On this page
  1. 1. First response time (median)
  2. 2. Resolution rate
  3. 3. Satisfaction (CSAT), with comments
  4. 4. Knowledge base hit rate
  5. 5. Conversations by hour
  6. What not to obsess over
  7. A 20-minute weekly review
  8. The takeaway

Support dashboards can show dozens of numbers. Most teams need five, looked at every week, with a question attached to each.

The five numbers to watch each week
  1. First response timeUse the median
  2. Resolution rateHow many were really solved
  3. SatisfactionRead the comments too
  4. Knowledge base hit rateWhether answers have a source
  5. Conversations by hourWhat to staff for

1. First response time (median)

Question: Are people waiting too long for the first human reply?

Use the median — one conversation forgotten over lunch shouldn’t hide a good week. Count only human replies when judging the team; automated greetings don’t count.

If it’s rising: check the busy-hours heatmap against who was online, and whether auto-assignment is on.

2. Resolution rate

Question: Are conversations ending with the problem solved?

A falling resolution rate often means conversations are being left open — or that new kinds of questions are arriving that the team can’t answer yet.

If it’s falling: read five unresolved conversations from the week. The pattern is usually obvious.

3. Satisfaction (CSAT), with comments

Question: How did it feel for the customer?

The average matters less than the low scores. Read every one- or two-star rating each week; each one is a specific story you can learn from.

If it drops: check whether it’s one topic (a shipping delay, a product issue) or one part of the process (slow hand-offs).

4. Knowledge base hit rate

Question: How often do we already have the answer written down?

This drives how much the AI agent can handle and how quickly new teammates can help.

If it’s low: take the not-answered list, add answers for the top three, and check again next week.

5. Conversations by hour

Question: Are we online when customers are asking?

A weekday × hour view shows mismatches instantly: a busy evening nobody covers, or a quiet morning with three people online.

If there’s a gap: shift one person’s hours, or let the AI agent cover that window.

What not to obsess over

  • Total conversation volume — more isn’t better or worse by itself.
  • Messages per conversation — long conversations are sometimes the most valuable ones.
  • Individual speed rankings — they push people to rush the hard conversations.

A 20-minute weekly review

  1. Look at the five numbers and their change from last week.
  2. Read the lowest-rated conversations.
  3. Add three knowledge base answers from the not-answered list.
  4. Adjust one thing: hours, routing, a canned reply, a page on your site.

The takeaway

Five numbers, one question each, one change per week. That’s enough to steadily improve support without drowning in dashboards.

Tagsmetricsreportscsat
Related featureLive Chat Analytics & Customer Service ReportsResponse time, resolution rate and satisfaction, what Cubby solved alone, busy hours, agent workload — and how visitors who chatted converted.See how it works

Related articles

AI customer service3 min read

Measuring an AI agent without fooling yourself

“The AI resolved 70%!” sounds great until you ask how it was counted. Four numbers that tell you whether AI support is actually helping.

Read article

Live chat craft3 min read

How fast is fast enough? First response time in live chat

Visitors on live chat expect a reply in moments, not hours. What a realistic target looks like for a small team — and five ways to hit it without hiring.

Read article

Sales & conversion2 min read

Proving what customer support is worth

Support is often seen as a cost. Connecting conversations to conversions — carefully — shows where chat earns its keep.

Read article

Try it on your own site

A free plan with no time limit, AI agent included. One line of code.

Start free