CSAT, short for customer satisfaction score, is the percentage of survey respondents who rate one specific experience, such as a delivery, a support call or a checkout, as 4 or 5 out of 5. It belongs on the scorecard of any digital strategy because it gives a direct reading of whether the customers your advertising brings in are happy once they have bought.
The formula takes one line. What makes a score useful is three decisions most guides skip: what counts as "satisfied", how many responses you need before a movement is real, and why your percentage cannot be compared with the satisfaction index scores South African brands publish. This post works through each with the arithmetic shown.
Quick Answer
CSAT is calculated as the number of 4 and 5 ratings divided by total responses, multiplied by 100. A vendor benchmark puts a good score at 75% to 85% for most sectors, but a score built on a small number of responses can swing by several points on chance alone. Survey straight after the interaction, ask one rating question and one open question, and track the score by channel, judging it over pooled periods that give each segment a few hundred responses rather than reacting to single months.
In This Post
Collecting satisfaction scores but not sure what to do with them?
Contact us with your current survey question and where it is sent, and we will review it and suggest what to track beside the score.
Get a Survey ReviewCSAT Formula: How Do You Calculate the Score?
The csat formula divides the number of satisfied responses by the total number of responses and multiplies by 100, where "satisfied" means a rating of 4 or 5 on a five-point scale. IBM's explainer gives it as: add the positive responses (ratings of 4 or 5 out of 5), divide by the total number of responses collected, then multiply by 100.
The inputs below are invented purely to show the arithmetic. They are not data from any business.
| Rating | Responses | Counts as |
|---|---|---|
| 5 (very satisfied) | 90 | Satisfied |
| 4 (satisfied) | 60 | Satisfied |
| 3 (neutral) | 25 | Not counted |
| 2 (dissatisfied) | 15 | Dissatisfied |
| 1 (very dissatisfied) | 10 | Dissatisfied |
| Total | 200 | Score = (90 + 60) ÷ 200 × 100 = 75% |
That 75% score hides a shape worth reading: 12.5% of respondents were neutral and another 12.5% were dissatisfied. The percentage is a headline, and the distribution underneath it tells you where to look.
Takeaway
The score is the share of respondents who chose the top two ratings. In the worked example, 150 satisfied responses out of 200 gives 75%, with the other 25% split between neutral and dissatisfied customers.
Which Scale and Which Ratings Count as Satisfied?
The scale and the definition of "satisfied" are the two choices that decide whether a csat survey can be compared month to month, because changing either one changes the score without changing a single customer's opinion. IBM notes that organisations use scales of 1 to 3, 1 to 5 or 1 to 10, or a percentage scale. Hello Customer adds that a 7-point or 10-point scale works as long as you define consistently what counts as positive.
The 1 to 5 scale with 4 and 5 counted as satisfied is the common convention, so it is the safest choice if you want others to read your number the same way. Pick it once and write it down.
Side calculation: what happens if you count the 3s
Take the same 200 responses. If neutral ratings are counted as satisfied, the score becomes (150 + 25) ÷ 200 × 100 = 87.5%, against 75% under the standard definition. Nothing about the customers changed; only the rule did. If a competitor, agency or software dashboard quotes a higher score than yours, check which ratings it counts before concluding anything.
Dashboards add a second trap. A dashboard may also show an average rating out of 5 beside the percentage, and the two are different measures. Report one of them consistently, and label it.
How Many Responses Do You Need Before a Score Means Anything?
As a working rule of thumb, you need about 200 responses per channel for a usable baseline, and well over 400 per period before a few-point change between two periods is readable. A satisfaction score is an estimate from the customers who answered, so the fewer responses behind it, the wider the range it could plausibly sit in. You can approximate the 95% margin of error for a percentage as 1.96 × the square root of (p × (1 − p) ÷ n), where p is the score as a proportion and n is the number of responses.
Using the 75% score from the example, the margin shrinks slowly as responses grow:
| Responses (n) | Approximate margin of error | Read it as |
|---|---|---|
| 50 | about ±12 percentage points | Too noisy to track movement; read the comments instead |
| 100 | about ±8 percentage points | Only large swings mean something |
| 200 | about ±6 percentage points | A usable baseline for one channel or team |
| 400 | about ±4 percentage points | A single score is fairly tight; a change between two periods still needs about 6 points |
| 1,000 | about ±3 percentage points | Tight for the whole base, but splitting it into segments divides n again |
At 100 responses, a reported 75% could sit anywhere from roughly 67% to 83%. So a fall from 75% to 72% in a small month is not evidence that anything changed.
