Per-seat pricing assumes the user is doing the work, and the value scales with the number of users. AI changes that assumption. When the product does the work, the value scales with the work done, not the seats logged in. Per-seat pricing for an AI product is a slow leak, both ways: customers feel overcharged for low usage and undercharged for high usage.
There is a darker version of the leak worth naming: an AI product that succeeds at its job often reduces the customer's headcount need in that function. Under per-seat pricing, your product's success shrinks your own contract at renewal. That is not a pricing inefficiency; it is a structural misalignment where your revenue is inversely correlated with your value, and it only gets worse as the AI gets better.
What outcome-based pricing looks like
Price the unit of value the product produces. Resolved support tickets. Drafts generated. Tasks completed. The customer pays for what they got, not for who could have logged in. The unit price reflects the cost of producing the outcome plus the margin the value supports.
Choosing the unit is the design decision that makes or breaks the model, and good units share four properties: the customer can count it independently (no black-box billing), can roughly predict it month to month, cannot easily game it, and recognises it as value rather than activity. "Resolved ticket" passes all four; "API call" fails the value test (activity, not outcome) and "revenue influenced" fails the counting test (attribution wars at every renewal). This is the logic behind pricing FlyWheel in Wheels: one target account through the full workflow is countable, predictable, hard to game, and unmistakably the unit of work a customer wanted done.
What it changes about the relationship
Customers stop treating the product as a fixed cost and start treating it as a variable input. Their adoption curve aligns with their value curve. You stop competing on per-seat parity and start competing on the value of each outcome. Expansion happens organically as they use the product more.
It also changes the internal economics of your own roadmap. Under seats, the incentive is to add surface area that justifies the price. Under outcomes, the incentive is to make each outcome cheaper and better, because margin improvement and customer value point the same direction; see cost-per-resolved-task for the metric that keeps this honest. Sales conversations simplify too: the pitch stops being a features tour and becomes arithmetic the buyer can check ("an outcome costs you X today; we charge a fraction of X").
Where the model breaks
If the unit is unpredictable to the customer, they will hesitate to scale. The fix is a hybrid: a small platform fee for predictability, plus outcome pricing for scaling. The customer gets a known floor and a known cost per unit beyond it.
Two more break-points deserve pre-emption. Quality disputes: if a delivered outcome can be contested ("that ticket was not really resolved"), define the unit with an acceptance rule up front (reopened within N days does not count) rather than litigating monthly. And bill shock: a genuinely successful month should never feel like a penalty, so cap exposure with volume tiers or a soft ceiling with overage packs, priced published, the way Wheels overage works. Predictability is not the enemy of outcome pricing; it is the feature that lets customers say yes to it.