The End of the Seat: What Usage Pricing Does to Your Bill

For two decades, buying software was easy to forecast: a price per person, per month. You knew the year's cost before you signed.

That is being dismantled, and not out of pricing ideology. It is a direct consequence of AI agents. When the work is done by software rather than by people, charging per person stops making sense for the seller — because the customer can shrink the team and consume just as much.

OpenView tracked adoption of usage-based pricing among software companies: 34% in 2020, 45% in 2021, 61% in 2022.¹ And Gartner, in a release dated 1 July 2026, put a number on what is in play: $234 billion of enterprise application software spend at risk from agentic AI, describing the shift as an existential threat to vendors defending seat-based models.²

What "per outcome" means in practice

The clearest verifiable example is Fin, Intercom's support agent: $0.99 per resolution, charged at most once per conversation, with a fifty-resolution monthly minimum.

The promise is appealing — you pay when it works. The small print less so: a "qualification" is billed separately at $9.99; inside Intercom you still need at least one paid seat; and the definition of what counts as a resolution belongs to the vendor.

In June 2026 Salesforce announced it was acquiring the company for roughly $3.6 billion. In the same month it launched its own support agent charging per resolution at about $2 — double the price of the product it was buying. That says something about the maturity of this market: nobody yet knows what an outcome is worth.

The counter-trend almost nobody reports

Here is the information that changes how to read this subject, and which is missing from nearly all coverage of it.

In December 2025, The Register reported the opposite movement: Salesforce, after experimenting with consumption and per-conversation pricing, moved back toward treating seat-based licensing — with adjustments — as the norm for AI agents. The stated reason was simple: customers wanted predictability.³

So "the death of the seat" is a contested thesis, including among those who announced it loudest. Anyone writing about it as settled fact is selling something.

Meanwhile, the price goes up

Whatever the model, aggregate cost has risen. Vertice, which measures software inflation across a base of 16,000 vendors — and sells spend optimisation, which should be said — recorded 8.7% in 2023 and 11.4% in January 2025, against 2.7% average G7 inflation over the same period.⁴

Software spend per employee moved from about $7,900 in 2023 to $9,100 at the end of 2025. And there is the phenomenon Vertice calls shrinkflation: 28% of renewals in the final quarter of 2025 delivered less for the same price.

The figure that matters most for you: companies of 100 to 500 employees spend around $6,700 per employee on software, against $4,700 at large enterprises. Less volume, worse per-head price.

Who wins and who loses

Wins: anyone carrying idle seats, anyone with light or seasonal use. Here usage pricing corrects a genuine unfairness in the old model.

Loses: heavy users. This is the counter-intuitive point — under outcome pricing, the better the tool works, the more you pay. A support team that resolves well pays more than one that resolves badly.

And everyone loses predictability. Recurring revenue becomes an estimate, the annual budget becomes a range, and the month-end bill becomes a surprise.

Six clauses worth having

1. A written definition of the billed unit. Exactly what counts as a resolution, a conversation, an execution. Without it, the metric is defined by whoever is charging.
2. A consumption cap. A ceiling above which nothing is billed without your approval.
3. A commit-to-consume discount. If you can forecast usage, that forecast is worth money.
4. A grandfathering clause. Locks your current terms for a period even if the vendor changes policy — especially relevant after an acquisition.
5. A notice window for price changes. Without it, you find out in the invoice.
6. A re-evaluation clause in any multi-year deal. This market is moving too fast to sign three years without an exit.

The general rule

Usage pricing is not a synonym for cheaper. It is a transfer of risk: away from the vendor, who previously had predictable revenue, and onto you.

Whether that is good or bad depends entirely on your consumption pattern. The calculation that decides it is simple and almost nobody does it: take the last twelve months of actual usage and re-price it under the new model. If you cannot run that calculation because you do not measure usage, that is the problem to solve before negotiating any price.