How to Calculate CSAT (and What to Do When It Drops)

Here is how I calculate CSAT. Customer satisfaction score (CSAT) is the share of survey answers that are satisfied: on a 1 to 5 scale, the 4s and 5s divided by all answers, times 100. If 1,920 of 2,400 customers answer 4 or 5, CSAT is 80.0%.

The free CSAT calculator does this from your rating counts. When I was head of customer support, CSAT was the first number my manager looked at every month. It took me one bad month to learn that the headline score says very little on its own. This is how I calculate CSAT, how I read it, and what I did the month it dropped six points.

How do you calculate CSAT?

CSAT (%) = Satisfied responses ÷ Total responses × 100

Satisfied responses = answers in the top two boxes of your scale
                      (4 and 5 on a 1–5 scale, 6 and 7 on a 1–7 scale,
                      thumbs up on a thumbs scale)
Total responses     = every answer received, including neutral and unhappy ones
                      (surveys that were sent but not answered stay out)

Neutral and unhappy answers stay in the total, so they pull the score down.

What does a CSAT calculation look like with real numbers?

Say 2,400 customers answered the survey this month:

Rating	Answers
5	1,310
4	610
3	230
2	110
1	140
Total	2,400

Satisfied = 1,310 + 610 = 1,920. CSAT = 1,920 ÷ 2,400 × 100 = 80.0%.

Some help desks average the raw ratings instead. The same month gives an average of 4.18 out of 5. I never mix the two methods. A 4.18 average and an 80% top-box score describe the same answers but are not comparable, so I pick one and keep it. To check your own month, enter the five counts in the CSAT calculator.

What rules should I agree before I report CSAT?

1. One survey per solved ticket, sent right after it is solved, never after every reply.
2. The same exclusions every month: no surveys on spam, duplicates or internal tickets.
3. Always show the number of answers next to the score. 80% from 2,400 answers and 80% from 40 answers are not the same result.

If you also track NPS or CES, CSAT, NPS or CES? How I choose the right metric explains what each one is for.

What is a good CSAT score?

There is no single answer, because channel, industry and survey timing all shift the score. Chat and phone usually score higher than email in my experience, and refunds always score lower than "how do I" questions. My working thresholds were 85% for all contacts and 80% for refunds. I treat outside benchmarks as a rough guide at most. Your own trend, split by contact reason, matters far more than someone else's average.

Why did our CSAT drop from 86% to 80%?

One month the score fell from 86.0% to 80.0%. The first reaction in the room was "agents are getting worse". The split by contact reason told a different story.

Refund contacts had dropped from 79.0% satisfied to 61.0%, and booking changes from 86.0% to 79.1%. Every other reason was flat or slightly up. A change in the refund timeline had gone live without any message to customers, so people asked "where is my money?" and gave low scores even when the agent did everything right.

Two numbers settled it:

• Mix: refunds went from 17.4% of all answers (410 of 2,350) to 21.7% (520 of 2,400). More of the month's answers came from the unhappiest reason.
• What if: if refunds and booking changes had kept last month's rates, the month would have scored about 85.8% (2,058 of 2,400). Those two reasons explain almost the whole drop.

No amount of agent coaching would have fixed that. It was a policy and communication problem.

How many CSAT responses do I need?

A score is only as solid as the number of answers behind it. With 2,400 answers, 80% is accurate to about ±1.6 points at 95% confidence. For refunds alone (520 answers, 61%), it is about ±4.2 points.

At agent level it gets worse. An agent with 12 survey answers moves 8.3 points with a single rating. I never rank or coach agents on CSAT with fewer than about 30 answers in the period. I look at their comments and QA reviews instead. The survey sample size calculator shows how many answers you need for the precision you want.

Response rate matters too. We sent about 16,000 surveys and got 2,400 answers, a 15% response rate. If that rate suddenly halves, the customers still answering may not represent everyone.

What I've seen in practice

The fix for the refund drop took six weeks:

1. Told customers the timeline upfront. The refund reply now states the processing time and the date to expect the money.
2. Sent a proactive update when the refund was approved, so customers did not have to ask.
3. Gave agents one clear answer with a template for the new timeline, instead of each agent explaining it differently.
4. Shared the low-score comments with the finance team every week, so the people who owned the policy saw its effect.

Two other mistakes I have made and stopped making: comparing CSAT across teams that survey at different moments, and celebrating a rise that came only from fewer refund tickets that month. Mix changes move the headline score as much as service quality does.

My CSAT dashboard

I update this table on the first working day of each month. "This month" is six weeks after the fixes; "last month" is the month of the drop.

Metric	Target	This month	Last month	Change
CSAT, all contacts	85%	84.5%	80.0%	+4.5 pts
CSAT, refunds	80%	74.0%	61.0%	+13.0 pts
CSAT, booking changes	85%	84.0%	79.1%	+4.9 pts
Survey answers	2,000+	2,300	2,400	−100
Response rate	15%	14.8%	15.0%	−0.2 pts

Refunds are still below target, so they stay first on the list for next month. For the relationship side, see how I calculate NPS, and for effort, the CES calculator.

Frequently asked questions

Is CSAT a percentage or an average?

Usually a percentage: satisfied answers divided by all answers. Some tools report an average rating instead. Both are valid, but never compare one with the other.

Do neutral answers count in CSAT?

Yes. A 3 on a 1 to 5 scale stays in the total but not in the satisfied count, so neutral answers lower the score.

How often should I calculate CSAT?

Monthly for the team, split by contact reason. Weekly only if each reason gets enough answers to be stable.

Can I judge an agent on CSAT?

Only with enough answers. Below about 30 in the period, one rating moves the score by several points, so I use comments and QA reviews instead.

Sources

• What is CSAT and how do you measure it?, Qualtrics
• Margin of error calculator, formula and examples, SurveyMonkey