Here's the short answer: CSAT is the percentage of respondents who picked "Satisfied" or "Very satisfied." That share is called the Top 2 Box (T2B). It is not the mean. If a 5-point scale produces an average of 4.2, that is the mean satisfaction score — not CSAT.
CSAT (Customer Satisfaction Score) traces back to Oliver (1980) and the expectation-disconfirmation theory, and was standardized through frameworks like Fornell (1992) and the ACSI (American Customer Satisfaction Index). The math itself is simple, but practitioners keep tripping over the number of scale points, the Top 2 Box boundary, and how the share is calculated. This piece walks through the 5 steps to compute CSAT correctly, with the "this is where many go wrong" pitfall called out at each step. How to interpret and operate the metric is covered in the CSAT survey design guide; here we stay focused on the math. It's a companion piece to our recent how to calculate NPS — both are CX-metric how-tos.
Step 1: Ask a satisfaction question on a 5-point scale
CSAT starts with the right question. The standard wording looks like this:
"How satisfied were you with this service?" 1 (Very dissatisfied) / 2 (Dissatisfied) / 3 (Neither) / 4 (Satisfied) / 5 (Very satisfied) — 5-point scale
A 5-point scale is the CSAT standard. Some practitioners use 7- or 10-point versions, but 5-point is the most widely used and makes Top 2 Box the easiest to interpret.
This is where many go wrong: asking on a 3-point ("Dissatisfied / Neutral / Satisfied") or 2-point ("Satisfied / Not satisfied") scale instead of the symmetric 5-point "Very dissatisfied to Very satisfied." Below 4 points, Top 2 Box stops working — you lose the gradation that distinguishes "satisfied" from "very satisfied." For scale design, see the Likert scale design guide. And always make the question's subject ("about what" — the transaction, support, product, etc.) explicit.
Step 2: Classify responses into the Top 2 Box (satisfied group)
Bucket the responses by score. The heart of CSAT is treating the top two scale points as your satisfied group:
- Top 2 Box (T2B / satisfied): respondents who picked 4 (Satisfied) + 5 (Very satisfied) on a 5-point scale
- Middle (neutral): 3 (Neither)
- Bottom 2 Box (B2B / dissatisfied): 1 (Very dissatisfied) + 2 (Dissatisfied)
The CSAT score uses only the Top 2 Box count. On a 10-point scale, the standard is Top 2 Box = 9 + 10 (neutral is 7-8, dissatisfied is 1-6).
This is where many go wrong: confusing "Top Box (5 only)" with "Top 2 Box (4+5)." They are different metrics — Top Box captures only "Very satisfied," while Top 2 Box includes everyone "Satisfied or better." Pick one — Top Box or T2B — and stick with it forever. Switching mid-stream breaks your time-series comparisons. And under no circumstances should "3 (Neither)" be folded into the Top 2 Box.
Step 3: Divide by the total to get a percentage
Divide the Top 2 Box count by the total number of respondents to get your CSAT score.
Example: out of 200 respondents, 40 picked 5 ("Very satisfied"), 80 picked 4 ("Satisfied"), 50 picked 3 ("Neither"), 20 picked 2 ("Dissatisfied"), and 10 picked 1 ("Very dissatisfied"):
- Top 2 Box = 40 + 80 = 120 people
- CSAT = 120 ÷ 200 × 100 = 60%
This is where many go wrong: dropping neutrals from the denominator — computing it as "Top 2 Box + Bottom 2 Box" instead. The CSAT denominator is always the full respondent count (neutrals included). Excluding neutrals inflates the number dramatically and invalidates comparisons with other companies or other periods. As with NPS, fix your denominator convention and never change it — that's the iron rule.
Step 4: Compare against industry and your own baseline
CSAT ranges from 0% to 100% (100% if everyone is in the Top 2 Box, 0% if nobody is). The unit is "%", and it's written like "CSAT 60%."
Calculating in Excel
In practice you can knock this out in Excel or Sheets. Assuming responses are stacked in a column (say column B):
- Top 2 Box count:
=COUNTIF(B:B,">=4") - Total responses:
=COUNT(B:B) - CSAT:
=COUNTIF(B:B,">=4") / COUNT(B:B) * 100
For a 10-point scale, swap in ">=9" — same formula. For end-to-end CSV import and aggregation steps, see the survey Excel aggregation guide.
