How to Calculate CSAT Score: Formula, Examples, and Benchmarks

CSAT % = (number of satisfied responses ÷ total number of responses) × 100. On a 5-point scale, "satisfied" responses are the ratings of 4 and 5. The result is a percentage between 0 and 100.

That single equation is the whole method. The only judgement call is which ratings you count as "satisfied," and the standard convention — used by Zendesk, Qualtrics, and most CX teams — is the top-two box: on a 1–5 scale, a 4 ("satisfied") and a 5 ("very satisfied") both count. Neutral (3) and everything below is excluded from the numerator but stays in the denominator.

Written out step by step:

1. Count every survey response you received (this is your denominator).
2. Count how many of those responses were a 4 or a 5 (this is your numerator).
3. Divide the second number by the first.
4. Multiply by 100 to turn the ratio into a percentage.

Zendesk's CSAT guide describes the same top-two-box approach on both 5-point and 10-point Likert scales, where a 9 or 10 out of 10 maps to "very satisfied." The scale you pick does not change the formula — it only changes which numbers land in the "satisfied" bucket.

Suppose you send a post-resolution survey and get 200 responses: 110 rated 5, 50 rated 4, 20 rated 3, 12 rated 2, and 8 rated 1. Your satisfied responses are 110 + 50 = 160. CSAT = (160 ÷ 200) × 100 = 80%.

Here is the same calculation laid out in a table so you can see exactly what enters the numerator:

160 satisfied divided by 200 total is 0.80, or 80% once you multiply by 100. If you would rather not do the arithmetic by hand, this CSAT calculator takes your response counts and returns the percentage plus a benchmark comparison. The point of running through it manually once is to see that the "neutral" 3s quietly drag your score down — they never count as satisfied, but they always count in the total.

Use the percentage (top-two-box) method for reporting and benchmarking, because nearly every published benchmark is expressed that way. An average-score method — adding all ratings and dividing by the number of responses — produces a 1-to-5 figure that is not comparable to industry CSAT numbers.

There are two legitimate ways to summarise CSAT responses, and mixing them up is the most common calculation error:

• Percentage / top-box method: (satisfied responses ÷ total) × 100. Output is 0–100%. This is what ACSI, Zendesk, and Qualtrics benchmarks use.
• Composite average method: sum of all rating values ÷ number of responses. Output is a 1.0–5.0 figure. Useful internally for tracking small shifts, but meaningless against a "77% national average."

Run the earlier example through the average method: (110×5 + 50×4 + 20×3 + 12×2 + 8×1) ÷ 200 = 848 ÷ 200 = 4.24 out of 5. That 4.24 and the 80% describe the same data, but you cannot compare 4.24 to a benchmark expressed as a percentage. Pick one method, document it, and never switch mid-quarter — a silent change from top-two-box to top-one-box (counting only 5s) can drop a healthy score by 20 points overnight.

A CSAT score between 75% and 85% is generally considered good, and above 90% is exemplary, according to Zendesk's CSAT guidance. As a reference point, the American Customer Satisfaction Index recorded a US cross-industry average of 76.7 on its 100-point scale in Q1 2026, so anything comfortably above the high-70s is doing well.

"Good" is relative to your industry, not the global average. The ACSI has published national scores every quarter since 1994, and the number has sat near 77 for over a decade — 76.9 across 2025 and 76.7 in Q1 2026 (ACSI, 2026). Individual sectors swing widely around that line.

Zendesk's own guidance stresses that the trend matters more than the absolute number: a score climbing quarter over quarter tells you more than a single reading. A 78% that has risen from 72% over three quarters is a healthier signal than a static 85%.

CSAT measures satisfaction with a specific interaction on a percentage scale, NPS measures long-term loyalty on a −100 to +100 scale, and CES measures how much effort an interaction required. The three use different formulas and cannot be converted into one another.

They answer different questions, which is why support teams often track more than one:

CSAT is transactional and immediate — you send it right after a resolved ticket. NPS is relational and periodic. If you are choosing between them, our breakdown of CSAT vs NPS vs CES and which one predicts churn covers the research on where each metric earns its place. A quick heads-up for anyone about to average CSAT and NPS into one "satisfaction" figure: the two scales share no math, so the combined number means nothing.

Send the survey immediately after the interaction you want to measure, while it is fresh in the customer's mind. The most common trigger is the moment a support ticket is marked resolved; other useful triggers are post-purchase, after onboarding, and after a customer uses a help-center article.

Timing changes the score more than most teams expect. A survey sent a week after a resolved ticket measures a faded memory, not the interaction. Zendesk recommends surveying while the experience is recent and automating the trigger so it fires the instant an interaction closes.

Practical rules for accurate collection:

• Attach the survey to a specific event, not a calendar date. "After this ticket closed" beats "end of month."
• Keep it to one question plus an optional comment box. Brevity is why CSAT gets higher response rates than longer surveys.
• Cap frequency so a single customer is not surveyed after every one of five tickets in a week — over-surveying collapses response rates and skews the sample toward the annoyed.
• Read the comments, not just the numbers. The rating tells you the score moved; the comment tells you why.

A platform that fires the survey automatically on resolution removes the biggest source of error — the human who forgets to send it. Converge, for example, sends a five-star CSAT prompt as a platform message when an agent marks a conversation resolved, so the survey lands in the same chat thread the customer was already using, at $49/month flat rate for up to 15 agents.

A low response rate makes your CSAT score unreliable because the customers who bother to respond are rarely a representative sample. People with a strongly positive or strongly negative experience are far more likely to answer than the indifferent majority, which pushes scores toward the extremes.

Zendesk's CSAT guidance flags non-response as one of the metric's real weaknesses: if only your most loyal repeat customers fill out the survey, the score reflects them, not your whole base. Two teams with an identical true satisfaction level can report very different CSAT numbers purely because one collected 8% response and the other 30%.

How to keep the sample honest:

1. Aim for a response rate you can defend. A one-question, in-thread survey typically clears the bar that long email surveys cannot.
2. Report the response count alongside the score. "82% CSAT (n=340)" is a claim; "82% CSAT (n=11)" is noise.
3. Watch for sudden rate changes. A jump or drop in response rate can move the score without any real change in satisfaction.
4. Segment before you conclude. A blended score can hide a channel or cohort that is quietly cratering.

The discipline is simple: never act on a CSAT number without knowing how many responses produced it.

The four mistakes that corrupt CSAT numbers are counting neutral responses as satisfied, silently changing which ratings count, comparing your score to a different scale, and reporting a percentage with no response count attached.

Each one produces a number that looks fine and misleads anyway:

• Counting 3s as satisfied. A neutral rating is not a happy customer. Including it inflates the score and hides dissatisfaction. Keep the top-two-box convention (4 and 5 only).
• Switching the "satisfied" definition mid-stream. Moving from top-two-box (4–5) to top-box (5 only) without flagging it makes a stable program look like it collapsed. Document your definition and hold it constant.
• Comparing across scales. An 8/10 average is not the same as an 80% top-box CSAT. Convert everything to the same method before benchmarking, or the comparison is decorative.
• Publishing a percentage with no denominator. A 95% from 20 responses is not better than an 82% from 400. Always ship the response count with the score.

None of these require statistical sophistication to avoid — they require writing down your method once and refusing to change it quietly. Do this, not that: report "80% CSAT (top-two-box, n=200)"; never report a bare "80%."