AI-generated article. This article was researched and drafted using AI tools and published automatically, and its featured image was generated by AI. Facts are drawn from the sources cited in the text.
Most small businesses budget for software the way they budget for rent. One fixed number, multiplied by twelve, done. That habit is quietly coming apart, because outcome-based AI pricing is replacing the per-seat subscription that made those flat numbers possible, and the replacement does not sit still from month to month.
Which is both good news and a genuine headache. Paying only for work that actually gets finished is fairer than paying for nine logins when six people ever sign in. It also turns a predictable line item into a variable one, and variables are harder to live with when there is no finance team watching them for you.
What replaced the seat
The shift is measurable rather than speculative. Gartner expects at least 40% of enterprise SaaS spend to move to usage, agent or outcome-based models by 2030. A Pilot study tracked seat-based pricing falling from 21% to 15% of SaaS companies inside twelve months, while hybrid models, meaning a lower fixed base fee plus variable billing on top, jumped from 27% to 41%. Zendesk started billing per resolved customer interaction instead of per user back in August 2024. Salesforce priced Agentforce at roughly $2 per AI conversation. At its Radiance conference in February 2026, HighRadius dropped per-seat pricing altogether and now takes a share of the savings it produces, with no implementation fee at all.
The logic behind it is blunt. When an agent does the work of five people, nobody needs five logins.
The bill you cannot see
The harder problem is not the headline rate, it is the part of the spend that never shows up on a subscription dashboard. Ramp’s 2026 benchmarks put token usage growth at 1,001% between January 2025 and April 2026, even as the price per token fell. Companies in that data expect consumption to grow around 78% over the next two years while per-token prices drop about 19%. Volume wins that argument, so unoptimized token spend still climbs roughly 44%. Agentic workflows are the reason: they consume five to thirty times more tokens per task than a single chat exchange, because the model is reasoning, calling tools and checking its own work. Uber reportedly burned through its entire 2026 AI coding budget in four months.
Visibility is the weak point. Research cited by PointFive found only 26% of organizations have full real-time visibility into what their AI systems cost to operate, and roughly one in four has delayed or canceled an AI project because of that blindness. For a ten-person company, this is the same problem in miniature, and it compounds quickly once several tools are billing by use. Anyone still working out what AI actually costs a small business is asking exactly the right question, just about a target that keeps moving.
Outcome-based AI pricing is not automatically worse for a small company
This is where the panicked reading gets it wrong. A five-person business was almost always the loser under per-seat pricing, paying for capacity headroom it never touched. Billing tied to resolved tickets, processed invoices or booked meetings can genuinely cost less at that scale, because a small business uses less.
The risk sits somewhere else. It is the absence of a ceiling. A per-seat contract had a natural maximum built into it, and a usage contract does not unless someone negotiates one.
Ask about the cap, not the rate
The most useful question at renewal is not what a resolution costs. It is what happens in the month everything goes wrong: a spike in support volume, an agent looping on a broken integration, a campaign that lands better than expected. Vendors moving to these models generally have an answer, in the form of committed-use discounts, spend caps or alerts. They tend not to volunteer it.
Overruns are now the norm, not the exception
Nobody should feel behind for finding this messy. CFO Dive reported that nearly seven in ten US companies, 68%, had at least some AI initiatives run over budget in the past year, and a third said overruns happened mostly or always. Riseup Labs put the average overrun at 42% above the original budget. Deloitte’s analysis of outcome-based pricing for agentic products is partly about a related accounting problem: when revenue depends on results, both sides have to agree on how results are counted and when.
Data preparation alone often eats 30% to 50% of an AI budget, which is one reason so many of these projects look cheap at the quote stage. It is also a decent argument for starting smaller than feels ambitious, and for being honest about whether building or buying makes sense before any of it gets priced.
One small thing worth doing this week
List every AI and software tool the business pays for. Next to each one, write a single word: fixed, per seat, per use, or per outcome. Most owners cannot answer that from memory for more than half the list, and finding out takes about twenty minutes with a card statement and an inbox search.
That list is the whole foundation. It tells you which bills can surprise you, which cannot, and where a spend cap is worth asking for. It also pairs naturally with getting clear on what an agent is allowed to do on its own, since permission and cost are the same conversation once the meter runs per action. Every little tweak in that direction counts, and starting the list at all puts a business ahead of most.
The uncomfortable question underneath all of this has no clean answer yet. When software charges by the result, who gets to define what counts as a result, and what happens when the buyer and the vendor disagree?
