Start from unit economics
Compute your true cost per user: AI tokens + infrastructure + support + payment fees. Then price for a target margin: price = cost ÷ (1 − margin). This is your floor, not your final price.
Then move to value-based pricing
Customers pay for outcomes, not tokens. If your tool saves an hour a week, it's worth far more than its cost. Anchor price to value delivered, and use cost only to ensure you're not underwater.
Protect against heavy users
AI costs are variable, so a few heavy users can erase your margin. Defend with a generous included allowance + metered overage (price overage ~3× raw cost), and a hard-capped free tier sized so each free user's cost is an acceptable marketing expense.
Choose the right model per plan
Offer cheaper models on lower tiers and premium models on higher tiers. This aligns cost with revenue and gives customers a reason to upgrade.
Watch the metrics
Track gross margin per user, cost per action, and the usage distribution (not just the average). The AI SaaS pricing calculator models all of this, including a plan profitability simulator.