Intercom charges $0.99 per resolved support conversation. Zendesk charges between $1.50 and $2.00 per automated resolution. HubSpot dropped its rate to $0.50 in April 2026. These numbers appear in nearly every article written about outcome-based AI agent pricing this year, and they explain why the model sounds appealing: you pay only when the agent actually finishes the job. What those articles rarely mention is what happens after you sign, once the agent gets better at its job.
What the Pitch Actually Promises
Outcome-based AI agent pricing ties your bill to a completed result rather than to time spent, seats purchased, or tokens consumed. A resolved ticket, a booked meeting, a reviewed document. If the agent does not finish the task, you owe nothing. Vendors describe this as risk shifting from buyer to seller, and on paper it is. The buyer no longer pays for effort; the buyer pays for output.
The Part the Sales Deck Skips: Your Savings Do Not Come Back to You
Under a flat per-outcome rate, a vendor that makes its agent faster or cheaper to run keeps the entire gain. Your price per resolution stays fixed even as the underlying cost to deliver that resolution drops. A support conversation that took the agent four steps in January and takes two steps by June still costs you the same $1.50, because the contract was never written around cost. It was written around price.
Forbes covered this dynamic under the label of the principal-agent problem: the party doing the work benefits from every efficiency gain, while the party paying for the work sees no change at all. The gap between what the vendor spends to deliver an outcome and what you pay for it, under outcome-based AI agent pricing, can widen for years without ever showing up on an invoice.
Who Actually Defines “Resolved”?
A second issue sits underneath the pricing model rather than beside it. Outcome-based contracts require someone to define success, and in nearly every case, the vendor writes that definition. A conversation the customer had to follow up on the next day may still count as resolved on the vendor’s dashboard. A document review flagged as complete may have missed a clause a human reviewer would have caught. None of this requires bad faith. It requires only that the definition of “done” was written by the party being paid when something counts as done.
The Cherry-Picking Problem Buried in Flat-Rate Outcomes
Research on agent workflows has found up to a hundred-fold cost difference between a simple task and a complex one, even within the same product category. A flat per-resolution rate cannot reflect that spread. It averages easy and hard cases into one number, which means a vendor paid the same $0.99 for a simple password reset and a genuinely difficult, multi-step billing dispute has a quiet incentive to route effort toward the cases that resolve quickly and let the hard ones drift toward human escalation, where the outcome no longer counts against the vendor’s numbers at all.
Four Questions to Ask Before You Sign
Negotiating outcome-based AI agent pricing well starts with a framework adapted from enterprise buyers who have already been through this.
- Ask for a written definition of the outcome you are paying for, specific enough that a dispute could be resolved by reading the contract.
- Ask what happens when a task partially succeeds, since real work rarely divides cleanly into done or not done.
- Ask for a cost estimate at your own expected volume, since the same headline price behaves very differently at ten resolutions a day than at ten thousand.
- Ask directly whether your price per outcome will change as the vendor’s own delivery cost changes, and get the answer in writing.
When This Pricing Model Genuinely Works in Your Favor
None of this means outcome-based AI agent pricing is a bad deal by default. For high-volume, repetitive work where the definition of success is genuinely simple and hard to dispute, such as basic form processing or straightforward data extraction, the model does what it claims. The risk concentrates specifically in variable, judgment-heavy work, where complexity swings widely between cases and where “resolved” is a matter of degree rather than a clean yes or no. Match the pricing model to the shape of the work, not to how appealing the pitch sounds.
The Bottom Line on Outcome-Based AI Agent Pricing
Outcome-based AI agent pricing shifts risk in the buyer’s favor on the surface, and it can genuinely do that for simple, well-defined tasks. For anything more variable, the fixed price per outcome quietly protects the vendor’s margin as their costs fall, while the definition of success stays in the vendor’s hands. Read the full Forbes analysis on this exact tension before your next renewal conversation. Ask the four questions above before you sign, not after the second invoice arrives looking exactly like the first.
For more on evaluating AI vendor pricing, see our guides on the shift to usage-based SaaS pricing and measuring AI agent ROI.