What Counts as a "Resolution"? Ask Before You Compare AI Helpdesks

When an AI helpdesk tells you it "resolved" 1,000 conversations last month, that number is only as useful as the definition behind it. Did a customer confirm the answer helped? Did a human never have to step in? Or did the customer simply go quiet?

Most vendors don't spell this out. Zendesk now has, at least for the resolution types that show up on its billing dashboard, and the definitions are worth reading closely whether you use Zendesk or not. They show how much a "resolution" can vary from one product path to the next.

What Zendesk's new help-centre article says

On 20 October 2026, Zendesk published a help-centre article titled "What do the different automated resolution types mean in Automated resolutions dashboard?" It lists seven resolution types and what each one counts. A few of them are defined by what the customer does next, and two are defined partly by the customer doing nothing:

• Auto-replies are resolutions where Zendesk's legacy AI agent "suggested help center articles over email or web form, and the customer either closed their request or engaged with the suggested article and did not reply again within 72 hours."
• For Web Widget (Classic), "resolutions are counted when the customer views a suggested article preview or marks the issue solved within a 2-hour session."
• Topic based triggers are resolutions from a Support trigger that sends article recommendations, and it "counts as resolved once the customer doesn't send another public reply within 72 hours."

The current AI agent types are described differently. Chat automation counts conversations where the AI agent "resolved the issue without handing off to a human agent"; Ticket automation counts tickets resolved "without a human agent needing to respond"; Voice automation counts calls resolved "without transferring to a human agent." The article does not apply the 72-hour or article-preview rules to those types.

Zendesk adds its own caveat: "Because Zendesk automation offerings have evolved over time, these resolution types may reflect both legacy and current product experiences."

Why the definition matters more than the number

Read those definitions as a support lead rather than as a buyer. A customer who reads a suggested article, gives up, and emails again from a different address next week can count as resolved under a "no reply within 72 hours" rule. So can a customer who closes the tab because the answer didn't help and they've decided to phone instead. None of that means the rule is unreasonable. Silence is often a sign the answer worked. But it is a different claim from "the customer confirmed it was solved", and a dashboard that mixes the two doesn't tell you which you're looking at.

It matters for budgeting too. When AI is billed per resolution, the definition of a resolution is also the definition of what you pay for.

Five questions to ask any helpdesk about its AI numbers

1. What event makes a conversation count as resolved? A customer confirmation, the AI not handing off, a set period of silence, or something else?
2. Does the definition differ by channel or product? Zendesk's article shows it can: chat, tickets, voice and legacy article suggestions are each counted their own way.
3. What happens when the same customer comes back? If they return after the window closes, is the first conversation still counted as resolved?
4. Can you see the conversations behind the number? A resolution count you can't sample is a number you can't check.
5. Is the thing you're billed for the same thing you're told was resolved? If they differ, you need both.

The useful habit is simple: every few weeks, pull a sample of "resolved" conversations and read them. Did the customer get what they needed, or did they just stop writing?

How DeskLeap counts

We should answer our own question. DeskLeap doesn't count or bill resolutions at all. The unit is an AI reply: each time the AI Agent sends a message to a customer, or drafts one for your team to approve, that counts as one reply, including when it asks a clarifying question. What the customer does afterwards doesn't change the count. The platform is 100% free with unlimited human agents; an optional AI Agent add-on is $29/mo each, and each one includes 500 AI replies a month.

That is a different trade-off, not a free lunch. A reply count tells you what the AI did, not whether it helped. So the same habit applies to us: read the conversations, not just the total.

Source

Zendesk, What do the different automated resolution types mean in Automated resolutions dashboard? (published 20 October 2026; read 22 October 2026). Quotations are verbatim from that article.