AI SaaS breaks the old seat math: your costs scale with their usage, but your price often does not. The heaviest 5% of accounts routinely drive over half of inference cost. If your pricing has no response to that curve, growth makes you poorer.
The fix is structural, not a price hike: metered tiers, fair-use gates, and overage paths your best customers accept because the value is visible. Design it before the power users arrive, not in a panicked quarter.
The questions that actually unblock pricing
Which workflows concentrate cost, and which customers live in them? What is the cheapest path to serve each tier without degrading the outcome? Where does the contract let you say no — and where should it, before finance has to?