Traditional SaaS has a beautiful property: near-zero marginal cost per additional user. You write the software once, host it, and each new subscriber is almost pure margin. This is why SaaS multiples are so high. The unit economics get better as you grow.

AI wrappers break this property.

When your product is built on top of a foundation model API, every user action costs money. More usage means more expense, proportionally. The analyst who uses your tool heavily — your best customer, the one most embedded in your workflow — also costs you the most to serve. The unit economics don’t improve at scale; they stay roughly constant, or worsen if power users drive disproportionate usage.

The numbers show up in churn. Products built on this model see 65% churn within 90 days. Users try it, use it a lot in the first few weeks, hit the limits of what it can do, and cancel. The API bills mount during the highest-usage period. The customers most likely to stay are the ones using it least.

The businesses surviving this aren’t solving it by making better wrappers. They’re solving it by using AI internally to deliver services that look like software from the outside. The AI generates the output; a human (or a well-designed process) validates and delivers it. The customer pays for outcomes, not for API access. The margin structure is more like a consulting firm than a SaaS company — but with software-level leverage on the delivery side.

This isn’t a compromise. It’s a different model with different strengths. Outcome-based pricing is stickier than subscription pricing, because the customer pays when they get value rather than monthly regardless of usage. The churn dynamics are inverted.

Building for outcomes instead of access is a design constraint that clarifies a lot of product decisions. What’s the specific outcome the professional is paying for? Can you guarantee it? Can you price for it? Those questions will get you further than optimizing the model selection. +++