Satisfaction Metrics

The record a product leaves behind only says what happened. Who signed in, what they clicked, how many days passed before they came back — all of it is stored. Whether the person sitting in front of that screen felt stuck or felt fine is stored nowhere. Teams that watch only the logs get blindsided by that gap.

Pagedeck learned it the hard way. Over one quarter, weekly active users grew from 8,000 to 9,400 and documents created rose along with it, and then eleven paying teams declined to renew. Mia spent days digging through the logs and found exactly one fact: those teams had used the product as much as ever in their final month. Heavy use and genuine liking turned out to be different stories, and no amount of further digging in the logs was going to separate them. Phone calls afterward produced the same answer from seven of the eleven. They kept using it, they kept losing time at the same spot, and eventually that added up to going and looking at something else.

Counting complaints that reach support doesn't close the gap either. The people who write in are either furious or unusually generous with their time, and the majority leave without a word. Twenty-three tickets in a month is not the number of people who hit a problem; it's the number of people annoyed enough to write. That leaves one option: ask.

The catch is that answers can't be pooled if every team asks differently. One team uses five stars, another collects free text, and nothing can be said about whether this quarter improved on the last. Satisfaction metrics are the bundle where the question wording, the scale, and the arithmetic are fixed in advance. Comparability across quarters comes from asking the same sentence, at the same moment, in the same way.

Within the metrics branch of the product field, this bundle is the one that measures by asking. Usage metrics, which count how much the product gets used, activation, the point where a new signup first feels the product's value, and retention, how many of them keep coming back, all come out of what the servers recorded. Satisfaction metrics are the only ones that require a user to answer, which is why they're the only part of product metrics carrying a response rate as a condition.

The four terms underneath split into two groups: three metrics that produce a score by asking, and one framework that decides where those scores belong.

The three metrics differ in what they ask about. One asks about the product as a whole, one about something that just happened, and one about how hard that something was.

1. Net Promoter Score (NPS) — asks how likely you are to recommend the product to a colleague, on a scale of 0 to 10. Answers of 9 and 10 are promoters, 7 and 8 are passives, and 0 through 6 are detractors. The score is the percentage of promoters minus the percentage of detractors, with passives dropping out of the math. Out of 200 responses, 90 promoters, 60 passives, and 50 detractors give 45 percent minus 25 percent, or 20. The result ranges from -100 to 100 and is read as a score, not a percentage.
2. Customer Satisfaction Score (CSAT) — asks whether one specific experience was satisfying, usually on a 1-to-5 scale. The score is the share of people answering 4 or 5. If 320 of 400 respondents answer 4 or above, that's 80 percent. Because it's asked right after a support reply, a payment, or a particular feature, it scores that single moment rather than the product.
3. Customer Effort Score (CES) — asks how much work it took to get something done, usually as agreement with a statement like "it was easy to handle this," rated 1 to 7 and averaged. Effort gets asked about instead of satisfaction because when you want to know whether someone will come back, the memory of a struggle predicts it better than the memory of a pleasant moment.

The framework is the HEART framework, Google's model for splitting user experience into happiness, engagement, adoption, retention, and task success. Each axis moves from a goal, to the signal that would show up in the product if the goal were met, to the metric that counts that signal. When Pagedeck applied it to document sharing, the happiness goal was that invited people don't get lost, the signal was the time between opening an invite link and opening the document, and the metric was a Customer Effort Score asked right afterward. The three metrics above live on the happiness axis; the other four axes are measured from records. A survey score on its own is hard to place, but inside the framework it becomes a number covering one axis out of five.

The arithmetic is the easy part. The hard part is writing down the three things that come before it: when you ask, who you ask, and how many answered. An NPS collected on day one of a trial and one collected after six months of use are different numbers. A score from paying customers only and a score that includes free users are different numbers. Without those three recorded, this quarter can't be compared with the last, and a number that can't be compared produces no decisions in a meeting.

