B2B Pricing Rebuild: Usage, Seat, or Outcome in the AI Era

The first commercial pricing experiments for enterprise AI in 2023 and 2024 mostly took the shape of a per-seat add-on, priced at a modest premium over the underlying software subscription. That structure was operationally convenient for both vendor and buyer, and it produced early revenue that helped fund the model bills. It has also, in the last two renewal cycles, begun to fall apart, and the way it is falling apart tells software leaders something specific about what B2B pricing looks like in the AI era.

The operational fact that broke seat pricing is that AI usage does not distribute like seat usage. In most enterprise deployments, a small number of power users produce the majority of tokens consumed, and a larger population uses the tool occasionally or not at all. Vendors priced against seats, and buyers, once they had usage data, correctly observed that they were paying for a large number of quiet accounts. The renewal negotiation that resulted has, in practice, produced three outcomes, none of them the original seat premium.

The first outcome is pure usage pricing, typically token- or call-based, sometimes with a committed floor. This works when the buyer is technical enough to model consumption and disciplined enough to govern it, which is a smaller subset of enterprise buyers than vendors have assumed. For a CFO buying on behalf of a business unit, a usage meter without a ceiling reads as an uncapped budget line and is rejected on that basis.

The second outcome is a hybrid that combines a smaller platform fee with a usage component metered against a specific outcome, most commonly tickets resolved, contracts reviewed, or drafts produced. This structure has worked well where the outcome is unambiguous and the vendor can price against a defensible reference cost, and poorly where the outcome is fuzzy or where the buyer distrusts the meter. The narrower the workflow, the more this model succeeds.

The third outcome, and the most talked about but least implemented, is genuine outcome pricing, where the vendor is paid a share of the customer value delivered. In practice, this appears principally in categories where the value is measurable in cash and where the buyer accepts that measurement, which today means contact-center deflection, revenue-cycle recovery in healthcare, and a specific set of sales-productivity plays. Every other outcome-pricing proposal we have seen in 2026 collapses in negotiation because neither side agrees on the counterfactual.

For a software CEO or CFO thinking about the next pricing rebuild, the practical guidance is honest. Do not try to price AI as a horizontal add-on. Segment the AI features by workflow, price the ones with measurable outcomes on the outcome, price the ones with clear technical consumption on usage with a committed floor, and price the ambient assistants inside the platform on the platform. A single AI pricing model across a product portfolio is almost always a compromise that leaves money on the table on the strong workflows and creates renewal risk on the weak ones.

The useful question for a board of a software company is whether the current AI monetization line, whatever it is called in the plan, distinguishes between the workflows the customer would pay a premium to keep and the ones the customer would drop tomorrow if they had to choose. If the model treats all AI revenue as one bucket, the renewal risk is being under-reported.