Open your last three invoices from any software you pay for monthly. Somewhere on at least one of them is a line for AI: an assistant, credits, a copilot, an add-on seat. It appeared at some point in the last two years, often as a free trial that quietly became a charge, and almost nobody audits it.
Here is a way to decide in ten minutes per tool whether it earns its place.
Ask what it replaced. Not what it does. What it replaced. If the answer is a task you used to do and now do not, keep it. If the answer is a task you still do, but faster sometimes, note the number of times you used it last month. Most tools can tell you. A figure under about five uses is a subscription to the idea of a tool.
Check whether it is sold per seat or per use. A per-seat AI add-on on a team of eight charges for eight people whether one of them uses it or eight do. Usage pricing can be cheaper and can also run away from you, which is a different problem but a visible one.
Find out whether the base plan already includes a worse version of the same thing. This is common. The summarizer in the add-on and the summarizer on the plan you already pay for are often the same feature with a different cap.
Try one month without it. This is the only test that gives a real answer. Cancel, work the month, and see whether anything hurts. Add-ons are nearly always reinstatable at the same price, and the ones that are not will tell you loudly before you cancel.
Watch for the second charge. A few tools now bill the add-on and then meter the usage inside it. Your invoice says one line. Your actual exposure is that line plus overage. If you cannot find your current usage against the cap in under a minute, that is worth knowing on its own.
The point is not that AI features are overpriced. Some are a bargain. It is that they were added to your bill in a period when nobody was reading the bill closely, and the ones you never use cost exactly as much as the ones you live in.