Response rate deserves its own look. The people who answer surveys are mostly the delighted and the angry; the middle quietly skips it. A score from a survey with a 2 percent response rate describes the people who spoke up rather than the user base. If 400 out of 30,000 answered, work out who those 400 are before saying the score moved five points. The way to check is to compare the makeup of the respondents with the makeup of all users: if paying customers are 12 percent of everyone but 40 percent of respondents, the score belongs to paying customers.

Benchmarks for a good number don't come from outside. The most common mistake is reading that the industry average NPS is 31, seeing 24 on your own dashboard, and declaring a problem — a comparison made without knowing when that 31 was collected, from whom, or in what words. Published figures do exist. Subscription software often treats an NPS above 30 as healthy, a CSAT near 80 percent as unremarkable, and a CES above 5 on a 7-point scale as easy going. Use them to locate yourself roughly, not as a pass mark. Three references do the real work: your own previous quarter measured with the same question, the spread between segments, and the link between the score and actual behavior. The third is the most useful. At Pagedeck, teams scoring 0 to 6 renewed 41 percent of the time the following year while teams scoring 9 or 10 renewed 88 percent of the time. Once that link was confirmed, the score stopped being a number that needed interpretation and became an early warning about renewals.

Every survey ends with one free-text box. The number says where to dig and the sentences say why. A score of 24 doesn't tell you what to fix tomorrow, but thirty sentences from the people who scored low will repeat the same word. That's also how the four terms get used together: the whole product tracked quarterly by NPS, a screen you just changed measured on the spot by CSAT, a path that might be rough checked with CES, and the coverage of all three confirmed against the HEART framework.

"NPS dropped from 32 to 27, but only 40 people answered. That's not enough to act on."

The sample is small enough that a five-point move can't be separated from noise. With 40 responses, two detractors appearing or disappearing swings the score by that much.

"We're tracking this redesign on CES, not CSAT. The goal isn't to delight anyone, it's to stop people getting lost."

What a team chooses to measure states what the work is actually for. Satisfaction can stay flat while the effort of getting through a flow drops by half, and for a redesign that is the result worth claiming.

"The average is in the eights and there are twelve comments describing the same problem."

A healthy average can still hide one group stuck in one place. The score says where to look; what to fix is written in the sentences.

Pagedeck runs NPS once a quarter and asks CES right after onboarding finishes. Mia owns the quarterly survey and Noa owns the onboarding one. Noa once polished the wording into a question about whether setup had been difficult, and the next month's average rose from 4.8 to 5.6. The product hadn't changed. Since then the wording stays frozen until the quarter ends, and any change runs as two versions side by side for a month so the difference can be measured first.

Grace uses these numbers in renewal planning. Teams scoring 6 or below go on a list and get contacted two months before their contract expires. The reason behind the score gets heard directly, and it either goes on the roadmap or comes back as an honest answer about what won't happen soon. Either way the conversation happens while there is still a contract to save, which is the part a quarterly report delivered after renewal season cannot do.

The most common error is turning the score into a target. A number invites a team to raise it, especially where every team owes a quarterly goal. But this number is manufactured from user answers, so it moves when the asking moves. A team handed that target starts sending the survey to people likely to be happy, or has support agents request a high score at the end of a chat, or softens the question wording. Six months later the score has climbed from 24 to 38 and renewals haven't budged. Worse, the history is ruined: the method changed, so last year's value can't sit next to this year's, and the rule that used to flag at-risk teams falls over with it. Turn a measuring instrument into a goal and it stops measuring. The thing worth targeting is the problem the score pointed at, and whether that got solved is a question for renewal or return-visit numbers.

The second is believing an industry average locates you. The figures are published and the shape matches, so comparison looks fair. In practice the same metric name covers different objects. One study asks right after signup, another asks people a year in. One includes only paying customers, another counts free users. The wording differs everywhere. When those three vary, a ten-point gap describes survey design rather than product quality. Judging by outside numbers leads to rebuilding something that worked, or relaxing about a real problem because the score beat the average. Treat external figures as rough terrain and make decisions from the trend in your own, measured the same way each time.