Survey Sample Size
Survey sample size is the number of responses you need to collect before a survey result — your CSAT score, your NPS, a customer-feedback finding — reliably reflects what your whole customer base thinks, rather than the luck of who happened to reply.
Get it right and a few hundred responses can speak confidently for tens of thousands of customers. Get it wrong and you make decisions on noise — reading a random wobble as a real shift in how customers feel.
This guide covers sampling for surveys and customer feedback. For the related question of how many calls to score in a quality assurance program, see our companion term, QA Sample Size.
What it is
The number of survey responses you need so the result represents your whole customer base, within a known margin of error and confidence level.
Why it's tricky
People dramatically underestimate how many responses they need — and a survey that feels "big enough" is often nowhere near it.
What is Survey Sample Size?
In plain English
You almost never hear from every customer.
Instead you survey a sample — a subset who respond — and use their answers to estimate what's true of your whole customer base. Survey sample size is simply how many responses you collect.
The bigger and more representative your sample, the more confident you can be that the score reflects how customers really feel. The smaller it is, the more your result is just luck of the draw.
The reason sample size has a whole field of statistics behind it is that there's a precise, mathematical relationship between how many responses you gather and how much you can trust the answer.
You don't have to guess — you can calculate the number you need before you send a single survey, based on how precise and how confident you want to be.
What it is
A deliberately chosen number of responses, set so the result represents your wider customer base within a known margin of error and confidence level.
What it isn't
It is not "however many people replied". A number you happened to collect, with no margin of error attached, isn't a sample size — it's a guess wearing a number.
Why It Matters
Survey sample size is the difference between a finding you can act on and a number that quietly misleads you. Too few responses, and you'll chase phantom trends; far more than you need, and you've spent goodwill and survey fatigue for precision that didn't change the decision.
🎯 For CX & insights leaders
A CSAT or NPS you can defend in front of the executive team — not a figure someone can wave away with "that's only a handful of responses". Sample size is what makes the number credible.
📊 For research & survey teams
Every survey result carries a margin of error whether you state it or not. Knowing your sample size lets you report "72% ± 4%" honestly, instead of presenting a fragile number as hard fact.
🏛️ For benchmarking
Comparing your CX metrics to a benchmark is only meaningful if both rest on adequate samples. A "gap" between two small samples can vanish entirely once you account for their margins of error.
What Determines Survey Sample Size
Four things set the number of responses you need. Once you understand them, the maths stops feeling like a black box.
Confidence Level
How sure you want to be that the true value falls within your range.
Common choices are 90%, 95% (the usual default) and 99%. Higher confidence needs more responses.
Margin of Error
How much wiggle room you'll accept around the result — the "± X%". A tighter margin (say ±3% instead of ±5%) demands a substantially bigger sample.
Population Size
The size of the customer base you're surveying. It matters most when that base is small; once it's large, the responses you need barely grow — which surprises people.
Expected Variability
How split the answers are. The most cautious assumption is a 50/50 split, which requires the largest sample — so it's the safe default when you don't know.
The counter-intuitive bit
To estimate a result at 95% confidence with a ±5% margin of error, you need roughly 384 responses — and that's true whether your customer base is 20,000 or 20 million.
Above a few thousand, the size of your base barely moves the number. That's why national polls of a couple of thousand people can speak for millions; representativeness matters far more than raw size.
How to Work Out Your Survey Sample Size
You don't need to derive the formula by hand, but it helps to know what's driving the number.
The underlying formula
For estimating a result in a large population, the standard (Cochran) formula is n = Z² × p × (1 − p) ÷ e², where Z is the z-score for your confidence level (1.96 at 95%), p is the expected proportion (use 0.5 when unsure, the most conservative choice), and e is your margin of error as a decimal.
For a smaller customer base, you then apply a "finite population correction" that reduces the number — which is why surveying a small list needs proportionally more of them to respond.
Do the maths in seconds
You don't have to work this out by hand. ACXPA's Survey Sample Size Calculator does it for CSAT, NPS and CES — tell it the accuracy you want and it returns the number of responses you need, or check how far you can trust a survey you've already run.
Sampling calls for quality assurance rather than surveys? Use the QA Sample Size term and the QA Sample Size Calculator instead.
In practice, the sensible steps are:
Decide precision first
Choose your confidence level and margin of error before you send the survey, based on the decision you're making — not afterwards, to justify the response count you happened to get.
Protect representativeness
A big sample skewed toward your happiest or angriest customers is worse than a smaller random one. Who responds matters as much as how many — so guard against self-selection.
