How to Price an AI Product Without Losing Margin
AI products die from two things: churn and gross margin. Pricing is the only lever that touches both at once. Here's the framework we use with portfolio companies.
Start with cost, not competitors
Model your fully-loaded cost per active user: inference, storage, egress, third-party APIs, human review. Then apply a 70–80% target gross margin. If competitors price below your floor, you don't have a pricing problem — you have a cost problem.
Three pricing models that work in 2026
Seat + credits: predictable base, elastic upside. Fits teams. Highest win rate for horizontal tools.
Outcome-based: pay per completed task/lead/ticket. Fits vertical agents where the outcome is measurable.
Consumption with caps: pay per unit but with hard monthly caps. Removes the 'scary bill' objection.
The pricing test that predicts churn
Ask 20 paying users to rate value 1–10. If your median is under 8, you'll churn — no packaging change fixes it. Fix product first, then re-price.
Key takeaways
- Margin is designed at pricing, not fixed at scale.
- Match model to buyer: seat+credits for teams, outcome for vertical.
- Add usage caps to remove buyer fear.
Frequently asked
Is per-token pricing dead?
For end-users, yes — nobody wants to count tokens. Under the hood you still meter tokens; on the invoice you sell credits or outcomes.