Customer satisfaction metrics are the numbers a support team keeps so it can tell how the customers are feeling about the help, how fast the help is arriving, and whether the answers held up after the conversation closed. Six or seven of them are worth knowing, hardly more than that. And none of them ask for special software or a data person, the raw material is sitting in your helpdesk already, it has been sitting there the whole time, and this guide only shows you where to look and what to divide by what.
But nobody goes looking for these numbers out of plain curiosity, people meet them the hard way. Picture a support lead at a small payroll software company, one Monday in the weekly meeting, when the founder turns to her and asks, are the customers happy with us. Fourteen months of answered email behind her, and she sat there realising she did not actually know. The inbox got emptied most evenings, that much she knew. Some weeks felt heavier than others, she knew that too. What she does not have is one number she could put on the table. This article is the set of numbers she went and collected over the following quarter. Each one of them went and misled her at least once along the way, and the misleading part gets equal space on this page, on account of it being the useful part.
What are customer satisfaction metrics, in plain words
Stand-ins for a feeling, that is the honest description of them. There is no gauge you can wire to a customer's mood, much as everybody would love one, so what gets measured instead is the signals lying around the mood. What a person says when you ask them straight. How long they sat waiting before anyone wrote back. Whether the fix held, or came apart in their hands a week later and brought them back crosser than the first time. Each metric is watching one of those signals and one only, blind to the rest of it, and this is the reason we keep telling teams to run a small group of them together rather than crowning any single one. A number that improves by accident, or by somebody quietly cheating, will get given away by its neighbours soon enough.
Our support lead began the way most people begin, she went and copied every metric she could find into a spreadsheet, eleven rows of the things, and the spreadsheet lasted about a week before it stopped being opened at all. Eleven numbers is a dashboard nobody reads, fair is fair. The set that survived was smaller. One satisfaction score, the two clocks, and a pair of quality numbers that keep each other honest, and the rest of this page walks through the survivors one at a time, in more or less the order she met them.
How to measure customer satisfaction with one small question
CSAT went on first, the customer satisfaction score, on account of it being the most direct question in the whole toolbox, you simply ask. A conversation closes, the customer gets one line asking how satisfied they were with the help on a scale of 1 to 5, and that is the survey, the whole of it, there is no page two. Her second month of running it, 200 customers answered. 162 of those gave a 4 or a 5, and you count the happy answers, divide by everyone who answered at all, multiply by 100, and it comes out at 81%. Which is a fine place to be sitting, since B2B software support mostly lives somewhere in the 75 to 85% band. You will meet teams claiming they hold above 90% for a good long while, and some genuinely do, though fewer of them than the case studies would have you believe, to be honest with you.
The trap introduced its self in the third month, when only 14 people answered and the score leapt to 93%. The founder was delighted, naturally, and she got the uncomfortable job of explaining why the leap meant nothing. These surveys hear back from maybe 5 to 15% of customers in a normal month, and think about who actually bothers clicking, the very pleased and the very cross, the two crowds with a reason to. Nothing scandalous about a thin month, mind you. But a score built on 14 replies cannot sit beside a score built on 200 and call its self a trend. The response count belongs next to the score permanently, the two of them read together or not read at all.
The customer effort score, and why easy beats impressive
CES came a month later, and it came because of one reply in particular, a customer who wrote, in full, I did get my answer in the end, but it took three emails and a reminder from my side. No satisfaction survey had caught that mood, the man had ticked a 4, funnily enough. The customer effort score asks the very question that reply was answering already. How easy was it to get your issue sorted, a 1 to 7 scale usually, and the score is nothing fancier than the average of whatever ratings come in. Gartner has spent years arguing that effort predicts loyalty better than delight ever did, and everything we watch over here at Maxdesk sits with that argument. A customer rarely leaves because nobody charmed them. They leave when getting help has started to feel like a second job on top of the one they already have.
A fair working floor is 5.5 on the 7 point scale, and the low ratings deserve reading one at a time, each being a short account of a wall somebody walked into. Two warnings though. The survey its self must cost nothing, one tap and done, since a three question form arriving after a one question conversation reads like a joke to a busy person, and the busy people were the ones you most wanted hearing from. The other warning is about what the number can do. It tells you the climb was hard. Which stair was loose, it cannot say, that part only comes out of reading the actual conversations behind the low scores, and no one has found a shortcut around the reading yet.
Where NPS belongs, and where it does not
The founder came home from a conference wanting NPS up on the office wall, and that is roughly how NPS enters most companies, somebody senior hears it spoken of somewhere. The survey asks how likely the customer is to recommend you, on a scale of 0 to 10. 9s and 10s are your promoters. 7s and 8s sit the whole thing out as passives. Everything from 0 down to 6 lands as a detractor, harsh as that sounds for a 6. Subtract the percentage of detractors from the percentage of promoters and you get a score somewhere between minus 100 and plus 100. Above plus 30 is a solid showing for software, and the famous names go clearing plus 50.
She put the number on the wall, fair is fair, and then she argued that it should not hang over the support desk in particular. She was right. NPS moves with pricing, with product quality, with the brand, the whole relationship of it, so it is a loyalty number rather than a service number. Hand it to a support team as a target and a gentle sort of gaming starts up on its own, nobody meaning to cheat exactly. Surveys start going out only after the wins, little hints get dropped about what a ten would mean to the team, and the score climbs while the loyalty underneath it sits exactly where it always was. Better to run it quarterly, let the company one and all own it, and judge support on the numbers support can actually move. Those are coming next.
