How to measure support quality

A feeling that “we answer fine, I think” is not something you can act on. Four numbers from the dashboard are.

Support is usually judged by complaints: as long as nobody is shouting, things must be fine. The problem is that most unhappy customers do not complain — they simply leave. To see that coming, four indicators are enough, and all of them are already in the dashboard.

First response time

The most honest chat metric. Look at the worst cases rather than the average: those are the ones that turn into lost deals. The “escalation (min)” setting is effectively the limit you promised yourself — cross it and the administrator gets an alert. Start at fifteen minutes and lower the bar until alerts become rare.

Share of unanswered conversations

Conversations where the last message is from the customer and it stayed unanswered. They hang in the list silently and appear in no report until you look for them on purpose. Going through that queue once a day fixes more than any written policy.

Customer ratings

Conversation ratings (Basic and above) show what response time cannot: fast does not always mean useful. One or two bad scores mean nothing, but a recurring low score for one operator or one topic is a clear pointer to where to look.

Repeat requests

If a person comes back with the same question, the first answer was poor — even if it arrived within a minute. Customer history is visible in the dashboard, and repeats usually cluster around one theme: delivery, compatibility, returns. That is a ready-made list of what to fix on the site or in reply templates.

What to do with the numbers

Metrics are not for a report but for three decisions: add a person to a shift, rewrite the page that makes everyone ask the same thing, or change the routes. If the data leads to none of those actions, you can stop collecting it.

On honest measurement

Do not tie bonuses directly to response speed: operators will learn to send “hello, checking now” within ten seconds and the metric will stop meaning anything. Watch response time and rating together — as a pair they are much harder to game.

Where to start

If you have had no metrics at all, start with one: the share of requests answered within the first five minutes. That number almost always explains both conversion and complaints. Once it stops sagging, it makes sense to add the rest — ratings, repeats, distribution across operators.

Talking to the team

Numbers are useful while they describe the process rather than people. “Forty per cent of our evening replies take over an hour” is a conversation about the shift schedule. “Your ratings are bad” is a conversation about nothing, unless you can show the actual conversations. So bring every metric with two or three examples: they usually make the fix obvious.