The 2027 AI budget is about inference, not licences: caps, contingency and a named owner

The 2027 AI budget conversation will not be about licences. It will be about who pays for every question your people ask.

Michael Domanic, Section's Head of AI, has published a budget framework that is worth reading even if you discount it as a vendor view. For a 1,000-person company going through a real transformation, he estimates about US$6.6 million a year, roughly what 60 additional employees would cost. Licences are only about 7% of that. Inference is about 80%. Enablement, the coaching, champions programs and change management that turn spend into value, is the last 13%.

Two parts are uncomfortable. His biggest funding lever is backfills: with 12% attrition, a 1,000-person company loses about 120 people a year, and closing out a portion of those backfills, probably 25%, frees up about US$3.25 million. And he warns that if you skip enablement, most of your workforce will still be using AI to write better emails at the end of 2027, with inference charged on every one.

The underlying shift is that AI has stopped behaving like SaaS. His advice is to set inference targets and enforce them with token caps, and to build in 25 to 30% contingency, because inference grows with the usage you are trying to increase. In his words, every bit of success shows up as budget variance.

That needs caps, contingency and a named owner, not a procurement line. And it needs an answer to the question a CFO will ask anyway: what changed in the work, and how will you know? He suggests committing now to quarterly reporting on AI value, including what is not working.

A vendor wrote this, so treat the figures as one model, not a benchmark.

Source: Michael Domanic, Section, "20 questions about 2027 AI budget planning" (3 October 2026), https://lnkd.in/gPCbCBwV

#AIStrategy #CFO #Transformation