Resolved or just deflected? How to tell with a small team
A quiet ticket is not a solved ticket. Here is how a small team can judge outcomes using the Metrics page, ticket history and a few honest checks.
A customer asks a question, gets an answer, and never writes back. Was that a win? Maybe. Or they gave up, or they emailed you directly, or they churned quietly. If your only number is "tickets we didn't have to handle", you can't tell these apart.
That gap is the difference between deflection and resolution. Deflection means the question didn't reach a person. Resolution means the customer's problem is actually gone. Big teams buy analytics tools to separate the two. A team of five can do a reasonable job with the screens it already has, if it knows what each one does and does not show.
What the numbers can and can't tell you
Be clear about one boundary first. Cerberius Agents has two things that talk to people, and they are measured differently.
The chatbot answers visitors on your website or portal, from your help pages only, and nobody approves its answers. The support agent works for your team through the Queue, and a person approves every reply unless you have turned on automatic replies.
The Metrics page, titled How the agent is doing, describes the support agent's work on tickets. It is not a scorecard for chatbot conversations. I'll say below what to do about that. For now, keep it in mind so you don't read a ticket number as a chatbot number.
Start with the Metrics page
Open Metrics and pick Last 7 days, Last 30 days or Last 90 days. Every number counts tickets that arrived in that period, plus everything that has happened on them since. Full definitions are on the Metrics and feedback page. Three cards matter most for judging outcomes.
Replies sent without edits
This is the share of sent replies that went out exactly as the agent wrote them. It tells you how often your team looked at a draft and found nothing to fix. It does not tell you the customer was happy. A reply can be sent unedited and still miss the point if the reviewer was in a hurry.
Root cause confirmed right
This is the share of thumbs up among all votes on the agent's analysis. You vote with the thumbs up or down next to Was the analysis right? on the ticket, once the agent has findings. Each person gets one vote per reply, and you can change it.
This is the closest the page gets to "did it understand the problem". It only works if your team actually votes. A dash (—) means there isn't enough data yet. A high percentage from three votes is not much to go on, so look at the "how many of how many rated" line underneath before you trust the bar.
Time to findings, and failed runs
Time to findings is a median, so a few slow tickets don't skew it. Agent runs that failed turns amber when any piece of work didn't finish. Neither says anything about resolution. They tell you whether the machinery is working.
Use ticket history for the part the numbers miss
Metrics tell you about the agent. Whether the customer's problem went away is a question about individual tickets, and the ticket page is where you answer it. See Tickets and the ticket page for the layout.
In What happened, read these signals:
- Edited before sending. Your team changed the reply. Open the ticket and compare it with the agent's draft, which stays under The agent's draft, before edits. The edits show you what the agent keeps getting wrong.
- Looked again after the requester's reply. The customer wrote back after an answer. That is a sign the first answer didn't settle it.
- Skipped replies. A reply nobody sent is a draft your team judged wrong. If the requester is still waiting, the ticket shows as Needs a person.
- How the ticket ended. A closed ticket shows Closed by who closed it, and when. A ticket closed for going quiet says it was closed automatically after the requester went quiet.
That last one is the trap. If you switch on Close quiet tickets automatically in Settings › Ticket system, tickets that were answered and then went silent close after the number of days you choose. That's a useful tidy-up. But "closed after silence" and "closed because the customer said it worked" are different outcomes. Don't count them together.
The closing that does carry a signal is the other one: when the requester says the problem is solved, the agent proposes Close the ticket with its reason, and you decide. Those are the tickets where a person said it worked.
A simple monthly check you can do in an hour
You don't need a dashboard for this. Once a month, set Metrics to Last 30 days, and write down three numbers: Replies sent without edits, Root cause confirmed right, and how many tickets arrived. Then do this:
- Open ten closed tickets. Mix ones closed by a person with ones closed after silence.
- For each, read What happened and ask: did the customer confirm it worked, go quiet, or write again?
- Sort them into three piles: confirmed, silent, came back.
- Compare the piles with last month's.
Ten tickets is a small sample, and I'd treat it as a prompt for questions, not a statistic. If the "came back" pile grows, look at what those tickets had in common. If most of the "silent" pile is one topic, that topic is worth a phone call to a customer or two.
When the numbers look good
The help suggests reading Replies sent without edits and Root cause confirmed right together. If both are high, the agent is doing well, and you may be ready for automatic replies on simple tickets. I'd add a condition from experience: only widen automation when your sample of closed tickets shows customers confirming or not returning, not just when your edit rate drops. A low edit rate can also mean reviewers stopped looking closely.
For the chatbot, the honest position is that you need your own signal. Because it answers only from your help pages, a wrong chatbot answer usually points at a gap or a stale page. Watch for repeat tickets on topics your help pages claim to cover. Tickets that arrive after a visitor has already been "answered" are the closest thing to a deflection test your ticket history will give you. See Help sources for how to fix the pages behind it.
Making the metrics improve
Votes don't change how the agent works on their own. To improve answers, correct it on the ticket in Write › To the agent, add background under Settings › Agent, or set rules under Organisation › Rules. Then check next month whether the numbers and the piles moved.
A good next step: open Metrics this week, set 30 days, and read ten tickets end to end.