Measuring support quality: CSAT, response time and resolution
Which customer support metrics are worth tracking, how to collect CSAT without annoying customers, and how to turn the numbers into weekly improvements.
"Are we doing a good job?" is a hard question to answer from gut feeling. A few well-chosen numbers make it much easier. The trick is to track a small set of metrics consistently, rather than every number a dashboard can show.
This post covers the metrics that matter most for chat support, how to collect them and how to act on them.
The core four
1. Customer satisfaction (CSAT)
CSAT asks the customer directly: how was this conversation? It is usually a simple rating after the chat closes. It is the closest thing you have to the customer's own verdict.
2. First response time
The time between the customer's first message and the first real reply from your team. It shapes the customer's first impression more than anything else. Automated greetings should not count.
3. Resolution time
The time from the first message to the conversation being closed. It tells you how long problems take to solve, not just how quickly you said hello.
4. AI resolution and handoff rate
If you use an AI agent, track how many conversations it resolves on its own and how often it hands over to a person. This shows whether the AI is genuinely taking work off the team.
All four are available in Inboxbee reports, per channel and per teammate.
Collecting CSAT well
A rating is only useful if customers give it. A few habits help:
- Ask once, right after closing. The conversation is fresh and the effort is small. In Inboxbee, the "Rating request after close" automation template sets this up.
- Keep it to one question. A rating plus an optional comment is enough. Long surveys get abandoned.
- Don't ask in every situation. If a conversation ended with a customer who was clearly unhappy for reasons outside your control, a rating request can feel tone-deaf. Use automation conditions to skip certain tags if needed.
- Read the comments. The score tells you something is wrong. The comment tells you what.
Response rates will never be complete, and customers at both extremes are more likely to answer. Treat CSAT as a strong signal, not a precise measurement, and look at trends over weeks rather than single days.
Median, not average
For time-based metrics, use the median where you can. A few very slow replies, for example overnight messages, can pull the average far from what most customers experience. The median tells you what a typical customer saw.
It also helps to look at the slow end separately. Which conversations waited longest, and why?
Split by channel
Customers expect different speeds on different channels. A minute feels long on website chat, while a few hours can be fine for email. Compare each channel against itself over time, not against the others.
Read metrics together
Each metric can mislead on its own:
- Fast first response, low CSAT: quick acknowledgements but slow or poor answers.
- High CSAT, very long resolution time: customers are patient, but you may be relying on goodwill.
- High AI resolution, falling CSAT: the AI may be closing conversations that should have gone to a person. Read a sample and check your handoff rules.
- Low AI resolution, high handoff: the AI is missing knowledge. Add help articles or Q&A pairs for the common questions.
Turn numbers into a weekly habit
A short weekly review beats a long quarterly one. Fifteen to thirty minutes is enough:
- Check the four core metrics against last week.
- Look at busiest hours and compare them with working hours and staffing.
- Read every low rating and its conversation.
- Pick one improvement for the week: a new saved reply, a help article, a routing tweak or a change to working hours.
Write down what you changed, so next week you can see whether it helped.
Don't turn metrics into targets for individuals
Per-teammate views are useful for coaching and for spotting overload. They become harmful when they turn into league tables. An agent who takes the hardest cases will often have slower times and sometimes lower ratings. Use the numbers to start conversations, not to end them.
Summary
- Track CSAT, first response time, resolution time and AI resolution.
- Ask for ratings once, right after closing, with one question.
- Use medians, split by channel and read metrics together.
- Review weekly and make one change at a time.
To act on response time specifically, read how to cut your first response time without hiring.