AI SaaS Pricing Models: Seats, Usage, Credits and Outcome-Based Pricing
How to price AI features in SaaS: seats, usage, credits, outcome-based and hybrid models, with real examples, margin risks and the metering they need.
The short answer
AI features change SaaS pricing because every use costs money, and that cost varies by customer. The main options are seats (simple, but heavy users can erode margins), usage-based pricing (fair and scalable, but harder for customers to predict), credits (a middle ground that bundles different AI actions into one currency), outcome-based pricing (aligned with value, but dependent on measuring outcomes precisely), and hybrids of these. Most products land on a hybrid: a subscription with included AI usage, plus credits or overages for heavy use. Whatever you choose, you need accurate metering of your own costs and customers' usage before you set prices.
This guide is for SaaS founders, product and finance leaders. It covers the models, public examples, margin risks and the metering and billing systems behind them. For the pricing page itself, see SaaS pricing page optimisation.
The pricing models compared
| Model | How customers pay | Strengths | Risks |
|---|---|---|---|
| Per seat (AI included) | Fixed price per user | Simple to buy and forecast | Heavy users can cost more than they pay; AI may reduce seat counts over time |
| Per seat + AI add-on | Seat price plus an AI upgrade per user | Separates AI value; easy upsell | Same cost exposure if add-on usage is unlimited |
| Usage-based | Per token, request, document, minute or task | Price tracks cost; scales with value | Bills are harder to predict; customers may limit use |
| Credits | Buy a pool of credits; actions consume different amounts | One currency for many features; room to reprice actions | Credits can feel opaque; needs clear dashboards |
| Outcome-based | Per resolved ticket, completed task or qualified lead | Strongest alignment with customer value | Outcome definitions, attribution and disputes |
| Hybrid | Subscription with included usage, plus overage or credits | Predictable base with protection against heavy use | More complex to explain and bill |
What well-known products have done
Public examples show how quickly AI pricing has evolved. These were checked in November 2026 and may change.
| Product | Approach |
|---|---|
| Intercom Fin | Charges per Fin outcome (US$0.99 at the time of writing), separately from helpdesk seats |
| Zendesk | Announced outcome-based pricing for AI agents in 2024, charging for automated resolutions |
| Salesforce Agentforce | Moved from a per-conversation price to Flex Credits in 2025, with each agent action consuming credits |
| GitHub Copilot | Introduced premium request allowances in 2025, then announced a move to usage-based AI credits priced by tokens from June 2026 |
| Cursor | Changed its Pro plan from request limits to usage-based limits in mid-2025 and later clarified the change after customer feedback |
Pro tip
The pattern across these examples is a move away from "unlimited AI" towards pricing that tracks usage or outcomes. The Cursor example also shows the reputational cost of changing pricing without clear communication.
Start from your unit economics
Before choosing a model, know what one unit of AI value costs you. For each AI feature, measure the cost per action or outcome (model usage, retrieval, infrastructure and any human review), and how that cost varies across customers. See AI FinOps for how to meter and attribute these costs.
Then check three things for each candidate pricing model: the gross margin for typical, heavy and extreme customers; what happens to margins if your model provider changes prices or you switch models; and whether customers can predict their bill well enough to buy with confidence.
Choosing a model
| If your product… | Consider |
|---|---|
| Uses AI lightly to improve an existing workflow | Seats with AI included, with fair-use limits |
| Has AI usage that varies widely between customers | Hybrid: subscription with included usage and overages or credits |
| Offers several AI actions with different costs | Credits, with a published table of what each action consumes |
| Delivers a clear, countable result (resolved ticket, processed invoice) | Outcome-based pricing, if outcomes can be measured reliably |
| Sells to developers or technical buyers | Usage-based pricing with transparent rates and budgets |
Outcome-based pricing: define the outcome like a contract
Outcome pricing is attractive but demanding. Write down exactly what counts: for a support assistant, does a conversation count as resolved if the customer doesn't reply? If they come back within 24 hours? If a person reviewed the answer? Record the evidence for each billed outcome, let customers see it, and have a clear process for disputes. Without this, outcome pricing turns into billing arguments.
Protecting margins
- Include usage allowances per plan rather than "unlimited" AI
- Set fair-use limits and quotas per user and account; see protecting AI applications from cost abuse
- Keep the right to adjust what a credit buys as model costs change, with notice
- Route simple tasks to smaller models and reserve expensive ones for complex work
- Watch for new free tiers or trials being used as a free model API
- Review margins per plan and per customer monthly
Adding AI to your SaaS product?
ZSpace Labs builds AI features together with the usage metering, quotas and billing integrations needed to price them sustainably.
The metering and billing system behind the price
Every usage-based, credit or outcome model needs reliable metering in your own product, not just your provider's invoice:
- Usage events recorded at the point of use, with customer, feature, quantity and an idempotency key so retries aren't double-counted
- Rating that turns events into charges according to each customer's plan, allowances and credit balance
- Customer visibility: usage dashboards, balance alerts and budget controls inside the product
- Billing integration: usage summaries sent to your billing system for invoicing; see subscription billing architecture
- Reconciliation between metered usage, invoices and your own provider costs
Build or buy billing
Billing platforms increasingly support usage-based pricing. Stripe, for example, completed its acquisition of the usage-billing company Metronome in January 2026 and now points new usage-based integrations to it, and open-source options such as Lago exist for teams that want to self-host. Buy where these fit. Build custom metering and rating when your pricing combines several dimensions the platforms don't handle well, when outcomes must be calculated from your own application data, or when usage must be enforced in real time inside the product. Most teams end up with custom metering feeding a commercial billing system.
Common mistakes
- Launching "unlimited AI" and discovering heavy users later
- Pricing on one provider's current token rates
- Credits with no clear table of what actions cost
- Outcome pricing without a precise definition and evidence trail
- No usage dashboards, so customers are surprised by bills
- Changing pricing without clear notice and transition terms
Conclusion
AI pricing works when it reflects how customers get value and what it costs you to deliver. Measure unit economics first, choose the simplest model that protects margins, make usage visible to customers, and build metering you can trust. For controlling the costs behind the price, see LLM cost optimisation.
Rethinking pricing for AI features?
Tell us about your product, AI features and current billing setup. ZSpace Labs will help you design the metering and billing to support it. Discuss your SaaS product.
Common questions.
The main ones are per-seat subscriptions with AI included or as an add-on, usage-based pricing by tokens, requests or tasks, credits that customers buy and spend across features, outcome-based pricing per resolved case or completed task, and hybrids that combine a subscription with included usage and overage charges.
Because AI costs scale with usage, not users. A few heavy users can cost far more than the seat price, and as AI does more of the work, customers may need fewer seats. Many vendors now combine seats with usage allowances or credits.
Charging for a defined result, such as a support conversation resolved by AI, rather than for access or usage. It aligns price with value but requires a clear, measurable definition of the outcome, reliable tracking and a process for disputes.
Track cost per unit of whatever you charge for, keep a margin buffer, and avoid fixing prices to one provider's current token rates. Credits and plan allowances give you room to adjust what a credit buys as costs change, with notice to customers.
Reliable usage metering in your own application, a billing system that can aggregate usage events and rate them against plans, customer-facing usage dashboards and alerts, and invoicing with clear line items. Options range from billing platforms with usage features to open-source and custom billing.