What is CSAT: the customer satisfaction score explained
A customer satisfaction score (CSAT) is a survey metric reporting the percentage of respondents who gave a positive answer when asked how satisfied they were with a specific interaction, usually on a five-point scale straight after the event. It measures a moment, not a relationship, and only among the people who replied.
What does CSAT actually measure?
One transaction, felt at one point in time. A warranty decision, a parts delivery, a support call, a site visit. The narrowness is the point: ask while the customer still remembers, and the answer traces back to a case, an agent and a process.
It does not measure loyalty, effort, or whether the problem is fixed. A customer can be entirely satisfied with how politely you told them the part is on back order.
The survey mechanics that decide the number
The standard form is a single question with a five-point scale, from very dissatisfied to very satisfied, sent within minutes or hours of the interaction closing. Almost every design choice around that question moves the score, which is why comparing CSAT across two companies is close to meaningless.
- Scale length. Five, seven and ten points produce different distributions from identical customers. Changing it breaks your own trend line.
- What counts as positive. Most teams count the top two boxes. Counting only the top box gives a lower, more sensitive number.
- Timing. At case closure you measure the interaction. A week later you measure whether the fix held. Both are valid and they are not the same metric.
- Trigger rules. Surveying only resolved cases, or skipping cases escalated to a manager, quietly deletes your unhappiest respondents.
- Wording. "How satisfied were you with your engineer?" and "How satisfied were you with the outcome?" score differently on the same visit.
How is CSAT calculated?
Divide the number of positive responses by the number of responses received, then multiply by 100.
Note the denominator: responses, not customers surveyed. That single choice is the source of most of the metric's trouble, and it belongs on every report you circulate.
A worked example, and the number underneath it
A warranty desk sends 1,000 surveys after closing claims. 180 come back. Of those, 144 chose satisfied or very satisfied, so CSAT is 144 divided by 180, which is 80%.
Now read it the other way. Of 1,000 customers, 144 are confirmed happy, 36 are confirmed unhappy, and 820 said nothing at all. If even a fifth of the silent group were dissatisfied enough to disengage, the true picture sits well below 80%. The score is honest about its respondents and silent about everyone else.
Why response bias makes CSAT flattering
Survey response is not random, and the ways it skews mostly push in one direction.
- Non-response bias. People who feel strongly reply more than people who feel nothing, but customers who have given up on you often just leave.
- Selection at the trigger. If the survey only fires on closed or successfully resolved cases, the worst outcomes are structurally excluded.
- Social desirability. A survey signed by the engineer who just spent an hour in the customer's kitchen collects politeness as much as judgement.
- Recency and halo. A courteous closing call lifts the score on a case that took three weeks longer than it should have.
- Survey fatigue. High-contact customers are surveyed most and answer least, so the accounts you know best are the ones you measure worst.
Two habits help without pretending the problem is solved. Publish the response rate next to the score every time. And track movement rather than level, because bias that stays roughly constant still lets a trend mean something.
How CSAT differs from CES and NPS
They are usually presented as rivals. They answer three different questions, and after-sales teams tend to need more than one.
- CSAT asks how satisfied you were with this interaction. Transactional, immediate, easy to attribute to a case or a person, and easy to lift with courtesy alone.
- CES, the customer effort score, asks how much effort the customer had to put in to get their issue resolved. It was introduced by Matthew Dixon, Karen Freeman and Nick Toman in "Stop Trying to Delight Your Customers" in the July to August 2010 Harvard Business Review, which argued that reducing the work a customer has to do predicts loyalty better than trying to exceed their expectations. For warranty and parts, effort is often the sharper diagnostic, because it exposes the four calls it took to identify one component.
- NPS, the net promoter score, asks how likely you are to recommend the company, and was proposed by Frederick Reichheld in "The One Number You Need to Grow" in the December 2003 Harvard Business Review. It measures the relationship, not the interaction, so a single support case rarely moves it and it cannot tell you which process is broken.
An honest reading: CSAT is the easiest to run and the easiest to flatter, CES is the most useful for finding friction in a service process, and NPS is a relationship indicator that gets misused as an operational one. For outside context, the American Customer Satisfaction Index put the US national score at 76.1 out of 100 in the second quarter of 2026, down from 76.7 in the first. A self-reported 90% on an internal dashboard is measuring something much narrower.
What CSAT is good for
Comparing you to yourself. Segment it by process, not just by agent: claims closed on first contact against those needing a return, cases where spare parts identification was right first time against those where the wrong part shipped, visits resolved on the day against repeat visits.
Read alongside operational data it gets sharper still. A falling score next to a rising RMA rate points at a diagnosis problem. A falling score next to a stable first-time fix rate points at communication. Neither conclusion is available from the score alone, which is the whole case for never reporting it alone.