Price the feature, not just the model
List prices tell you cost per token. Unit economics tell you whether seat or usage revenue covers that cost after adoption and heavy users.
Watch AI COGS % of revenue
Many teams target roughly 10–25% of feature revenue for model API spend. If you are above that, route to a cheaper model, raise prices, or reduce tokens per use.
Plan for adoption spikes
Power users can dominate spend. Model uses per adopting user, not just MAU, before you launch unlimited AI in a flat seat plan.
FAQ
Should AI be priced per seat or per use?
Seat pricing is simpler when usage is predictable. Per-use pricing protects margin when a minority of users generate most of the token volume.