The table gives the margin on one score. A change between two periods carries more uncertainty, because both scores are estimates: with equal responses in each period, the margin on the difference is about 1.4 times larger. At 200 responses per period a movement needs to be roughly 8 points, at 400 roughly 6 points, and at 1,000 roughly 4 points before it clearly exceeds chance. Splitting by channel, product or agent divides your responses among the segments, so check each segment against the table, not the total.
Treat these margins as a best case. They assume the people who answered are a random sample of your customers, and they are not. IBM warns that customers are more likely to complete a survey after an exceptionally good or exceptionally bad interaction, which slants results towards the extremes. The real uncertainty is wider than the table.
Takeaway
With 100 responses a 75% score has a margin of roughly 8 percentage points either side, and about 3 points at 1,000 responses. A change between two periods needs roughly 1.4 times that margin to stand out, so pool several weeks before judging it and treat a small monthly dip as noise.
Not enough responses to trust your score?
Tell us how many orders or enquiries you handle a month and how the survey is sent, and we will talk you through what a reporting cycle can support.
Check My Response VolumeWhen and How Should You Send the Survey?
The survey should reach the customer straight after the interaction it measures, with one rating question and one open question, because the experience is freshest then and a short form is easier to finish. Hello Customer's guidance is to ask "within hours of a support contact, within a day of a delivery, straight after checkout."
A standard wording is "How satisfied were you with this experience?", which Thematic's comparison lists as the typical CSAT question on a 1 to 5 scale. Name the specific moment instead: "How satisfied were you with your delivery?" produces a score you can act on, while a vague question produces a mood.
- Pick the moments that matter. For a South African online store that is usually checkout, delivery and a returns or support contact. For a service business, it is after the first call, after the quote and after the job. If calls are a major channel, call tracking shows which enquiries came from which campaign, so you can match the score to the source.
- Ask one rating question. Keep the same scale across all your surveys.
- Add one open question. "What is the main reason for your score?" IBM points out that the rating alone is not in-depth enough to explain why customers are satisfied or not. The comment does that job.
- Send it where the customer already is. Use the channel the customer already used to buy or contact you, so the invitation arrives where they are paying attention.
- Record the score against the order, the channel and the date. A score you cannot segment is a number you cannot use. Our guide to Google Analytics 4 covers tying behaviour to acquisition source.
Mind the consent rules. Section 69 of POPIA prohibits electronic direct marketing unless the person has consented or is a customer. The customer route applies only where you obtained their details in the context of a sale, you market your own similar products or services, and they had a chance to object when the details were collected and on each message. Every message must also name the sender and give an address for stopping further messages.
Whether a pure satisfaction survey counts as direct marketing is a legal question we found no source to settle, so put it to a POPIA adviser. Keep the invitation free of promotion either way, include a clear opt-out, and see our POPIA first-party data guide for how survey data fits into your wider data practice.
NPS vs CES vs Satisfaction Scores: Which Measure Answers Which Question?
Each of the three common customer experience measures answers a different question, so the choice depends on whether you care about one interaction, the whole relationship or the effort involved. Hello Customer sums up the first split: CSAT measures satisfaction with one interaction, while NPS measures loyalty to the company as a whole.
| Measure | Typical question and scale | How it is scored | Best for |
|---|---|---|---|
| CSAT | "How satisfied were you with this experience?" 1 to 5 | Percentage choosing 4 or 5 | Checking one moment: delivery, support, checkout |
| NPS | "How likely are you to recommend us to a friend or colleague?" 0 to 10 | Percent promoters minus percent detractors, from -100 to +100 | Tracking loyalty across the relationship |
| CES | "[Company] made it easy to handle my issue." Usually 1 to 7 | Average response | Finding friction in a process such as returns or onboarding |
For most small and mid-sized South African businesses, the sensible start is a short rating survey at two or three points in the customer journey, and NPS added later once volumes justify it. Our post on marketing goals vs KPIs explains where a satisfaction measure sits against the numbers you report on, and customer experience in South Africa covers the wider discipline.
Why Can't You Compare Your Percentage With the SA-csi Index?
The South African Customer Satisfaction Index (SA-csi) reports a 0 to 100 index built from a causal model, which is a different measure from a csat percentage, so the two numbers cannot be set side by side. According to Moonstone's account of the index, SA-csi uses an ACSI-based causal model that links customer expectations, perceived quality and perceived value to satisfaction, which in turn is linked to complaints and loyalty intentions.
Use it for context, not as a target. In the 2025 short-term insurance results reported by Money Marketing, the industry scored 69.5 across 6,384 consumers, and the leader, Auto & General, scored 73.7. Those are index points, not the share of customers giving 4 or 5.
The same report holds a lesson that does apply to any satisfaction survey: problems and how they end. Only 7% of customers had a problem, but 60% of those were resolved to the customer's expectations, 27% were somewhat resolved and 13% stayed unresolved. The report said unresolved issues should sit below 5%, and that anything above 10% creates a pool of upset customers.
