CSAT is short for customer satisfaction score, and the whole of it is one question put to somebody at the end of a conversation. How satisfied were you with the help, answered on a scale of one to five. Count the fours and the fives, divide by everybody who answered at all, multiply what you get by a hundred, and the figure left in your hand is your CSAT. There is no more arithmetic in the whole of it than that, and the sum was never the difficulty. The difficulty is that the number gets read as a mark out of a hundred for your service, when the thing it measures is a good deal narrower, and narrow in a direction that goes quietly against whoever on your desk is straightest with people.
The word itself gets used two ways in the one meeting with nobody stopping to separate them out. One person says CSAT and means the survey, that little one line thing going out after a ticket closes. Somebody across the table has the report in mind instead, the monthly figure that comes back off all those answers once they are added up. Both of them are right enough, truth be told. It is the second one that ends up printed beside somebody’s name at review time, and the trouble starts there.
The ranking that came out backwards
Roisin sells spare parts for washing machines and tumble dryers out of a unit outside Athlone, a business her father started and she took on in her thirties, and most of the trade comes in by mail now from people trying to keep a nine year old machine going one more winter. Two people answer that mail alongside her, Senan four years at it now and Clodagh in since the summer before last, and the three of them between them will clear most of a day’s post before the dinner.
Last October she was working out a Christmas bonus. Small money, a few hundred each, and she wanted it split by something better than a guess, so she went and opened the satisfaction report on her desk software for the first time in months and sorted the thing by person her own self. Clodagh came out at ninety one. Senan came out at seventy four.
She sat looking at that for a good while. Because she knew, the way you know things about your own place without needing them proved, that Senan was the better of the two on the mail, he was the one she handed anything awkward to, he was the one people wrote back to by name months afterwards, and seventeen points is not the sort of gap you can shrug at either, it is the sort you act on, and had she acted on it she would have gone and rewarded the wrong person and told the right one to take himself away and improve.
So she went reading instead of acting. An hour of it or more, down through the ones and twos Senan had collected, and every last one of them had the same thing sitting underneath. The part is gone. Not made any more, no substitute worth selling you, your machine is nine years old and this is the end of the road for it, and there is no soft way round the rest of it. He said it plainly and he said it early, which is exactly what Roisin would want said in her name, and nearly everybody he said it to reached for a one or a two on the way out the door.
What the rating lands on, and what it misses
Nothing was the matter with Clodagh either, and that part took longer to see. Clodagh was the newer of the two, so the easy mail went to her by arrangement, anything with the part sitting on a shelf forty feet away. She was answering the yeses, near enough the whole of them, while anything with a no waiting in it went to Senan. Neither of them had gone near a single figure, nor would either of them have known how to, and the ranking still came out backwards on account of the routing rather than anything either of them had done with the work.
The reason for it sits in the timing of the survey its self. It goes out at the end, hooked onto the closing of the conversation, and it gets answered in about four seconds by a person who has one hand on the door already. Four seconds is not long. What anybody feels in those four seconds is mostly about whether they got the thing they came for. The care taken over the reply is in there somewhere, and the speed of it, and the manners of it, only the outcome sits on top of the lot and drowns out the rest of it.
Take two conversations handled the very same way, both answered inside the hour, both written in plain warm English by somebody who read the question properly before starting. One of them finishes with the part going in the post on Tuesday. In the second, nothing will fit that drum any more and the woman is being told her machine is past saving. The second customer is not rating the help at all when she reaches for a two. She is rating Tuesday, or the want of a Tuesday, and there is not a box anywhere on that survey for telling those two things apart, nor was there ever going to be.
So a good share of the time CSAT is really a product number that has been handed to the support desk to carry. When the thing you sell breaks, or runs out, or will not do what people keep writing in and asking it to do, the desk carries the score for the whole of it, and nobody further upstream ever sees their own fingerprints anywhere on the number.
What counts as a good csat score
No answer to that travels properly between companies, and anybody handing you one figure to aim at is selling something. All the same, a bit of rough furniture helps. Business to business software support mostly sits somewhere in the seventy five to eighty five band. Below that, go looking. Teams do hold above ninety and some of them honestly, though a long run up there usually says more about a narrow spread of questions coming in the door than about a desk doing something the rest of us never thought of.
The other half of reading a score is the count standing behind it. Five to fifteen percent of people answer these things in a normal month, so the number is built off a slice, and the slice was never a random one. The two crowds who bother tapping are the delighted and the annoyed, and the great quiet mass of ordinary satisfied customers in the middle of it say nothing at all. Eleven answers in a thin month cannot be set beside two hundred from a busy one and read as a trend, mind you, and small desks do exactly that and give themselves a fright over nothing.
So the comparison worth having is your own desk against your own last quarter, with roughly the same mix of questions arriving. Shift the mix and the figure moves without one person’s service having changed by an inch. That is most of what this article is about, and worth stopping over for a minute before you go any further. The score was never meant to be read on its own in the first place, no two ways about it, it belongs in a small set of numbers that keep each other honest, which we went through with the formulas and the traps over in customer satisfaction metrics.
The month it climbed and nothing at all had improved
February proved the point to Roisin without her asking it to. The score across the desk went up to eighty eight and sat there the month, and she had done nothing whatever to earn a point of it. A supplier in Italy had restarted a line of drum bearings that had been out of production two years, so a whole class of mail that used to end in a no ended in a yes instead. Same two people, the very same writing, and fourteen points on the score.
