The plan price was set before anyone knew
You launched with a simple monthly price per seat or per account, like most SaaS products. The AI features were the reason people signed up, and they use them. Some use them lightly. A few use them all day, run bulk jobs, upload enormous documents, or have wired your product into their own automation so it calls the AI far more than a person ever would.
The model provider's invoice arrives as one total. You can see it rising, faster than revenue some months, but you cannot say which customers are behind it. Your gut says a handful of accounts are costing more than they pay. You have no way to prove it, so you cannot decide what to do about it.
Why flat plans and AI features clash
Traditional SaaS pricing assumes a heavy user costs you roughly the same as a light one, because serving another page is close to free. AI features break that assumption. Every request has a direct cost, and it grows with how much a customer sends and how often.
- Plans were priced on competitor prices or gut feel, before usage data existed.
- Model calls are not recorded against the account that made them.
- There are no limits, so one account can consume without any ceiling.
- Longer inputs and bigger outputs cost more, and heavy users tend to send both.
- Customers who automate against your API or run bulk jobs use far more than any person clicking.
The founder's question is not really technical. It is whether the pricing model still makes sense. But that question cannot be answered without the numbers, and the numbers need building.
What not knowing is doing to the business
| Blind spot | What it leads to |
|---|---|
| Cost per account unknown | Unprofitable customers look like your best ones |
| No usage ceiling | One account can move your monthly bill on its own |
| Plans priced by guesswork | Light users overpay, heavy users underpay |
| No usage data at renewal | Nothing to base a price change or upsell on |
| Investor questions | Gross margin figures you cannot stand behind |
There is a trap in the obvious fix. Raising prices for everyone punishes the light users who were paying their way, and they are the ones most likely to leave. Without per-account data, you cannot target the change where it belongs.
How we measure AI usage and price for it
- Route every model call through one place. A thin layer in your code that all AI requests pass through, whichever provider they go to, such as OpenAI or Anthropic Claude.
- Record usage against the account. Each call is logged with the account, user, feature, model and the tokens or units it used, so cost can be worked out from the provider's own rates.
- Show usage and cost per account. A dashboard that lists accounts by AI usage next to what they pay, so the heavy users and the unprofitable plans are visible for the first time.
- Reduce cost where it is waste. Caching repeated work, trimming oversized inputs and using a cheaper model for simpler tasks, so heavy use costs less before any price changes.
- Add limits that fit your product. Monthly allowances per plan, soft limits that warn before hard ones, and rate limits on automated use, set as configuration rather than code.
- Build the pricing mechanics. Usage tiers, credit packs, an add-on for heavy AI use or metered billing through Stripe, depending on what suits your customers.
- Tell customers where they stand. An in-app usage meter and warnings as they approach a limit, so nobody hits a wall without notice.
We give you the data and the mechanisms. Which pricing model to choose is your decision, and we will lay out the trade-offs of each option with your own usage figures in front of you.
Pricing on evidence
You can see, account by account, what AI usage costs you against what each customer pays. The accounts that were quietly costing too much are known, and you can talk to them with facts rather than a blanket price rise. New plans are designed around real usage patterns. Limits protect you from a single account running up the bill, and customers can see their own usage, which makes an upgrade conversation straightforward.
Margin figures for the board or investors come from records of actual calls, not an estimate spread evenly across customers.
Could this be your pricing problem?
- Your plans are flat and include unlimited or loosely limited AI features.
- The model provider's bill is growing faster than revenue.
- You cannot say which accounts account for most AI usage.
- Some customers automate against your product or run large batch jobs.
- You are considering a price rise but do not know who it should apply to.