Why AI Breaks the Classic SaaS Pricing Model
For two decades, SaaS pricing rested on one comfortable fact: serving one more customer cost almost nothing. Once the software was built, an extra user was close to free, so per-seat subscriptions made perfect sense. Charge a flat monthly fee, add seats, watch the margin stay healthy.
AI has broken that assumption. Every query to a model costs real money, and a heavy user can cost you many times what a light user does. If you wrap that in a flat per-seat price, your best customers, the ones who use the feature most, become your least profitable. Roughly 70% of companies shipping AI features now report margin pressure for exactly this reason.
The Shift to Usage-Based
This is why the industry is moving toward usage-based pricing at speed. By 2027, usage-based approaches, whether prepaid credits, post-paid metering, or a blend, are projected to make up around 62% of all AI product pricing. The logic is simple: when your cost scales with usage, your revenue should too.
But pure metering has its own problem. Customers hate unpredictable bills. A finance director who cannot forecast next month's invoice will either cap usage aggressively or churn. So the honest answer is rarely "pure usage."
Buyers vs. Vendors: The Hybrid Compromise
There is a genuine tension here. Buyers prefer postpaid, consumption-based models, they only pay for what they use. Vendors prefer prepaid subscriptions, predictable revenue they can plan around. Both are right from where they sit.
The resolution that most successful AI products are landing on is a hybrid: a base subscription that includes a generous usage allowance, with metered pricing only beyond it. The customer gets a predictable floor and the comfort of "fair use." You get baseline recurring revenue plus protection against the heavy user who would otherwise erode your margin.
A Practical Framework
If you are pricing an AI feature, I would work through these four steps in order:
- Know your unit cost. Before anything else, calculate what a typical action actually costs you in model calls. You cannot price what you cannot measure, and this number changes as models get cheaper.
- Pick a value metric that tracks cost. Choose something the customer understands and that rises roughly with your expense: documents processed, minutes transcribed, actions taken. Avoid metrics that feel arbitrary.
- Set a base plan with a real allowance. Include enough usage that most customers never hit the meter. The allowance is what makes the bill predictable, and predictability is what closes the deal.
- Meter the overage transparently. Show usage in the product, warn before limits, and make the overage rate easy to find. Surprise bills destroy trust faster than any price point.
Communicate the Value, or Nobody Pays
This is the step most teams skip. Buyers are open to paying more for AI, but only when the benefit is tangible. If your AI feature saves someone two hours a week, say so, in their terms, at the point where they feel the pain. A price is a claim about value, and the claim has to be believable before anyone signs.
Get the model right and AI becomes a margin story instead of a margin problem. Get it wrong and your most engaged customers slowly turn into your most expensive ones.
Working out how to price an AI feature without eroding your margins? Let's talk it through. It is one of the most common questions I help teams answer. And if you want a broader look at where your product stands, book a free product scan.