Explore: the metrics that matter in support

Explore can measure almost anything. That’s the trap. Open a prebuilt dashboard for the first time and you drown in numbers: tabs, tiles, curves. The real work isn’t measuring more, it’s measuring less. A handful of metrics tell you whether your support is healthy. The rest look good in presentations and never change a decision. This card separates the one from the other.

What Explore is

Explore is Zendesk’s reporting tool, and it rests on three terms. A dataset is a collection of metrics and attributes that queries data from one Zendesk product: Support tickets, AI, Knowledge, each its own dataset. A prebuilt dashboard is the ready-made report on top of it: read-only, but on most paid plans you can pull an editable copy. And a custom report is what you build yourself. Every report needs at least one metric, and it lives or dies by the right dataset: reach for the wrong one and you simply won’t find the number you’re looking for.

To start, the Zendesk Support dashboard is enough. It has nine tabs: Tickets, Efficiency, Assignee Activity, Agent Updates, Unsolved Tickets, Backlog, Satisfaction, SLAs and Group SLAs. For getting started, Tickets, Backlog, Satisfaction and SLAs are enough. That covers most of it before you ever build a report of your own.

You no longer have to build that first report of your own by hand. With Quick Reports you describe in one sentence what you want to see, and Explore generates the report from it. That takes the mechanics off your hands. The selection stays your job: which few numbers change a decision is not something the prompt answers. It only draws on the Support tickets dataset.

The metrics that matter

Six numbers, and why:

• First reply time: the time between ticket creation and the first public agent reply. This is the number your customers feel most directly. One caveat: Explore counts either calendar hours or business hours. Measure without business hours and you punish yourself for every night and every weekend.
• Full resolution time: from creation to the last solve, reopenings included (not to be confused with first resolution time, which counts only up to the first solved). The stricter measure, because it doesn’t reward tickets closed too early.
• SLA achievement rate: the share of tickets that meet their SLA targets. Where you’ve committed to reply or resolution times, this number shows whether you keep the promise. It checks the raw time metrics against the target you set yourself.
• CSAT: the share of good ratings among those submitted. Explore computes good ratings divided by submitted ratings, times 100. On the 1–5 scale, 4–5 counts as good, 1–3 as bad; the scale can also be switched to two or three levels. The denominator is only the submitted ratings, not all tickets, which is why the bare percentage is worthless without the response rate: 95% CSAT at a 3% return measures your loudest fans, not your support.
• Backlog: the unworked tickets over time. As a snapshot dataset it shows the trend: when the queue fills up, that’s an early warning no reply-time curve shows as soon. Because it’s a snapshot dataset, it can’t be mixed with event-based datasets in one report.
• One-touch rate & reopen rate: the share of tickets solved with one (or no) agent reply, against the share that reopens after being solved. Read together, they’re a quality signal: many one-touch and few reopens means clean resolution; many one-touch and many reopens means closed too hastily.

Where AI Agents are involved, a seventh joins them: the Automated Resolution. It counts only what an LLM has verified as handled with no human involved, and it’s the billing unit at the same time. What counts as «solved» there is in the note What «resolved autonomously» really means.

The discipline is in the selection: two or three of these numbers you look at weekly, the rest you deliberately leave aside.

And the ones that mislead

Ticket volume as success. More solved tickets isn’t good news when the volume itself is rising: often it just means a product problem is producing support. Volume is a diagnostic, not a performance measure.

Averages on time metrics. A single ticket that sits untouched for three weeks lifts your average resolution time so far that it says nothing about the normal case. For time metrics, Zendesk therefore recommends the median as the aggregator: where an outlier drags the average up to 20 hours, the median stays at 4. It shows the typical ticket, not the outlier.

Vanity CSAT with no response rate. See above: a satisfaction number without the number behind it, how many even replied, is decoration.

Everything at once. A dashboard with thirty tiles doesn’t get read, it gets skimmed. Four to six numbers that lead to an action beat any completeness.

The best report is the one you regularly cut something from. Once a metric goes months without changing a decision, it’s out.