💡 Build the capability in your team
Reading survey results honestly — margins of error, representativeness, what a score can and can't tell you — is a core CX skill. CX Skills runs Customer Experience training courses covering how to measure and act on customer feedback.
Common Pitfalls
Most survey sample-size mistakes come from collecting too few responses — or a skewed set — and then trusting the result far more than it deserves.
Reporting a score off too few responses
Announcing "CSAT is 82%" off 30 responses feels solid, but the margin of error on that figure can be enormous — the true number could sit anywhere from the low 70s to the low 90s.
The month-to-month swings you then see are mostly noise.
Non-response & self-selection bias
If only your delighted or your furious customers bother to reply, even a large sample misrepresents everyone in the middle.
A big but biased sample doesn't average out — it just makes a skewed read look convincing.
⚠️ Most CX survey results are reported without a margin of error
This is ACXPA's position: too many CSAT and NPS numbers are presented as precise facts when the sample behind them can't support it.
A score built on a few dozen responses is dominated by random variation, yet it gets used to declare trends, set targets and judge teams.
Before you act on a survey number — especially a quarter-on-quarter NPS movement — ask whether its sample could survive a margin-of-error test. Many can't.
💡 Right-size for the decision, not the calendar
The question isn't "how many replies did we get?" — it's "how confident do we need to be, and what's the cost of being wrong?".
A directional read for a team conversation can run on a smaller sample than a board-level investment decision. Match the rigour to the stakes.
If you need help designing surveys or sizing and analysing samples properly, the ACXPA Supplier Directory lists CX Strategy consultants who do exactly this.
Frequently Asked Questions About Survey Sample Size
What is survey sample size?
Survey sample size is the number of responses you collect so that a survey result — such as a CSAT score or NPS — reliably represents your whole customer base, rather than reflecting the random chance of who happened to reply. It's set by the confidence level and margin of error you want, not by how many people you sent the survey to.
How many survey responses do I need?
As a rule of thumb, around 384 responses gives you 95% confidence with a ±5% margin of error — and, surprisingly, that holds whether your customer base is 20,000 or 20 million. If you want tighter precision or higher confidence you need more; for a quick directional read you can accept fewer, as long as you're honest about the wider margin of error that comes with it.
How do I calculate survey sample size?
Choose a confidence level (commonly 95%) and a margin of error (commonly ±5%), then apply the standard formula n = Z² × p × (1 − p) ÷ e², using p = 0.5 when you're unsure of the split. For a smaller customer base you apply a finite population correction that reduces the number. Most people use a sample size calculator rather than doing this by hand.
Does a bigger customer base need more responses?
No, and this surprises people. The size of your customer base matters most when it's small. Once it's into the thousands, the responses you need barely grow — roughly 384 gives you 95% confidence and a ±5% margin whether the base is 20,000 or 20 million. Above a certain point, representativeness matters far more than the raw size of your list.
Does my response rate matter, or just the number of responses?
Both, because they affect different things. The number of responses sets your margin of error. The response rate matters for bias: if only a small, unrepresentative slice of customers reply — your happiest or your angriest — even a large sample can misrepresent everyone else. A good survey sample is large enough for precision and representative enough to avoid non-response bias.
Can I trust an NPS or CSAT from a small sample?
Treat it as a rough signal, not a precise fact. A score built on a few dozen responses carries a wide margin of error, so quarter-on-quarter movements are often just random variation rather than a real change in sentiment. Before acting on a small-sample score — especially for targets or comparisons — check whether the sample is large and representative enough to support the claim.
How is this different from QA sample size?
This page covers sampling for surveys and customer feedback — how many responses you need for a CSAT, NPS or feedback result to be trustworthy. The related question of exactly how many calls to score in a contact centre quality assurance program, and how to make those scores fair and defensible, is covered in our companion term, QA Sample Size.
Where to Next
Summary: Survey Sample Size
Survey sample size is how many responses you need so that a survey result reliably represents your whole customer base, within a known margin of error and confidence level.
It's driven by four things — confidence level, margin of error, the size of your customer base, and expected variability — and the maths is settled: you can work out the number you need before you send the survey.
The most common and surprising lesson is that representativeness matters more than raw size — a few hundred well-chosen responses can speak for millions — but only if the sample is large and unbiased enough for the claim you want to make.
ACXPA's position is that too many CX survey results are reported without a margin of error.
Before you set a target, declare a trend or compare to a benchmark on a CSAT or NPS number, ask whether its sample could survive a margin-of-error test. For the call-review side of sampling, read our companion term, QA Sample Size.
