What to measure in customer support, and what gets measured only because everyone does

Almost any helpdesk ships with a dashboard carrying a dozen numbers. It gets looked at in the first month, shown in some meeting, and then stops being opened, because nobody knows what to do when one of those numbers moves. The problem is not measuring too little: it is measuring things that lead to no decision.

It is worth cutting the list down. These five do change something when they change, and each comes with its own trap.

Volume handled

How many cases come in and how many get closed, over the same period. It is the simplest number and the most informative, because it does not measure quality: it measures whether the role is keeping up.

What has to be read is not the total, it is the difference. If more come in than get closed, the queue grows even when everything else looks fine, and the queue is what turns into angry customers two weeks later. If more get closed than come in for a while, backlog is being recovered — and that has an end date, it is not the normal pace.

What distorts it: the way of counting. A case the customer reopens three times can show up as three cases or as one; a conversation across two channels, as two. Before comparing months, settle how you count and do not touch it again.

First response time

How long the customer waits to hear back from a person. It is the metric the person writing in notices most, and usually the cheapest one to improve: it almost never depends on resolving better, but on the queue being sorted and on somebody looking at it first thing.

It has a well-known trap and it is worth saying out loud: it can be improved without improving anything. An auto-reply, or a «hello, we are looking into it» after thirty seconds, sinks the number and moves the case nowhere. If you are going to measure it, measure the first useful reply: the one that says something about the actual case.

What it has to be read alongside: resolution time. A very good first response with bad resolution is the profile of an operation that answers fast and closes nothing.

Resolution time

How long from the case coming in to it actually being closed. It is the number closest to what the customer feels, and the hardest to interpret on its own, because it contains things that do not depend on whoever is answering: a carrier taking three days to reply, a supplier who does not answer, a decision waiting on someone in your own team.

So look at the median before the mean. The mean is moved by one monstrous six-week case; the median tells you how the normal case goes. And if you can separate time waiting on third parties from your own time, separate it: they are two different problems and they are fixed along different paths.

What distorts it: closing by silence. If cases are closed automatically when the customer stops replying, the number improves and reality does not.

Escalation rate

How many cases are not resolved in the role and have to go up to someone else. It is the only one of the five that should not be driven to zero, and this is where a lot of people get it wrong.

A very high rate means one of two things, and they are worth telling apart before drawing conclusions: either the criteria are not written down — the person does not know how far they can decide, so they escalate to be safe — or the permissions are missing — they know what to do and cannot do it. The first is fixed by writing the policy; the second, by granting the permission with a limit.

A very low rate is not good news in itself either. It can mean somebody is resolving on their own cases that should be going through you, and that gets discovered late and badly.

What is worth watching is the trend: in a role that is going well, escalations drop over the first two months and then flatten out. If they do not drop, it is not a problem with the person: it is that the criteria were never written down.

What keeps repeating

This is the one almost nobody measures, and the only one that fixes the future instead of describing the past. It is not a number on the dashboard: it is a list of the questions that arrive over and over.

The same question three times is a warning, not a coincidence. It usually means there is a product page that does not say what it should, a page that is missing, a confirmation email that reads badly, or a step in the checkout nobody has reviewed in two years. Every one of those repetitions can be removed at source, and when it is removed, volume falls — which is the only healthy way for volume to fall.

For this to work, somebody has to note it down while answering, rather than reconstruct it at month-end. It is one of the things a customer support role really does: close the day's cases and leave a note of what should never have come in.

And the satisfaction survey

The score the customer leaves when the case closes deserves its own paragraph, because it is the one most often shown and the one most often misread. It does not measure the quality of your support: it measures the quality of your support among the people who answer the survey, and those people are not a random sample. The one who had a bad time answers, and so does the one who had a pleasant surprise; everyone else closes the tab.

It is useful, but for something else: for reading the comments. One low score with a specific comment is worth more than thirty ratings with no text. Use it as a source of cases, not as the month's grade.

What not to conclude from these numbers

Two warnings, because those five numbers get misused often.

First: there is no universally good figure. A fashion ecommerce with returns and a business with technical incidents have nothing in common, and within the same sector two shops with different policies do not either. The useful reference is not anybody else's: it is your own previous month, counted the same way. How this looks in a shop with a peak season is covered in running an ecommerce operation.

Second: none of the five measures quality. You can answer fast, close fast, escalate little and be leaving customers unhappy in every case. Quality is seen by reading conversations — a small sample, every week — and no dashboard replaces that.

Why a month-end report is not follow-up

A report tells you what already happened. It arrives on the 5th, says the month was worse and does not say why. By the time anyone reads it, the next month has spent a week behaving exactly the same.

Performance follow-up is something else, and it happens with the person in the room: looking at their numbers together, reading two or three specific cases, deciding what changes this week and who changes it. Without that, the data is only good for being right afterwards. With it, each month starts different from the last.

And there is a part of the follow-up that is not the client's job. Directing the day-to-day work is done by whoever delegates; making sure the person is well directed, trained and equipped to do their job is done by us, every month, against the objectives you have set. The exact split of who does what is in how a team works.

Where to start

If you measure nothing today, start with two: volume in against volume closed, and first useful response. With those two you know whether the role is keeping up and whether the customer is waiting.

Add the list of what keeps repeating as soon as there is someone permanently answering, because that is the one that reduces the work instead of describing it. The rest will come on its own, and it will arrive with a specific question behind it — which is the only decent reason to add a metric.