the tool used to absorb. Congratulations: resubscribe with evidence. You now know what the tool is actually for, which is worth more than the two weeks cost — it tells you which features justify the price, which tier you need, and what would have to appear elsewhere before you’d cut it.

Why absence measures better than use

While a tool is running, its value and its availability are impossible to tell apart — everything gets routed through it because it’s there, and the routing itself feels like value. Absence unbundles that. The tasks that genuinely needed the tool announce themselves as friction; the ones that merely defaulted to it flow silently somewhere else, and you learn that the somewhere-else was fine.

That’s also why the test beats introspection. Asked directly, you’ll defend the subscription — you chose it, and choices recruit their own lawyers. The log has no client. It just has Tuesday.

The rules that keep the test honest

One tool at a time. Two absences at once blur into general grumpiness and you can’t attribute the friction.
Log at the moment of friction, not at the end. End-of-week memory smooths exactly the spikes you’re trying to measure.
The workaround counting as “fine” is a verdict, not a failure of taste.
Don’t run it during an unusual fortnight — a launch week or a vacation measures the fortnight, not the tool.

The one-page version

Pick the subscription you’d defend but can’t remember using.
Downgrade, don’t delete. Two weeks. Change nothing else.
Log friction the moment it happens, ending each line with what the workaround cost.
Empty log → cancel. “Workaround was fine” → cancel. Real cost → resubscribe with evidence.
One tool at a time, in a normal fortnight.

The cancel test costs nothing, risks nothing, and every outcome is a win: either you keep the money, or you keep the tool and finally know why.