Plan requirement
| Subscription | Suite Team or higher, Support Team or higher |
| Access | Admin |
An automated resolution is a conversation an AI agent finished without a person, and it is the unit your AI agent usage is measured and billed in. Escalated conversations are never evaluated for one, and the rest are checked by a language model before they count.
What has to be true
| Not escalated | Only conversations that were not escalated are evaluated at all. |
| Something was answered | Either at least one use case was identified and a response given, or a sufficient knowledge answer came from agentic reasoning or generative replies. |
| An LLM agrees | It reads the text of the conversation to confirm the request was satisfactorily resolved without human intervention. Conversations that pass become a Verified resolution; conversations that fail become a Contained resolution, and that tier does not draw on your allowance. |
Counted per conversation, not per ticket
One person's visits across channels, browsers or devices are separate interactions. That is why a single ticket can show a resolution alongside work an agent did: those were two conversations.
What does not count
- A conversation handed over to an agent, regardless of how much the AI did first.
- An exchange where the AI only gathered information before passing it on, which is the same thing.
- Anything the verification step judges not to have been resolved.
Note what is not on that list: a customer who comes back later with the same question. The verification looks at the conversation as it ended, so a resolution that did not hold still counted, and you still paid for it. That is the case worth hunting for.
Why the unit is a good one
You pay for outcomes rather than conversations attempted. An AI agent that talks to a thousand people and helps none of them consumes nothing, which aligns the billing with whether the thing works. Plans include a resolution allowance, with overage beyond it.
What it does not tell you
Whether the customer was satisfied. The check is that nobody escalated and the exchange reads as resolved; a customer who gave up politely can look the same as one who was helped.
So read resolutions next to what happens afterwards: repeat contacts on the same subject, and what customers say when asked.
Silence counts as success, which is the sharpest edge of that. A conversation that goes quiet is checked once it counts as ended, and when that is depends on the channel: email after 72 hours, messaging after 2 hours by default, and voice the moment the caller hangs up. Two hours is short enough to catch somebody who stepped away from their desk. It can be raised to 72 hours for messaging; ask us and we will arrange it. A reply that arrives later lands on the ticket but not back in the agent's conversation, and the resolution cannot be undone. A customer who went on holiday mid-conversation therefore reads as a success in the figures.
Watch the trend, not the total
Resolutions rise with contact volume. What matters is the share of conversations resolved, and whether that share moves when you improve an article or add a procedure.
Where to see it
In the AI agents reporting, alongside how many conversations were handled and how many were escalated. Those three read together tell a clear story; any one alone does not.
Comments
0 comments
Article is closed for comments.