Zendesk Automated Resolution Pricing: What Counts

zendesk, pricing, ai-support, vendor-evaluation

Zendesk charges $1.50 per automated resolution on committed volume, $2.00 pay-as-you-go, and the meter starts the moment your plan's monthly allowance runs out — 5 per agent on Suite Team, 10 on Suite Professional. But "automated resolution" is not "the AI agent replied." Since May 18, 2026, Zendesk sorts every AI-handled conversation into one of three tiers, and only one of them is billed. The other two are free — and the line between the free tier and the billed one is drawn by a second AI model reading the transcript, not by the customer saying anything at all. That mechanic is also the whole answer to "can I dispute this": there is no documented dispute process, because there is nothing to argue with except another model's opinion, and you only get to see that opinion three days after the fact.

This is Zendesk's own automated resolution pricing as its documentation describes it, not the price alone that every comparison post already covers.

The three tiers, and what actually separates them

Before May 2026, Zendesk billed a single kind of outcome: a conversation the AI agent touched and a human didn't finish. That single bucket is gone, replaced by three:

Only the bottom row draws from your allowance or your overage bill. The first row is an easy call — a human agent closed the ticket, so there's a record of what happened. The second and third rows are not: they're the same customer behavior (the AI answers, nobody replies) sorted into different buckets by a model's read of the transcript.

Silence reading as success is not new to the tiered system — it's the assumption the old single-bucket model ran on too, and Zendesk's own documentation is explicit that it carries forward: a customer who stops replying is treated the same as one who confirmed the answer helped, provided a second, automated check doesn't override it. The tiers didn't remove that assumption. They added a model's opinion on top of it, and made that opinion the billing event.

It's the second mandatory rewrite Zendesk customers are working through this year: the other one retires plain API tokens for OAuth on a fixed deadline through April 2027. Neither is optional, and neither is something a customer opted into on its own timeline.

What decides "verified": two checks, neither of them the customer

A conversation becomes a Verified Resolution through two separate checks, both automated. The first is a rule-based filter: was this a genuine support request, and did the AI agent handle it end-to-end without a human touching it? The second is the one doing the actual judgment call — a language model reads the full conversation and decides whether the customer's problem looks solved.

Neither check involves the customer confirming anything. There's no "was this helpful?" prompt required, no follow-up survey the billing depends on. The signal is the transcript, evaluated by a model whose only job is to guess whether the human on the other end walked away satisfied or gave up.

Zendesk does expose why a conversation was counted — the reporting dashboard shows reasoning data, the model's own explanation for the call — but only for conversations already marked resolved, and only after a 72-hour confirmation window closes. Nothing you did today shows up as billed usage until three days from now. If you want to catch a wrong call before it's final, you have a three-day window to read reasoning data on conversations you don't yet know were billed.

This is the same mechanic Intercom's Fin bills under the name "assumed resolution" — silence after an answer, charged the same as an explicit "yes, that helped." Zendesk's version adds a verification model on top, which is a real improvement on billing every silence unconditionally. It is not the same as giving the customer, or the admin, a say in the call before the invoice reflects it.

What Zendesk's automated resolution pricing actually costs

The included allowance is small by design: 5 automated resolutions per agent per month on Support Team and Suite Team, 10 on Suite Professional. Above that, every Verified Resolution bills separately — around $1.50 on committed volume, $2.00 pay-as-you-go, reported by third-party pricing breakdowns and consistent with Zendesk's own allowance-overage documentation.

The part worth checking in your own account is the default. Since January 1, 2026, overage above your allowance auto-bills unless an admin has turned on a pause setting — and that setting ships off. The only pause option Zendesk documents stops the AI agent from responding at all once the allowance runs out; there's no documented middle setting that throttles or caps spend while leaving the agent running. So the choice an admin actually has, out of the box, is auto-billed overage with no ceiling, or an AI agent that stops answering customers mid-month. Neither is "review each charge before it's billed."

A high-traffic week makes this concrete. If your team handles three times the usual volume in a busy month, the allowance doesn't stretch to match — every resolution past it bills at the same per-unit rate as the first one, and the tiering rules above decide which resolutions count against it without anyone reviewing the individual calls until the invoice is already final.

How to check your own account before the invoice, not after

1. Pull the tier breakdown, not the resolution rate. Reporting on Contained versus Verified is now separated in Zendesk's AI agent reporting. A resolution rate on its own hides how much of it is the free tier versus the billed one.
2. Read reasoning data on a sample of Verified Resolutions, not just the ones that look wrong. The dashboard shows the model's stated reason for each call; a pattern of thin reasoning ("customer did not respond, likely resolved") on conversations you'd guess were abandoned is the thing worth escalating to your CSM, even without a formal dispute process to route it through.
3. Confirm who can see the pause setting, and whether it's on. If nobody on your team turned it on, overage has been auto-billing since January 2026 by default — check the invoice history against the allowance, not just this month's total.
4. Budget for the 72-hour lag. A traffic spike this week shows up as a bill you can't reconcile until three days after it happened, by which point the month may already be closing.
5. If a third-party connector or integration triggers AI replies, confirm its conversations get attributed the same way manual ones do — Zendesk's per-resolution billing wasn't built with a specific integration in mind, so the tiering rules apply uniformly whatever originated the conversation.

Where Replium stands on this

Running the same questions on ourselves, starting with the ones where the honest answer is no.

Do we have automated resolutions to bill? No. We have not connected an LLM provider to the product, so there is no AI agent deciding whether a conversation looks resolved — every conversation in Replium today is closed by the human operator who handled it, the same person whose name is on it.

Do we bill per outcome at all? No. There is no billing in Replium yet — no plans, no invoice, nothing to reconcile against an allowance.

Who should stop reading here? Anyone evaluating whether an AI agent's per-resolution pricing is predictable enough to budget for. We don't have an AI agent to compare, so we're not a data point on that question — Zendesk's own documentation and the checklist above are what to bring into that conversation instead.

What we are: a small multi-tenant help desk — shared inbox, an embeddable chat widget, a public knowledge base that's server-rendered so search engines read the actual text — in English and Russian, with support requests handled by people, not billed by a model's opinion of a transcript. The trade that's actually true about a vendor this size: your report doesn't queue behind an enterprise account base, it reaches whoever wrote the code, and a fix that would sit in a backlog elsewhere can ship in days.

If a human-operated inbox without a resolution meter is what you need today, create a workspace — self-serve, an email confirmation, no card, and nothing to audit a bill against because there isn't one.