Plan requirement
| Subscription | Suite Team or higher, Support Team or higher |
| Access | Admin |
Low resolution rates have a small number of causes, and they are diagnosable in an hour by reading transcripts.
Symptoms
- Most conversations end in a handover.
- The agent frequently says it cannot help.
- Ticket volume has not changed since it launched.
Start by reading twenty handovers
In full, from the first message. The cause is usually obvious within ten of them, and every alternative to reading them is guesswork.
Is the knowledge base too thin?
Check this first. The questions arriving have no article behind them, so the agent has nothing to answer with.
The fix is writing articles, and no configuration substitutes for it. Take the five most frequent unanswered subjects and write those.
Are the articles unusable rather than absent?
They exist, but the answer is buried, or spread across an article covering ten things, or written for somebody who already knows the context.
Rewriting three articles so the answer is stated plainly near the top often moves the number more than any setting.
Is the scope too narrow?
The agent was pointed at a small section of content, or restricted to a few use cases, and everything else falls through.
Widen deliberately, one subject at a time, checking transcripts as you go.
Are customers asking about something else entirely?
You automated what you expected; they are asking about something you did not. The subjects in the transcripts settle this immediately, and it is a good outcome: you now know what to build.
Read the number the right way
A resolution is only counted for a conversation that was not escalated and that then passes an LLM check confirming the request was resolved. So a low rate is not the model being harsh: it is a faithful count of conversations that ended without a person, and every handover you read is one that could not.
Those handovers are not invisible in the data, which saves you reading everything. Zendesk grades each conversation into a tier, and Assisted escalation is exactly this group: the AI contributed and a person finished. Counting that tier against the other two tells you the size of the problem before you open a single transcript, and the transcripts then tell you the cause.
What is not the cause
The persona, the avatar and the greeting. Those affect how the exchange feels, not whether an answer exists.
Comments
0 comments
Article is closed for comments.