How to Price an AI SaaS: Usage, Seats, Credits and Hybrids
Price an AI SaaS by anchoring on the value unit customers recognize, then protecting margin against variable model costs. Most teams land on a hybrid: a base subscription with seats or a platform fee, an included allowance of usage or credits, and metered overage for heavy users.
Why is pricing an AI SaaS different from classic SaaS?
Classic SaaS has near-zero marginal cost per user, so a flat seat price is safe. An AI product pays for every request: model tokens, embeddings, vector search, GPU time or third-party APIs. A single power user can cost more than the whole plan brings in.
That shifts the pricing question from “what will they pay” to “what will they pay, and what does serving them cost us”. You need both numbers before you publish a price page.
The second difference is volatility. Model prices change, new models appear, and customers change how they use the product once it works. Your pricing structure has to survive those moves without a forced migration every quarter.
Which pricing models work for AI products?
There are four main families. Each one maps cost to revenue differently, and each one sends a different signal to the buyer about what they are paying for.
| Model | How it bills | Best for | Main trade-off |
|---|---|---|---|
| Per seat | Fixed fee per user per month | Team tools where usage per person is predictable | Heavy users erode margin; needs fair-use limits |
| Usage-based | Per request, token, minute, document or task | APIs and developer products with measurable units | Unpredictable bills scare finance teams |
| Credits | Prepaid bundle spent across actions | Products with several AI actions of different cost | Credits feel abstract unless the exchange rate is clear |
| Outcome-based | Per resolved ticket, qualified lead or finished report | Agents replacing a clearly measured task | Hard to define and audit the outcome |
| Hybrid | Base plan plus included allowance plus overage | Most B2B AI SaaS | More complex price page and billing logic |
How do you pick the value metric?
The value metric is the unit that grows when the customer gets more value. For a transcription tool it is audio minutes; for a support agent it might be resolved conversations; for a writing assistant it is often seats.
A good value metric is easy to understand, easy to predict, correlated with your cost and hard to game. You rarely get all four, so rank them. Buyer comprehension usually beats perfect cost correlation.
Avoid billing directly in tokens for non-technical buyers. Tokens are your cost unit, not their value unit. Translate them into documents, pages, messages or tasks.
Credits let you price actions of very different cost with one currency. A quick summary might cost one credit and a long research task twenty. You can change the exchange rate internally when model prices drop.
The risk is opacity. If users cannot predict how many credits a task will burn, they stop using the product to save credits. Show the credit cost before an action runs and show the balance everywhere.
Pure usage billing is more transparent for developers, who are used to per-request APIs. Pair it with spending caps and alerts, because a surprise invoice is the fastest way to lose a customer.
How do you set prices step by step?
- Measure cost per unit: log tokens, model, retries and tool calls per action in your own product, then compute average and 90th percentile cost.
- Estimate usage per customer segment: light, typical and heavy. Use real beta data, not guesses, as soon as you have it.
- Choose a target gross margin and work backwards to a floor price for each plan’s included allowance.
- Research the alternative the customer is replacing: a human hour, an agency invoice or an existing tool. That sets the ceiling.
- Place plans between floor and ceiling, with the included allowance sized so typical users rarely hit overage.
- Add hard or soft limits on every plan, including the top one, so no account can run unbounded cost.
- Publish, watch conversion and margin by plan for a few billing cycles, then adjust allowances before headline prices.
How do you protect margins when model costs change?
Treat model choice as part of pricing. Route simple requests to smaller or cheaper models, cache repeated prompts and results, and cap context size. Each of these changes cost per unit without touching the price page.
Keep a per-plan cost dashboard. If one plan’s gross margin drops, you can tighten its allowance, change its default model or introduce an add-on, rather than raising everyone’s price.
Consider bring-your-own-key for technical customers. They pay the model provider directly and you charge for the software, which removes most variable cost from your side.
Common mistakes when pricing an AI SaaS
- Unlimited plans with no fair-use clause: one automated script can wipe out a month of margin.
- Pricing from competitor pages alone without knowing your own cost per unit.
- Charging per token to buyers who think in documents or tickets.
- Too many plans at launch; three is usually enough to learn from.
- No spending caps on usage plans, which turns every invoice into a support ticket.
- Changing headline prices often instead of adjusting allowances and limits.
- Forgetting that free trials and free tiers also consume model credits, so they need their own caps.
Where does open source fit into pricing?
If your product is built on open-source components, your cost stack is mostly hosting and model usage rather than licence fees, which gives you more room on price. It also means competitors can assemble a similar stack, so the price has to reflect workflow, data and support rather than the model call alone.
RepoLoot’s catalog tags open-source projects by licence and difficulty, which helps when you estimate what it would cost a customer to build the same thing themselves. That build-it-yourself cost is a useful sanity check on your ceiling price.
What should each plan include?
A three-plan structure is easy to read and gives you room to learn. The entry plan wins individuals and small teams, the middle plan is where most revenue should land, and the top plan exists for organizations with governance needs.
Differentiate plans by allowance, collaboration and control, not by removing the core AI capability. If the cheapest plan cannot show the product’s main value, trials will not convert.
Keep one lever for enterprise deals outside the price page: custom limits, dedicated capacity, data retention terms or a private deployment. These are negotiated, not listed.
| Plan element | Entry plan | Team plan | Enterprise plan |
|---|---|---|---|
| Included usage | Small allowance with hard cap | Larger pooled allowance | Negotiated volume |
| Overage | Upgrade prompt instead of overage | Metered overage with alerts | Committed spend or custom rate |
| Models | Default model only | Choice of models | Custom routing or bring your own key |
| Collaboration | Single user or small team | Shared workspaces, roles | SSO, SCIM, audit logs |
| Support | Docs and community | Email support | SLA and named contact |
How do you communicate AI pricing to buyers?
Translate every limit into something the buyer already understands. “Around 200 long documents per month” is clearer than a number of credits, even if credits are what you meter internally.
Show a usage meter inside the product and send alerts at sensible thresholds, such as halfway and near the limit. Surprises damage trust more than the price itself.
Publish a short fair-use policy that explains what happens at the limit: slower responses, a smaller model, a pause or an overage charge. Buyers accept limits they were told about in advance.
For annual contracts, offer a true-up clause instead of hard cutoffs. The customer commits to a volume, and you reconcile actual usage at renewal, which keeps procurement simple on both sides.
Frequently asked questions
- Should an early AI startup use usage-based pricing?
- Usage-based pricing suits developer-facing products with clear units such as API calls or minutes. For business buyers, a subscription with an included allowance is usually easier to sell, because budgets are approved annually and finance teams dislike unpredictable monthly bills.
- How much margin should an AI SaaS target?
- There is no universal number. Start from your actual cost per unit and decide what margin funds support, sales and development. Many AI products accept lower margins than classic SaaS early on, then improve them through routing, caching and model changes.
- Is outcome-based pricing realistic for a small team?
- It can work when the outcome is objective and logged by your system, like a resolved ticket or a processed invoice. It gets difficult when the customer disputes what counts as success, so define the outcome precisely in the contract before launching it.
- How often should I change my AI SaaS pricing?
- Review pricing data every quarter, but change headline prices rarely. Adjust allowances, limits, default models and add-ons first. When you do change prices, grandfather existing customers for a period and explain the reason clearly to avoid churn.