There is a second way a score climbs and this one costs you something, so it is worth knowing the smell of it, on account of hope rating better than an answer does nearly every time. A reply saying leave it with me and I will see what I can find ends the conversation on a warm note, and the customer taps a four on the strength of the warmth, and the bad news lands eleven days later in a thread nobody surveys or after the person has quietly given up and gone off elsewhere. Nobody sets out to do this. Saying no early feels expensive, plain and simple, and the score agrees with that feeling, so a desk under pressure about its numbers drifts toward the softer ending without a single meeting ever being held about it.
Which turns the whole business on its head. A CSAT that slips a few points after you have asked a team to be straighter with people, and sooner, may well be reporting back that they went and did it. That is not a comfortable sentence to put in front of a founder in a Monday meeting. It is true all the same, and it is the reason we would not hang money on this number head by head.
Roisin split the bonus evenly in the end and told the pair of them why she had, which she says was the easiest awkward conversation of her working life.
Cutting the score by what the person came in for
What she set up in January took about twenty minutes and she has kept it running since. Every conversation that closes gets one word put on it, yes if the person got the thing they were after and no if they did not. That is the entire system, there is no third box and no scoring of degrees, and the two of them fill it in them selves as they go, without thinking hard about it at all.
Then the report gets read twice over. The yes conversations came out at ninety three across both people. The no conversations came out at sixty one. Thirty two points between the two piles. And once it was cut that way, Senan and Clodagh landed within two points of one another on the yes pile, while on the no pile Senan came out four points ahead of her, which is what Roisin had believed about him all along with nothing to lean on but her own judgement.
The fairness of it was not the useful part though. What mattered was the no pile turning into a list. Something like two out of every five noes in that first quarter were the one situation over and over, an obsolete part with a substitute that would in fact do the job, known to Senan, written down nowhere, and never mentioned unless he happened to be the one who picked the mail up. She had him write up nine of those substitutes across two afternoons. They went into a knowledge base so either of them could drop the link into a reply without leaving the reply, and a decent share of what used to be a flat no went and turned into an awkward yes.
The headline score did not leap. It went from seventy nine to eighty three across about five months, which is nothing anybody would put in a case study. The mail got easier all the same, and what noes were left in the pile were real noes rather than gaps in what anybody happened to know.
What we sell, and what is traded against it
We may as well do the selling here in the middle rather than slipping it in at the end when your guard is down. What we make is Maxdesk, a shared inbox with a help desk sat over it, and the only bit of that worth your attention today is the satisfaction tracking, which we put on every plan we do, the free one as well. Ratings come to rest beside the conversation that earned them, so reading the thread behind a poor one costs you a click and nothing else, no exporting, no spreadsheet, no afternoon gone on the joining up of it. And the bill attaches to the workspace its self, never to the number of heads sat round it, so taking on a second or a third person for your heavy month puts nothing extra on the bill at all.
Now the trade on that free plan, before anybody signs up on the strength of the paragraph above. Supporting ads appear in the workspace, the very one your own staff have open in front of them the whole day. Mail going out carries a line of Maxdesk branding along the bottom. Your history rolls back three months and not an hour further. For twenty dollars a month the ads go and the branding becomes your own, with the history opening out to a full year, while ninety nine adds the AI layer and stretches the memory to two years. That is the pricing, the whole of it, and we will stop selling now.
The parts of this we cannot do for you
The yes and no cut, which is the one thing in this article that made Roisin’s number readable, is manual. You put that word on the conversation your own self, or you set up a tag and remember to use it. We do not work out for you whether a customer got what they came for, and I have not seen the tool that can, on account of the answer living in the meaning of the thread rather than in anything countable about it. If somebody sells you that, ask hard questions before you believe them.
The three months of history on the free plan bites harder here than on any other page we have written. A satisfaction score is a thing you read across a year and against your own season, and a quarter of memory will not let you do it. That one goes down as a genuine cost against us for anybody whose trade has a busy end and a quiet end to it, and it is the honest reason the paid plan exists, rather than any clever feature we could put in a list.
We are email only as well. No chat, no phone line, no portal for customers to log into. If half your conversations arrive over a telephone then the CSAT built out of your inbox goes by the name of the whole desk while having only ever seen the half of it, which is worse in some ways than carrying no number at all, on account of it looking finished. It is a real hole. We do not close it, and I would rather tell you here than have you find out on us in month two.
The biggest one of the lot has no software in it anywhere. What actually moved the number out in Athlone was nine substitute parts getting written down. No support tool was going to find those, ours no more than anybody else’s, on account of them living in Senan’s head and nowhere else, and the way into a man’s head is to go and ask him rather than to go and buy software. And whether people stay with you after a straight no is a separate question again, a harder one, with more of it in what we wrote about churn rate analysis.
Twenty ratings and four minutes
Do not start by changing the survey. That is what everybody reaches for first, a second question added on, or a move to five stars, or the little faces instead of the numbers, and not one of those goes anywhere near the thing this article is about.
Open your last twenty rated conversations instead. Beside each one put a single word, yes or no, on whether the person got the thing they wrote in asking for. Then work the score out twice, once across the yeses and once across the noes. Four minutes with a pen, and the back of an envelope will hold it.
If the two come out close together, within five or six points of each other, then your CSAT is doing the job everybody assumes it does and you can go and use it with a clear conscience. A gap of fifteen or twenty points says something else entirely. The score you have been reporting all year is mostly a report on your answer mix, and any target hung on it lands on your suppliers and on your product rather than on the people answering the mail. Twenty is enough to tell you which of the two you are living in. Do the hundred later, if the twenty surprises you.
Roisin finished the year at eighty three. It had read ninety one for one of her two people the October before, and seventy four for the other, and neither of those numbers was true about either person. She would tell you the report only started being any use to her once she stopped trusting the single figure sitting at the top of it. There is a quarter’s work in her no pile yet, which does not bother her in the least, since at least it is a pile she can see now.