AI Agent Pricing: Per Seat, Per Task or Per Outcome?
7 min readHow GaaS (Agentic as a Service) pricing works compared with per-seat SaaS, and the questions to ask before agreeing to pay per task or per outcome.
Zunkiree Labs Team
· Updated
In short: SaaS usually charges per seat, meaning per person with a login. GaaS (Agentic as a Service) usually charges for work done, either per task or per outcome. The difference sounds small but changes what you are buying, what can go wrong, and what you should ask a vendor before signing. This post is practical buying guidance, not market data.
Key Takeaways
- Per seat charges for access. You pay whether or not the tool does anything useful that month.
- Per task charges for each unit of work the agent attempts or completes, such as a document processed or a ticket handled.
- Per outcome charges only when a defined result is delivered, such as a qualified lead or a reconciled invoice.
- The hard part of outcome pricing is not the price. It is agreeing on what counts as an outcome and who decides.
- Ask what happens to failed attempts, retries and edge cases before you compare price lists.
Why SaaS Charges Per Seat
A SaaS product is a tool that a person operates. Value scales roughly with how many people use it, so charging per seat is simple to explain and to forecast. If you want the full contrast between the two models, see the key differences between GaaS and SaaS and the GaaS explainer.
The weakness appears when software does the work itself. If one agent can process what several people used to process, charging per seat no longer tracks the value delivered, and buyers start asking to pay for the work instead.
The Three Models Compared
| Per seat | Per task | Per outcome | |
|---|---|---|---|
| You pay for | Access for a named user | Each unit of work | A result you defined |
| Cost predictability | High | Medium, depends on volume | Medium, depends on success rate |
| Who carries the risk of failure | You | Shared, depends on the contract | Mostly the vendor |
| Hardest question | How many seats do we need? | What counts as one task? | What counts as a delivered outcome? |
None of these is always best. Per seat suits tools your team operates by hand. Per task suits steady, repeatable work. Per outcome suits work where success is easy to verify and attribute.
Where Outcome Pricing Gets Difficult
Outcome pricing sounds like the fairest model, and it can be, but only when the outcome is well defined. Four things tend to go wrong:
- Vague definitions. "A qualified lead" means different things to marketing and sales. Write the definition into the agreement, with an example of a qualifying and a non-qualifying case.
- Who judges. If the vendor decides whether it delivered, the incentive is obvious. Agree on a check you can both run, such as a sample you review.
- Failed attempts. Some work needs retries or hits dead ends. Ask whether failed attempts are free, capped or billed, because this changes your real cost per outcome.
- Attribution. If a result depends on several systems and people, it is hard to say what the agent contributed. Narrow outcomes are easier to price than broad ones.
Questions to Ask Before You Sign
- How is a task or outcome defined, in writing?
- Who decides it was delivered, and how can I check?
- What happens to failed, retried or partial work?
- Is there a minimum monthly commitment or a volume cap?
- What does it cost if volume doubles or halves?
- Can I see a log of every action the agent took on my behalf?
- What happens to my data and workflows if I leave?
For a wider checklist on choosing a provider, see how to choose a GaaS provider. For a look at how seat-based pricing is changing across the software industry, see SaaS pricing in 2026: two studies.
A Simple Way to Compare Offers
Pick one real workflow you run today and estimate its monthly volume and its current cost, including staff time. Then ask each vendor to price that same workflow under their model. Comparing quotes for the same job is more useful than comparing rate cards, because the models measure different things.
The Short Version
Per-seat pricing charges for access, per-task pricing charges for work attempted, and per-outcome pricing charges for results delivered. Outcome pricing aligns incentives well, but only if the outcome, the check and the treatment of failures are all defined up front. If you are exploring GaaS for your own workflows, start from one process and ask what a clean definition of "done" would look like. You can talk to the team through the contact page.
Frequently asked questions
What is the difference between per-seat, per-task and per-outcome AI agent pricing?
According to the post, per-seat pricing charges for access for a named user, per-task pricing charges for each unit of work the agent attempts or completes, and per-outcome pricing charges only when a defined result is delivered, such as a qualified lead or a reconciled invoice. The post calls this practical buying guidance, not market data.
Why does SaaS charge per seat?
The post explains that a SaaS product is a tool a person operates, so value scales roughly with how many people use it, which makes per-seat charging simple to explain and forecast. The weakness appears when software does the work itself, because per seat then no longer tracks the value delivered and buyers ask to pay for the work.
What makes outcome-based pricing for AI agents difficult?
The post names four problems: vague definitions of what counts as an outcome, who judges whether it was delivered, how failed attempts are billed, and attribution when a result depends on several systems and people. It advises writing the definition into the agreement and agreeing on a check both sides can run, such as a sample you review.
What should you ask before signing an AI agent pricing contract?
The post lists questions including how a task or outcome is defined in writing, who decides it was delivered, what happens to failed, retried or partial work, whether there is a minimum monthly commitment or volume cap, what it costs if volume doubles or halves, whether you can see a log of every action, and what happens to your data if you leave.