Customer Satisfaction (CSAT) Surveys: A Complete Guide with Worked Examples
What CSAT Actually Measures (and What It Doesn't)
CSAT — Customer Satisfaction Score — measures how satisfied someone is with a specific, recent interaction: a purchase, a support conversation, an onboarding call, a delivery. It is deliberately narrow and deliberately immediate. That narrowness is its strength: when a CSAT score moves, you usually know which touchpoint moved it.
What CSAT does not measure is loyalty. A customer can rate every support interaction five out of five and still churn next month because a competitor is cheaper, or rate one delivery poorly and stay for years. Treating CSAT as a loyalty metric leads teams to celebrate scores while customers quietly leave. Use it as a thermometer for touchpoints, not a forecast of retention.
The Standard Question and Scale
The canonical CSAT question is short: 'How satisfied were you with [the interaction]?' answered on a five-point scale from 'Very dissatisfied' to 'Very satisfied'. The label wording matters more than the number of points — respondents anchor to the words, and 'Very satisfied' sets a higher bar than an unlabeled 5. Label every point, not just the ends.
Whatever scale you choose, freeze it. The moment you switch from five points to seven, or reword the labels, your trend line breaks and every before/after comparison becomes an argument about methodology instead of a fact about customers. Consistency over time is worth more than any single improvement to the question.
Calculating CSAT: A Worked Example
CSAT is reported as the percentage of respondents who chose one of the top two answers. Say 200 customers answer after a support interaction: 62 choose Very satisfied, 58 Satisfied, 40 Neutral, 25 Dissatisfied, and 15 Very dissatisfied. The satisfied count is 62 + 58 = 120, so CSAT = 120 / 200 = 60%.
Notice what the percentage conveniently ignores: the 40 dissatisfied and very dissatisfied customers. The mean score here is 3.64 out of 5, which sounds gentler than '20% of customers left unhappy.' Report the top-two-box percentage as your headline, but read the full distribution before deciding anything. If you want to check the arithmetic on your own data, the free CSAT calculator on the SurveyExtreme tools page does the top-two-box math for you.
CSAT vs NPS vs CES: A Decision Rule, Not a Debate
The three metrics answer different questions. CSAT asks 'how was this interaction?' — use it per touchpoint. NPS (Net Promoter Score) asks 'how do you feel about us overall?' — use it once or twice a year for the whole relationship. CES (Customer Effort Score) asks 'how hard was it to get this done?' — use it after service and support flows, where effort predicts repeat behavior better than delight does.
The decision rule: pick the metric whose question you would actually ask a customer face to face in that moment. Right after a support ticket closes, 'how hard was that?' is natural — that is CES territory. After onboarding completes, 'how satisfied are you with how that went?' fits — CSAT. 'Would you recommend us?' the day after one delivery is premature — that is a relationship question.
Whatever you choose, do not stack all three on the same customer in the same week. Metric sprawl exhausts respondents, and the falling response rate quietly biases every score you collect afterward.
When to Ask: Timing Windows That Preserve Signal
Ask within hours of the interaction, while it is still concrete. A customer rating this morning's support conversation is recalling actual events; a customer rating it two weeks later is recalling a general impression of your brand. The longer the delay, the more your CSAT converges toward a fuzzy relationship score you could have gotten from NPS.
Delay also changes who answers. As the memory fades, the people who still bother to respond skew toward those with strong feelings — often the angry. An immediate ask samples the satisfied middle; a late ask samples the extremes. If your CSAT dropped after you moved the send time, check the timing before blaming the product.
Where CSAT Programs Go Wrong
The most common failure is chasing the score instead of the cause. Once a team is bonused on CSAT, agents start saying 'you'll get a survey — anything less than a 5 hurts me.' The score rises, the underlying experience does not, and leadership makes decisions on corrupted data. If you tie compensation to CSAT, audit for begging and gaming from day one.
The second failure is surveying every interaction. A customer who contacts support three times in a week and receives three surveys learns to ignore all of them. Throttle to one ask per customer per period, and accept a sampled view; a 30% response rate on a sample beats a 6% response rate on the census of the annoyed.
The third failure is ignoring who did not answer. If only one in five customers responds, your CSAT describes the kind of customer who answers surveys. Track the response rate next to the score, and treat a falling response rate as a data-quality alarm, not a footnote.
Read the Distribution, Not Just the Score
Two teams can post the same CSAT and need opposite interventions. Team A's 200 responses: 90 Very satisfied, 30 Satisfied, 10 Neutral, 20 Dissatisfied, 50 Very dissatisfied — top-two-box 120/200 = 60%. Team B's 200 responses: 40 Very satisfied, 80 Satisfied, 60 Neutral, 15 Dissatisfied, 5 Very dissatisfied — also exactly 60%.
Team A is polarized: it delights half its customers and infuriates a quarter of them. Something in its process works brilliantly and something fails badly — find the fork. Team B has almost no angry customers and a large mediocre middle: nothing is broken, nothing is memorable. Team A needs firefighting; Team B needs raising the ceiling. The single number 60% cannot tell you which team you are.
Segment Before You Conclude
An overall CSAT is an average over channels, products, regions, and staff — and averages hide forks. Slice the score by the dimensions you can act on: support channel, product line, plan tier, individual touchpoint. A flat 74% overall can decompose into 88% for chat and 55% for phone, which turns a vague 'improve satisfaction' goal into a specific phone-queue problem.
In SurveyExtreme, the answer segment filter does this directly on the results dashboard: filter all responses by how people answered any other question — channel used, plan type, first contact or repeat — and watch the satisfaction distribution recompute per segment. One caution: keep an eye on segment sizes. A segment of 12 responses swinging 20 points is noise, not narrative; wait for the base to grow before reorganizing a team over it.
Turn Open-Text Follow-Ups into Action
Always pair the score with one optional open question: 'What is the main reason for your score?' The score tells you where to look; the verbatims tell you what to fix. Keep it optional — forcing an essay after a five-second rating costs completions — and keep it singular. One good open question outperforms three lazy ones.
Reading a thousand verbatims does not scale, which is where automated help earns its keep. SurveyExtreme's AI sentiment analysis classifies open-text answers as positive, neutral, or negative alongside each question's chart, and the AI summary condenses recurring themes from text responses. Use the machine pass to find the themes, then read twenty raw verbatims per theme yourself — the specifics that make a fix obvious live in the raw text, not the summary.
Build the Habit: Baselines, Cadence, and Trends
A CSAT score means little in isolation; the trend is the product. Establish your own baseline over a few periods before reacting to any single reading, and mind seasonality — support satisfaction routinely dips in launch weeks and holiday rushes for reasons that have nothing to do with your team. The response trend view matters more than any individual month.
Resist anchoring on published industry benchmarks. Methodologies, scales, timing, and populations differ enough that a benchmark comparison is rarely apples to apples. The comparison that matters is you versus you last quarter, on a frozen question, at a stable response rate — that number you can defend in any meeting, and act on.