Customer Satisfaction Score

For a long time the only way to learn how a product felt was to read its logs. How many people arrived, which button they pressed, how many seconds they stayed — all of that gets recorded. Whether the person walked away relieved or annoyed gets recorded nowhere. Logs say what happened, never how it landed.

That gap showed up in meetings in a recognizable shape. A team changes the checkout screen, the share of people who complete a payment rises three percent, and support tickets for the same week rise forty percent. One signal says better, the other says worse, and two hours of argument produce no conclusion. Asking the people who were there would have taken thirty seconds, and nobody had asked.

The customer satisfaction score (CSAT) is that question cut down to a single item and attached to the moment itself. Instead of running a separate survey, you put one question on the screen where the support conversation ended or the payment went through. Asking before the memory cools keeps the answer tied to that one moment, and a single item is cheap enough to answer. Narrowing the subject from the whole product down to one scene decides everything else about how this metric behaves.

Divide the number of satisfied responses by the number of responses and write it as a percentage. If 400 people answered and 320 of them picked a satisfied option, that's 80 percent.

The division is the easy part. Three decisions in front of it do the real work.

1. Fix the moment you ask about. Right after a support conversation closes, right after a payment goes through, right after someone uses a particular feature for the first time. Change the moment and you have a different metric, so once it's set, leave it.
2. Fix the scale and the cutoff. The usual setup is a 1-to-5 scale with 4 and 5 counted as satisfied. A 3 does not count. Whether you use five points or seven, and where the cutoff sits, is the team's call — but moving the cutoff down to include 3 even once lifts the score by ten-odd points on its own.
3. Record how many answered and how many were asked. Next to that 80 percent, write 400 responses and an 8 percent response rate. Those three numbers have to travel together or next month's value has nothing to be compared against.

Reporting an average instead is possible, but the ratio is safer. An average lets one person who scored a 5 erase one person who scored a 1. A group of 100 fives and 100 ones and a group of 200 threes both come out at 3.0, even though the first group contains something badly broken and the second is merely flat. As a ratio, the first reads 50 percent and the second reads 0 percent.

Don't start by comparing against an industry average. Subscription software tends to treat something near 80 percent as unremarkable, but if you don't know which moment that figure was collected at or how the question was worded, you can't call your own 76 percent low. A score collected right after checkout and a score collected right after an outage ticket were never on the same footing.

Three things make usable baselines: your own value from last month at the same moment with the same wording, the gap that appears when you split respondents into groups, and whether the score connects to what people actually do next. The third one earns its keep. Once you confirm that people who scored the checkout low churn at a higher rate the following month, the number stops being something that needs interpretation and becomes an early warning.

Read the response rate next to the score, always. People who answer surveys skew toward the delighted and the furious; the middle goes quiet. An 89 percent built on a 2 percent response rate describes the people who felt like answering, not your users. Comparing the makeup of respondents against the makeup of all users exposes this quickly. If paid customers are 12 percent of your base but 40 percent of your respondents, the score leans toward paid customers.

Keep a free-text box under the question. A number tells you where to dig; sentences tell you why. A 72 percent by itself doesn't decide what to fix tomorrow, but reading thirty comments from the people who scored low makes the repeated word obvious.

"That checkout screen we changed last week — satisfaction right after it dropped from 88 to 72."

The change probably made the experience worse, so let's decide whether to roll it back.

"It's 89 percent this month, but the response rate is 2 percent. I'd rather not report that number."

Too few people answered for the result to stand in for everyone.

"That's measured right after a support conversation, so don't read it as a score for the whole product."

Correcting how much of the experience the metric actually covers.

Pagedeck runs this score in two places. One is support. When a conversation closes, a single question appears, and Sofia reads through the comments attached to low scores every week. If the same complaint shows up more than three times, it goes to the product team. The other is feature releases. When Mia ships a new screen, the question is shown only to people who used that screen, and it's switched off after two weeks.

The two scores stay separate. Merged into one number, a rise could mean support improved or the new screen landed well, and a number you can't act on is not worth collecting.

The common mistake is reading the score as a report card for the product. The name invites it. What was actually asked about is one moment, and that moment may be a small share of the experience. A product with 92 percent satisfaction on support replies can still churn heavily — the replies were kind, but there was far too much to ask about in the first place. Report this number as the product's headline and you spend a quarter polishing reply templates while whatever generates the tickets stays where it is.

Counting 3s as satisfied is the other one. Somebody who picked 3 didn't say they were unhappy, so excluding them feels harsh. But on a five-point scale a 3 is closer to undecided about coming back, and in the following month those people behave more like the 1s and 2s than like the 4s and 5s. Including them jumps the score by ten-odd points at once, and from then on it can't be compared with last quarter. A number that moved because the formula changed is not an improvement.

Of the three, the net promoter score is the closest relative: both are surveys and both report a ratio. What separates them is timing — one asks about something that just happened, the other about an impression built up over months. The customer effort score (CES), asked while the experience is still fresh, sits on the same side of that line.