Plan requirement
| Subscription | Suite Professional or higher |
| Access | Admin |
AI-handled tickets sit in the same data as everything else. Separating them is what stops your averages lying to you.
Why it matters
An AI agent resolves the easy cases in seconds. Left in the same average, resolution time falls sharply and your team appears to have become much faster, when what happened is that the quick work left the queue.
The remaining human tickets are harder and take longer, and that shows as a decline unless you separate them.
How to tell them apart
Zendesk does this for you, and it is worth using before building anything of your own. Tickets carry two fields:
- Resolution type, which reads Automated where an automated resolution happened.
- Resolution tier, which says how much automation it took: Assisted escalation, Contained resolution or Verified resolution.
Channel works as a rough cut where an AI agent covers one, but the two fields are exact and need no configuration.
A tag, only for what the fields do not say
Earlier advice here was to set your own tag. Do that only for a distinction Zendesk does not already make, such as marking a particular campaign or pilot. Duplicating Resolution type with a tag of your own gives you two sources that can disagree, and the tag is the one that will drift.
Three groups, not two
Fully automated, started by AI and finished by a person, and purely human. The middle group behaves differently from both and is where most of the interesting change happens.
Resolution tier already draws that line: Assisted escalation is the middle group, and the two resolution values are the automated one. So the split you want is a field, not a calculation.
Report the split, then the averages
Lead with how work divided between those three, then give times within each. A single average across all of them answers no question anybody has.
Fix the baseline once
Agree how AI tickets are identified before anybody builds reports on them. Redefining it afterwards invalidates the comparison you were trying to make.
One hard limit on the baseline: automated resolutions data exists only from 7 October 2024. Anything earlier cannot be split this way, so a before-and-after that reaches further back has to be built some other way or not promised.
Comments
0 comments
Article is closed for comments.