TechCrunch reported on September 3 that 77% of enterprises re-evaluate their AI vendors every six months or on a rolling basis, a pace legacy software contracts never demanded.
That kind of churn is pushing more vendors toward outcome-based AI pricing, where the bill tracks what the software actually delivers instead of how many seats or tokens it burns through. Budgets aren't the problem here.
That same reporting cited Madrona Venture Capital's survey of 150 enterprise IT professionals, in which 74% said they plan to expand AI spending over the next year. Growing budgets and stable revenue for the vendor turn out to be two different things.
Why Buyers Are Pushing for Outcome-Based AI Pricing
Andreessen Horowitz's growth team ran a survey of 50 technical AI buyers in August, and the preference split clearly favored work over raw compute.
Token pricing exposes a buyer to a vendor's own falling infrastructure costs. Pricing tied to recognizable work gives a finance department something it can actually defend at renewal time.
Futurum Research's survey of enterprise software decision makers found the same shift already underway, with per-user licensing losing ground fast to models tied to usage or results.
What Each AI Pricing Model Actually Buys
| Pricing model | What it charges for | What buyers said |
|---|---|---|
| Token or usage | Raw compute consumed | 14 of 50 AI buyers preferred it, per Andreessen Horowitz |
| Credits for recognizable work | Completed units of work, not raw compute | 27 of 50 AI buyers preferred it, per Andreessen Horowitz |
| Per-user license | Number of seats | Fewer than 1 in 5 buyers still prefer it, per Futurum Research |
| Outcome-based | Results actually delivered | 27% of buyers prefer it, per Futurum Research |
The ROI Problem Behind the Churn
Part of the reason buyers keep checking back traces to a figure the same TechCrunch piece raised: an MIT analysis found that 95% of enterprise AI pilots delivered no measurable financial return.
When most pilots don't pay off, a finance team has every reason to put even the ones that do work back under review every few months. Locking into a multi-year deal in a category with that failure rate is a hard sell internally, whatever a vendor's own numbers show.
How to Price an AI Feature So It Survives Re-Evaluation
None of this means AI features can't earn steady revenue. It means the pricing has to survive a buyer who checks in every six months instead of signing once and moving on.
Start by pricing at the highest layer of value you can actually measure, rather than defaulting to tokens because they're the easiest thing to track. Inside an app or SaaS product, that's usually a finished unit of work, not raw compute.
Build the instrumentation to prove that unit's value into the feature itself, not as something bolted on once a renewal gets shaky. This is also where AI automation work earns its keep beyond the sales pitch.
An automation that visibly clears a backlog or cuts a response time gives a finance team a real number to defend, instead of a vague productivity claim. That proof needs to exist before the renewal conversation starts, not during it.
That's the same thinking behind a SaaS marketing plan built around retention rather than new logos alone.
Per-seat pricing isn't disappearing overnight. But a vendor still selling AI purely by seat count is competing for a shrinking slice of the market, against buyers who have already told researchers what they would rather pay for instead.
Cover photo by Matheus Bertelli on Pexels
Sources
- Startup ARR Is Less Secure Than Ever, New Research Shows — TechCrunch
- You Are Not a Model: Don't Price Per Token — Andreessen Horowitz