This is where many go wrong: confusing "mean" with "CSAT." A calculation like "the 5-point average is 4.2 so CSAT is 84%" is completely wrong. Mean and Top 2 Box measure different things — it is entirely possible to have a high mean but a low Top 2 Box (everyone clustering at 3-4). CSAT is a "share," not an "average" — never lose sight of that.
Step 5: Read the score against industry benchmarks
So how do you read the resulting CSAT? Don't get emotional over the absolute number — that's the practitioner's rule.
The rough benchmarks vendors publish:
- Below 60%: needs improvement
- 60-75%: average
- 75-85%: good
- 85% and above: excellent
These come from English-speaking-market vendors, and in the Japanese market a central-tendency bias (respondents cluster toward the middle) pushes absolute scores lower. Skip cross-country absolute comparisons; anchor on time-series movement and relative comparison against direct competitors.
This is where many go wrong: judging "good or bad" from a single absolute measurement. The value of CSAT lies in continuous tracking of the change over time, plus diagnosing why low-scorers scored low. Scoring interpretation, benchmarks, and operational guidance are organized in the CSAT survey design guide — read it together with this piece once you've gotten the math down.
Editor's take — 3 things that actually matter when calculating CSAT
From continuously tracking industry cases and practitioner voices, here are the 3 things that genuinely matter when computing CSAT.
1. Never, ever confuse "mean" with "CSAT"
This is the most common accident. "The mean is 4.2 so CSAT is 84%" is just plain wrong. CSAT is a share (%), not an average. If you report a mean as "CSAT" in an executive meeting, every industry benchmark and competitive comparison falls apart. As with NPS (% promoters minus % detractors), getting the formula wrong means the metric stops working as a metric.
2. Always report N (sample size) alongside the score
Because CSAT is a share, it swings wildly when N is small. "CSAT 70%" with N=30 can easily move ±10 points with a handful of responses. Always report it as "CSAT 70% (N=30)" — show the sample size every time, and treat low-N scores as directional only. For thinking about required sample sizes, see how to determine survey sample size.
3. Always collect the reasons from low-scorers
The number itself matters less than what the Bottom 2 Box (dissatisfied) respondents are unhappy about — the open-ended comments are what drive actual improvement. Don't stop at the score; add branching so only respondents who picked 3 or below get asked for their reasons, and harvest the improvement signals. For design details, see the open-ended question design guide and the survey branching logic complete guide.
Summary — the 5 steps to calculate CSAT
- Ask on a 5-point scale — symmetric "Very dissatisfied to Very satisfied." Below 4 points, Top 2 Box stops working
- Classify into Top 2 Box — 4+5 on a 5-point scale, 9+10 on a 10-point scale. Don't confuse with Top Box (5 only)
- Divide by total respondents — denominator includes neutrals. Excluding them inflates the score
- CSAT = Top 2 Box ÷ Total × 100 — result is a %. One COUNTIF in Excel does it
- Read it as a time series and relatively — avoid cross-country absolute comparisons. Pair with the dedicated guide for interpretation
Calculating CSAT is not hard once you anchor on "Top 2 Box share." Where it goes wrong is always one of three things: confusing it with the mean, confusing T2B with Top Box, or getting the denominator wrong. Avoid those, and anyone can compute it correctly. Once you can compute it, the next step is reading the score, benchmarking it, and operationalizing low-score follow-up. For the full landscape of CX metrics, see the CX metrics comparison guide.
If you want to build and run a CSAT survey, try Kicue, a free survey tool. From a single account you can create the 5-point satisfaction question, add branching so only low-scorers get asked for their reasons, and export a respondent-ID-tagged CSV — everything you need to launch a CSAT survey and prepare for aggregation. (For Top 2 Box itself, the reliable workflow is to compute the exported CSV with Excel COUNTIF.)
References (2)
- Oliver, R. L. (1980). A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions. Journal of Marketing Research, 17(4), 460-469.
- Fornell, C. (1992). A National Customer Satisfaction Barometer: The Swedish Experience. Journal of Marketing, 56(1), 6-21.