Zendesk announced its Outcome AI Platform on November 10 at its AI Summit in San Francisco, and the headline change is billing. Instead of paying per agent seat, customers pay for each eligible issue the AI resolves or task it completes, drawn down from an Outcome Budget they set up front. Sierra has argued for the same model, and OpenAI has tried it with some customers.
On paper this is the pricing every buyer of AI support has asked for. You stop paying for software that sits there and start paying for work that got done. We like the direction. But there's a detail founders and support leads should read twice: Zendesk decides what counts as resolved, and it checks each resolution with its own AI evaluation model. Low-value exchanges are excluded from billing, which is good. The exact criteria aren't public, which is less good.
The meter and the grader are the same company
With seat pricing, the vendor's incentive is boring: get you to add seats. With outcome pricing, the vendor's revenue depends on a classifier saying "yes, this was solved." Nobody has to be acting in bad faith for this to matter. The number on your invoice now comes from a model you don't control, judging against a definition you didn't write.
Think about how a support conversation actually ends. A customer asks where their refund is, the bot explains the policy, the customer stops replying. Resolved? Maybe. Or they gave up and emailed again two days later from a different address. A grader looking at one transcript can't see that second email. Your data can.
Write your own definition first
Before you turn on any outcome-priced agent, Zendesk or otherwise, write down what resolved means for your business in terms you can measure without the vendor. A workable starting point:
- No repeat contact from the same customer on the same topic within 7 days
- No reopen of the ticket and no escalation to a human
- No refund, chargeback or cancellation tied to that conversation in the following 14 days
- A completed action where one was needed (the address actually changed, the order actually shipped)
Then log the ID of every interaction the vendor bills you for, and run a weekly job that checks each one against your definition. Think of it as the same reconciliation your finance team already runs against a payments provider. If 92% of billed resolutions pass your checks, you've got evidence the spend is working. If it's 70%, you have a concrete conversation to take to your account manager, with ticket IDs attached.
Zendesk's Agentic Analytics lets managers check whether changes reduce repeat contacts, so some of this lives inside the product already. Keep a copy outside it anyway. The audit you need most is the one the billing party doesn't run.
Set the budget like a rate limit
An Outcome Budget is a spending cap, so treat it like one. Start it low, raise it in steps tied to your reconciliation numbers, and alert when the burn rate jumps. A sudden spike in "resolutions" after a model update could be the AI getting better. It could also be the grader getting more generous. You want to know which before the month closes.
The same thinking applies to Zendesk Bridge, which lets outside agents like ChatGPT take actions in your support system. It ships with permission checks, human approval where you require it, and a kill switch. Decide in advance who is allowed to press it and what triggers that call.
If you're the one selling outcomes
A lot of the teams we talk to are building AI products and eyeing this pricing model for themselves. Go for it, but ship the evidence alongside the invoice. Give customers a per-outcome log with the transcript, the action taken and the reason it qualified. Publish your definition. Let them dispute line items. Vendors who make outcome billing auditable will win procurement conversations, because the first question any CFO asks is "how do I know?"
Zendesk's own survey found 84% of consumers want a human available at all times. Outcome pricing doesn't change that. It just makes being vague about what "done" means a lot more expensive.
We're here to help founders and teams design and build digital products that are built to scale with you, not slow you down. If you're looking to build something, get in contact with us today!