Skip to content
Resolvara

Zendesk AI agents: how a resolution is billed

How Zendesk AI agents defines a resolution, what it bills, and how to audit your own export.

Billing unit
per Verified resolution (since 2026-05-18), drawn from a dollar allowance
Price
Not published
Silence counts as resolved
Yes, a customer who stops replying can be billed as resolved.
Waiting window
Conversation end: email 72 hours after the first email; messaging 2 hours after the first message by default (extendable to 72 hours); voice at hangup.
Checked
2026-10-24
Resolvara support
Detected automatically. You can still correct the column mapping.

Facts on this page were checked on 2026-10-24. Zendesk changed its model on 2026-05-18, and many third-party articles describe the earlier one. Use the Zendesk pages listed below.

How Zendesk counts a resolution

Since 2026-05-18 automated resolutions are split into Contained and Verified. Zendesk states that you are billed only for Verified resolutions. Contained means the AI agent handled the interaction to completion without the customer requesting further assistance; it is not billed. Verified means the same, plus an LLM reviews the ended conversation and passes it. No explicit customer confirmation is required.

A conversation ends 72 hours after the first email, 2 hours after the first message in messaging (extendable to 72 hours), or at hangup for voice. Asking for clarification, giving feedback on the answer, or asking for a human prevents a Contained or Verified outcome.

Zendesk does not state a per-resolution price on the help pages read. The included allowance is a dollar pool per seat. Third-party price figures may predate May 2026, so Resolvara sets no default price for Zendesk: enter your own from your contract or invoice.

How to export

  1. Usage export. Admin Center > Account > Usage > Resolutions > Actions > Export contributing resolutions. A CSV is emailed and the link is valid for 14 days. It lists billed resolutions with ticket id, tier, channel group, brand, timestamp and cost. Header names are not documented.
  2. Ticket export. Export your tickets (CSV or API) with requester, created and solved dates, reopens and satisfaction. The ticket fields Resolution type and Resolution tier are available in Explore custom reports.
  3. Join the two on ticket id.

Legacy accounts may still carry the tag ai_agent_automated_resolution, added 72 hours after the last user interaction under the older model.

How to audit with Resolvara

  1. Choose the Zendesk preset and drop the ticket export. If you also have the Usage CSV, add it as the second file so billed ticket ids can be matched.
  2. If you have no Usage CSV, billed is inferred from Resolution tier = Verified (or the legacy tag). Check this in the mapping.
  3. Enter your price per resolution.
  4. Agent replies are off by default. The ticket field Agent replies is a count for the whole ticket. It cannot show that a reply came after the AI agent took part, and the Standard only counts a human reply that came after the AI. So for Zendesk exports the human-reply signal (human_handoff) is off by default, and the report carries a warning saying how many AI tickets have an agent reply that was not counted. To include them, turn on "Count agent replies as human handoff (cannot confirm order)" in the parameters. The report then says some counted replies may predate the AI. If your own export has a column that is only filled when an agent replied after the AI, map it as "Human reply after the AI" and it counts without the option.
  5. Review the flagged list. Contained and Assisted escalation tickets are not billed, so they appear as "not strict" but never as phantom.

Caveats

  • Native JSON ticket export is reported not to include AI agent tickets (third-party report, unverified). Use CSV, the API or Explore.
  • With the agent-reply option off (the default), a ticket that only a human reply would flag can still pass as strict. Expect fewer phantom resolutions than with the option on, and read the warning in the report.
  • Reopen count and satisfaction come from the ticket data and are not confirmed in the pages read here.
  • The Verified tier already includes an LLM check, so the number of phantom resolutions may be lower than for vendors that bill on silence alone. The Standard still flags repeat contacts and human replies.