Find the floor, then price well above it
Start with what one user actually costs you in a month: the AI usage, plus hosting, support, payment fees and a slice of overhead. Divide that by (1 − the margin you want) and you've got your floor. But treat it as exactly that - a floor. Customers buy results, not tokens. If your tool saves someone an hour a week, it's worth far more than it costs you to run, so set the price on the value it delivers and use the cost only to check you're not losing money.
The thing that quietly wrecks AI pricing is uneven usage. A few heavy users can swallow the profit from everyone else. So give each plan a generous allowance, then charge extra above it (around three times your raw cost). Cap the free plan hard, small enough that each free user costs about what you'd happily spend on an ad. Matching models to plans helps too - a cheaper model on lower plans, a premium one higher up - which lines your costs up with your revenue and gives people a real reason to upgrade. And watch the spread of usage, not just the average, because the average hides the customers who hurt. The AI SaaS pricing calculator works all this out for you.
The average can lie to you
Two products can have the same average cost per customer and completely different risk. In one, nearly everyone uses it about the same. In the other, most people barely touch it while a small group runs thousands of expensive jobs. The second one looks profitable right up until those heavy users find it.
So look at a typical user and your heaviest legitimate one. Build plans around a clear allowance, and decide what happens at the edge: charge extra, slow them into a queue, drop to a cheaper model, or nudge them to upgrade. Blocking someone's work with no warning is rarely a good way to keep them.
A free plan needs a job
A free plan is a marketing cost, so decide what it's for: letting a serious buyer test quality, helping a small project grow into a paying one, or just spreading the word. Then give it just enough to reach that moment and no more.
Put a monthly cost ceiling on it, guard against abuse, show its limits clearly, and build a clear path to paying. If your free plan is generous only because a competitor's looks generous, you're copying their headline without knowing their costs.
Price the result, explain the limit
People understand "documents processed", "support chats" or "hours saved" far better than tokens. Use the unit that matches the job, and quietly translate it into cost behind the scenes.
Be straight about limits. "500 reports a month" is easier to plan around than a vague pool of "credits". If usage varies a lot, show what a typical action uses and let people see where they stand before they hit the wall.
A tiered example
A simple setup: a free taster, a starter plan with a modest allowance, a team plan with collaboration and a cheaper overage rate, and a business plan with higher limits and support. The premium model might be optional on the starter plan and included higher up.
The point isn't to force any particular number of tiers - it's to line up "more value to the customer" with "more cost to you". Things like audit logs, team permissions, integrations and support often justify a higher price without adding much AI cost at all.
Review pricing by signup group
Group customers by the month they signed up and the plan they're on, then compare revenue against AI cost, hosting, fees, support time, upgrades and cancellations. A healthy overall margin can easily hide one plan or one type of customer that loses money.
And don't drop your prices every time a provider cuts a rate. That saving can become margin you spend on reliability and building the product. Reprice when the value, the market or your real economics genuinely change.
Questions to answer before you publish a price
- What result does the customer think they're buying?
- What does a normal, a heavy, an abusive and an occasional user cost you?
- Which limit is easy to understand and easy to enforce?
- What happens when someone hits it?
- Does the price still work after support, fees, failed jobs and refunds?
Run your numbers through the pricing calculator, but talk to real customers before you treat the result as final. A spreadsheet can find your floor. It can't tell you what the product is worth.