First reply time, the number customers feel in person
First reply time is the one metric on this page your customers experience in person, waiting is waiting, nobody anywhere has learned to enjoy it. The clock runs from a message arriving to the first human reply going back, and we lean on the word human deliberately, some tools count the automatic acknowledgement as a first reply, and once that happens the whole measurement has turned its self into fiction without anyone noticing. Go and check what your own tool counts before trusting a single week of it.
Business hours, never clock hours, that is the first rule of measuring it. Say an email lands at six on a Friday evening. The office opens Monday at nine and the reply goes out at ten, so in clock hours the reply took sixty four, while against the business hours you actually publish it took one, and whichever of those two readings you put in the report decides whether the report understands weekends or goes about treating every Monday like an emergency. Quote the median beside the mean while you are at it, one stuck ticket goes and drags an average somewhere unfair. On targets, a reply inside the same business day is a respectable floor for email support, four business hours starts feeling properly fast to the person on the other end, and past that you are into bragging territory. Speed proves very little on its own, mind you. A quick shallow reply that asks the customer something they already answered in the first message will score beautifully and help nobody, and that right there is why the quality numbers further down have to travel together with this one.
Resolution time, first contact resolution, and the reopen rate
A short story about an agent now, because the three quality numbers only make sense as a pack. First contact resolution is the share of issues sorted fully in one single conversation, no follow up, no second ticket, and the firms who study service desks for a living, SQM Group being one, put the typical figure somewhere in the high 60s to low 70s as a percentage, the excellent teams pushing on toward 80. The reopen rate sits right beside it, counting the conversations marked resolved that a customer had to come back and open again, and single digits is healthy there, under 5% ideally. Now the story its self. An agent gets told his first contact number looks low. So he starts marking things resolved a touch earlier than he honestly should, and there is no malice anywhere in this, just a person wanting a number to look nicer. For two months his figure climbs. And right behind it the reopen rate climbs too, all those customers whose tickets closed early coming back in, and the pair of numbers caught what either one on its own would have waved straight through. Track them together or do not track them.
Full resolution time rounds out the pack, the clock from the first message to genuinely done, summed and divided the same way as the reply clock, median quoted. It is allowed to run long on the hard problems. What it should not do is run long on the easy ones, so cut it by issue type before judging anybody on it, a password reset and a payroll data migration have no business sharing an average between them, no two ways about it.
Ticket volume and backlog, the two quiet numbers underneath
Neither of these two measures satisfaction on its own, and both of them push every number above around, so they have earned their short section. Volume is just the count of new conversations arriving per week, watched over time and cut up by topic, and the shape of the thing tells you far more than the size of it ever will. Month five of our support lead's little project, one question about a new payslip layout began arriving forty times a week. The cheapest satisfaction improvement she made all year was getting that one confusing screen fixed. Fewer tickets after that, the very same team, faster replies on everything else, and every metric on this page moved at once without one person working any harder than before.
Backlog is the count of conversations sitting open right now, this minute. It moves everything else, and quietly. Backlog swells, first replies stretch. Stretched replies pull the CSAT down, agents start hurrying, hurried agents close things early, reopens climb, and the whole chain of it began at one number nobody had been watching. So keep a working ceiling on open conversations per agent, and when a week runs above that ceiling, read it as a staffing question and not as an effort question. Asking a team to simply try harder is how one bad month turns its self into a bad quarter.
How to run customer service metrics without building a data project
The machinery for all of this is smaller than the vendors would have you believing, plainly said. One survey question on the back of each closed conversation gives you CSAT or CES, and pick one of the two to begin with rather than both, the response rate is the asset you are protecting and every extra question spends some of it. The clocks ask for no survey at all, they fall out of the timestamps that a proper help desk software was recording anyway, as a side effect of the team simply working. And that, quietly, is the whole argument against running support out of a shared personal login, since nothing over there gets counted unless somebody gives up an afternoon to the copying and the pasting of it. Set your business hours once and the reports do the rest of it.
Three numbers to start, and refuse the rest of them for a quarter, however tempting the dashboard menu gets. CSAT for the feeling, the median first reply time for the speed, the reopen rate for the quality. And when one of the three moves, make the meeting wait, no explanations until somebody has gone and read ten of the actual conversations sitting behind the move. A score tells you where it hurts and nothing further. The why lives in the transcripts, and every hour our support lead gave to reading transcripts repaid her better than any hour she ever gave to arranging charts.
The part where these numbers deserve some suspicion
Every trap on this page has the same shape underneath, and it is worth saying once in general form. The moment a number turns into the thing a person gets judged on, that person starts serving the number, mostly without noticing they have started, and the number drifts off from whatever it was standing in for. Economists have a name for the pattern, Goodhart's law, and honestly a support queue might be the clearest place the law plays out anywhere. Hold the scores loosely then. Show them to the whole team out in the open, be slow about tying bonuses to any single one, and the week a metric improves suspiciously fast, go and read the conversations before congratulating anybody, on account of the improvement sometimes being a person rather than a service.
And a plain word about our own selves, since this is sitting on our blog after all. Maxdesk ships the dashboard for everything above on the free plan, reply clocks, resolution clocks, SLA breaches, the satisfaction tracking, the lot of it, and no plan of ours charges per agent, your whole team comes in at $0 the same as it would at any price. There is a trade, and we would rather you hear it from us over here than go finding it on the pricing page with an eyebrow up. The free workspace shows supporting ads, and it gives you a 3 month view of your history. If your reporting wants a longer memory than a quarter, well, that is precisely what the paid plans exist for. And if a quarter is plenty, every number this article walked through can be measured this very week without paying anybody anything at all.