Your AI costs grow with use. Does your price?
Every use of an AI feature costs you money, and some uses cost far more than others. Three checks before you put a price on one.
Abhijit Chatterjee Chattopadhyay

On 13 October, Stripe wrote that "Two-thirds of the companies on the Forbes AI 50 list now use some form of usage-based pricing, reflecting AI's variable costs and unpredictable demand."
That is from Stripe's post on the economics of AI monetization (opens in a new tab), a write-up of its Monetize conference. The leading AI companies have stopped charging a flat fee for work whose cost they cannot predict. If you are adding an AI feature to your own product, the same problem is now yours.
The cost changes with every use
A normal software feature costs you almost nothing each time a customer uses it. An AI feature costs you money every time it runs, and the amount depends on how much work it does.
Stripe's post gives a clear example. Replit used to charge a fixed amount per agent checkpoint when its agents ran for a few minutes. As runs grew longer, some jobs cost Replit many times what the user paid for them. It moved to usage-based billing to close the gap.
A flat price works until your heaviest users find the feature. I covered why cheaper models do not fix this on their own in Cheap tokens, expensive bill.
Three checks before you set a price
- Know your cost per use, including the expensive cases. Not the average. Look at the longest and heaviest runs too, because those are the ones a flat price does not cover.
- Price what the customer values, not what you spend. Clay's head of finance said the company does not want to be in the business of "upcharging and taxing tokens." Customers buy a result, such as a resolved support conversation or a finished report. Charge for that, and treat model costs as an input.
- Give customers control over what they spend. One heading in the post says it plainly: "Predictability is becoming a product feature." Usage alerts, spending limits and a view of what drove the bill make usage pricing something a customer can say yes to.
Do this today
- Pull the cost of your AI feature for its ten heaviest users last month, not just the average.
- Write down what result each of them was actually paying for.
- Check whether a customer can see and limit what they spend before the bill arrives.
Your next step
If you want to know what your AI feature really costs to run before you price it, book a free 30-minute 1:1 assessment. I look at your setup with you and tell you what the work would take.
Drafted with AI help, edited and approved by me.