Two years ago AI spending in a small company was a rounding error hidden inside somebody’s expenses. It is now a real number, and most teams still have no idea what it is.
The problem is fragmentation
AI costs rarely arrive as one bill. They arrive as a seat upgrade on the CRM, a premium tier on the design tool, a per-message charge on the support desk and two or three individual subscriptions on personal cards. Each looks small. Added together they routinely land between five and twelve per cent of software spend.
Three approaches that work
Consolidate first, then negotiate. Most teams are paying for overlapping capability across three vendors. Cutting to one usually saves more than any discount you could negotiate.
Put a cap on usage-based tools. Anything billed per token or per request should have a hard spending limit configured on day one, not after the first surprising invoice.
Review quarterly, not annually. This category changes fast enough that a twelve-month commitment to a specific tool is a genuine risk. Prices and capabilities have both moved sharply within single quarters.
What to measure
Hours saved is a difficult metric to trust because everyone reports it optimistically. A more honest one: if the tool disappeared tomorrow, how long would it take to get the work done another way? Anything nobody can answer is a candidate for cancellation.
The line item worth creating
Give AI spend its own budget line, even a small one. Not for control — for visibility. You cannot manage a cost you cannot see.
