AI subscriptions add up quietly

The individual number is always small. Twenty a month, thirty a month, a hundred for the team plan. Nobody escalates a twenty dollar decision, which is exactly why it is the number these things are priced at.

Then you look at the card statement.

Two things make this worse than normal software creep

First, per seat pricing on tools that spread by word of mouth. Someone tries it, likes it, four colleagues get added, and nobody ever removes the person who left in March.

Second, usage based API billing sitting next to the subscriptions. That one does not even have a fixed number to notice. It is fine for months and then a script loops, or a feature ships and traffic triples, and the bill is four times what it was with no decision behind it.

The audit that takes an hour

List every AI tool, what it costs, how many seats, and when it was last opened. The last column is the one that hurts. In most lists I have seen, somewhere between a quarter and a third of paid seats belong to people who stopped using the thing months ago, or who have another tool that does the same job.

Duplication is the common finding. Two writing assistants. Three transcription services because different people picked different ones. A general assistant that already does what the specialised tool was bought for.

What to do about the API line

Put a hard spend cap on it, not an alert. Alerts arrive after the money is gone and are usually read the next morning. A cap is annoying exactly once, when it stops something, and that is a cheaper way to find out than the invoice.

Also worth knowing your cost per unit of whatever you are producing. Cost per thousand words, per support ticket, per generated image. Without that you cannot tell whether the bill going up is a problem or just proof that people are using something you wanted them to use.

Not an argument for cutting

Some of these tools are worth many times what they cost. That is the reason to know the number, not a reason to skip knowing it. Right now most teams cannot answer what they spend on AI in a month, and it is nearly always more than they guess.