Prof Adré Schreuder of the University of Pretoria, quoted in a Moonstone report on the SA-csi insurance results, said: "This year's index was also telling in the impact that complaints handling and speedy claims resolution has on overall customer satisfaction." For a small business the practical reading is to set a fixed reply window for every 1 or 2 rating and treat the resolved case as part of the score, not a separate department.
Takeaway
An SA-csi score such as 69.5 is an index out of 100, not a percentage of satisfied customers. Compare your own score only with your own earlier scores and your own definition of satisfied.
What Is a Good Score, and What Should You Do With It?
A good score is one that holds steady or rises on your own definition and your own customers, with the public range of 75% to 85% as a loose reference point only. Hello Customer says a good result sits between 75% and 85% for most sectors. That is one vendor's benchmark, not a South African standard, and it does not say which survey design produced it.
Two things move the number besides service quality. IBM notes that outside events or a customer's personal situation can skew a score, and Hello Customer describes the metric as mood-sensitive, with scores dipping after price increases, in peak-season queues and, for delivery businesses, in bad weather. Read a dip against what was happening that month before blaming a team.
For a South African online store, load-shedding days or courier delays are plausible examples of outside events that can drag down delivery ratings without any change in how the store performs. That is a reason to record the delivery method and the date beside each score, so a bad week can be matched to its cause. It is an illustration of the mechanism, not a measured effect.
- Segment first, within the limits of your volume. Split by acquisition channel, product line, delivery method and support agent, but only as far as each segment keeps enough responses. One blended score hides the part that is broken.
- Read the neutral and dissatisfied comments. In the worked example, 25 of 200 customers rated 1 or 2. Reading those 25 comments is a short job and says more than the percentage.
- Close the loop on low scores. Reply to the customer who rated 1 or 2, fix the problem and record the outcome.
- Review quarterly against the sample table above. Act on shifts that exceed the margin on a change (about 1.4 times the single-score margin), not on every wobble.
Want satisfaction tracked beside your acquisition numbers?
Request a free audit and we will show you how to line up scores by channel with your existing reporting.
Request a Free AuditWhy South African Businesses Choose Growth Pulse Media
Growth Pulse Media is a Johannesburg agency led by Dirk van Greuning, who built and scaled a large South African ecommerce business before founding it. All work is executed in-house and the client load is kept limited, so senior attention goes to every account. Our digital strategy service starts from the numbers a business should track, then sets the plan, and satisfaction measures sit alongside revenue and acquisition cost on that scorecard.
Who This Is NOT For
This guide is not for businesses that want a satisfaction number to put on a slide without acting on it, or that cannot yet collect enough responses to read one.
You want a score to quote in marketing. A self-reported 75% from an unrepresentative sample, with a definition of satisfied you chose, is a weak claim and an easy one to challenge.
You handle only a handful of orders a month. The margin-of-error table shows that a score from 50 responses carries about ±12 points. Read individual comments and phone the customers instead.
You will not follow up on low ratings. A low rating is the one moment a customer has told you exactly where you failed. Collecting it and never replying wastes the chance to fix the problem.
Frequently Asked Questions
What does CSAT stand for?
CSAT stands for customer satisfaction score. It is the percentage of survey respondents who rate a specific interaction as satisfied, usually 4 or 5 on a five-point scale.
How do you calculate a CSAT score?
Divide the number of 4 and 5 ratings by the total number of responses and multiply by 100. For example, 150 satisfied responses out of 200 gives a CSAT score of 75%.
What is a good CSAT score?
Hello Customer puts a good CSAT between 75% and 85% for most sectors. Treat that as a vendor benchmark rather than a South African standard, and judge your own score mainly against your own history and definition.
How many responses do you need for a reliable CSAT score?
As a rough guide, a 75% score from 100 responses carries a margin of about 8 percentage points, and about 3 points at 1,000 responses. A change between two periods needs about 1.4 times that margin to stand out, and all of it assumes a random sample, so real uncertainty is wider.
What is the difference between CSAT and NPS?
CSAT measures satisfaction with one interaction, while NPS measures how likely customers are to recommend the company as a whole, on a 0 to 10 scale. CSAT suits checking a specific moment, and NPS suits tracking loyalty over time.
Is CSAT the same as the SA-csi score?
No. The South African Customer Satisfaction Index (SA-csi) is a 0 to 100 index built from a causal model of expectations, quality, value and loyalty. A CSAT result is the share of respondents who gave a 4 or 5, so the two cannot be compared directly.
Turn satisfaction scores into decisions
Whether you run a Shopify store or a service business, we will review your survey and reporting and show you what to track. No obligation — we'll get back to you within 24 hours.
Talk to Growth Pulse Media

