European software companies are repricing, quietly, because AI features carry a cost per use.

The per-seat subscription is the most successful pricing structure in the history of business software, and it rests on one assumption: an additional user costs almost nothing to serve.

A product that calls a model on every meaningful interaction violates that assumption. The heaviest users become the least profitable ones, and the customers most enthusiastic about the AI feature are the customers eroding the margin fastest.

Three responses in the market

Credit systems, where the subscription includes an allowance and overage is purchased. Transparent to the vendor, disliked by buyers who cannot forecast consumption.

Tiered seats, where AI capability sits in a higher priced seat class. Familiar to buyers, and it caps exposure only if the tier is priced against realistic heavy usage rather than average usage.

Absorbed cost, where the feature is included and the margin is taken on the hope that inference prices keep falling. Common, and a bet rather than a strategy.

The European difference

European vendors interviewed for this piece have moved more slowly toward consumption pricing than their American counterparts, for two reasons that are both commercial rather than cultural.

European enterprise buyers, particularly in the public sector and in regulated industry, procure against fixed annual budgets and penalise unpredictable line items. And multi-year contracts, more common in European enterprise sales, make mid-term repricing awkward.

What the data suggests

Across the vendors who shared figures, the distribution of AI feature usage is severely skewed: a small minority of users generate the large majority of inference cost. That shape argues against absorbed cost and in favour of a tier or an allowance.

It also argues for measuring before pricing. Several vendors repriced on an assumed distribution and discovered their actual one looked nothing like it.