Finding the AI spend outside your API bill
Updated September 1, 2026 · first published September 1, 2026
Most organisations measure AI cost by opening the provider console. That number is real, it is well instrumented, and in a lot of companies it is the smaller half of the total. The other half is on expense reports: individual subscriptions bought by people who needed a tool and had a card.
Why it is invisible to platform teams
It never touches your API keys, your gateway or your telemetry. It arrives as a per-seat charge on a corporate card, coded to software or occasionally to office expenses, in amounts small enough to clear approval automatically. Two hundred people at twenty to thirty dollars a month is a six-figure annual line that no engineering dashboard will ever display.
And this is not a story about rule-breaking. It is what functioning employees do when a tool is useful and procurement takes six weeks.
What it actually costs you
The direct overspend is the least interesting part. Individual plans are priced well above negotiated enterprise rates for the same product, and paying twice for the same vendor — once through a team agreement, once through scattered personal seats — is common. Multiple tools that do the same job proliferate because nobody is comparing them.
The harder costs are elsewhere. Consumer tiers carry different data-retention and training terms than the enterprise agreement your legal team actually reviewed, so business data ends up under terms nobody signed off. There is no SSO, so access does not end when employment does. And you cannot forecast, negotiate volume, or make a build-versus-buy decision about spend you cannot see.
How to find it
Ask finance, not engineering. Search card and expense data for the known vendor names, then for the pattern itself: recurring charges between roughly ten and fifty dollars, monthly, to software-adjacent merchants. Cross-reference SSO and IdP logs for applications nobody formally onboarded.
Then respond by making the sanctioned path easier than the card, not by banning the card. Enforcement without a fast alternative produces the same spend with worse visibility. A team plan that people can join today, with the enterprise terms and SSO attached, converts shadow spend into negotiated spend and usually pays for itself on the seat consolidation alone.
Related
Related
Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →