Your AI support vendor bills per resolution now. What actually counts as one?

Zendesk's CEO didn't hedge about it. At HumanX in Amsterdam on September 23, 2026, Tom Eggemeier told the room: "Unequivocally: seats are dead." Sierra co-founder Clay Bavor said the same thing from the other side of the table — his company charges only when an agent resolves an issue or closes a sale, nothing for the seat, nothing for the attempt. Both predicted AI will handle the majority of customer interactions within a few years. Neither was talking about a future pricing model. Salesforce had already moved Agentforce's pricing inside the seat three weeks earlier, on September 3, and Zendesk changed its own action-credit meter the same month. The shift away from per-seat billing for AI support agents isn't a prediction anymore — it's happened at three of the category's biggest vendors in the same few weeks.

If you're evaluating or already running an AI support tool, the pricing model under discussion is very likely the one you're about to sign or renew. The number on the vendor's pricing page — $0.99 per resolution here, $1.50 there — looks like a clean, comparable unit. It isn't, and the gap between vendors is wide enough to move your monthly bill by a large multiple before you've changed anything about how the AI performs.

What actually counts as a "resolution"?

This is the question that matters most, and it's the one vendor pricing pages answer least clearly. The industry default, documented across multiple vendor comparisons, treats silence as success: if the customer doesn't reply again within some window, the ticket bills as resolved — including the customer who simply gave up and called a competitor instead of writing back. A vendor that instead requires a confirmed fix, or a customer's explicit "yes, that solved it," is measuring a materially smaller number under the exact same rate.

Two vendors can publish the identical $0.99-per-resolution sticker price and land on invoices that differ by a large margin, purely because one counts differently. You can't tell which one you're buying from the pricing page. You find out from the contract language, or from the first invoice.

Does escalating to a human still cost you?

Sometimes. A conversation the AI touches and then hands to a person bills as a resolution on some meters and stays free on others — one clause, and it can move your bill by half. This is worth asking as a direct, written question before signing, not something to infer from a sales call: "if your system starts a conversation and a person finishes it, do we pay for that interaction?" Get the answer in the contract, not the pitch deck.

Why can better performance mean a bigger bill?

Because the unit you're billed on is the same one you're trying to improve. If deflection climbs from 30% to 60%, the number of billable resolutions can roughly double — the invoice grows in exactly the direction your project was supposed to succeed in. That's not necessarily a bad deal if the AI is genuinely closing more issues. But there's a second number that quietly works against you: repeat-contact rates after AI handling run around 2.3x industry-wide, meaning a meaningful share of what gets billed as "resolved" comes back as a new ticket — sometimes a new billable resolution — because the first answer didn't actually stick. A resolution rate that looks great on the vendor's dashboard and a cost line that's climbing faster than ticket volume can both be true at the same time, and the gap between them is exactly the repeat-contact rate nobody's billing you to measure.

The three questions worth asking before you sign

1. What's the exact definition of a billable resolution — in the contract, not the pitch? Silence-based, confirmation-based, or something else. Get the definition in writing.
2. Does an AI-touched-then-escalated conversation bill as a resolution? Ask for a yes or no, not a "it depends on the workflow."
3. What's our repeat-contact rate on AI-resolved tickets, and does a repeat count as a new billable event? If the vendor can't answer the first half, they can't answer the second, and that's itself useful information.

Why a fixed-price build sidesteps this entirely

The pricing tension above only exists because the billing unit and the performance metric are the same number. A project built and owned outright — priced once, for the system, not metered per outcome — doesn't create that incentive: improving resolution rate doesn't generate a bigger invoice next month, because there isn't a next-month invoice tied to volume. It's the same tradeoff we've written about on the finance side, where teams comparing build-vs-buy on AP automation find the same gap between a per-unit SaaS meter and a scoped, owned build. On the support side, the pattern holds: on an industrial components distributor's ticket automation build, 85% of tickets are created and classified automatically, at a price that doesn't move because the system got better at its job.

What to do before your next renewal

If you're already on a per-resolution contract, pull the definition clause and the escalation clause out of the agreement this week — not at renewal, when you have less leverage. If you're evaluating a new vendor, ask the three questions above before the demo, not after. And if the honest answer is that no metered vendor gives you a bill that doesn't grow with your own success, the free AI Audit is a five-minute way to see what a fixed-price build of the same workflow would look like instead. Prefer to talk it through first? Book a call.