How to calculate survey sample size
Sample size comes down to four inputs and one formula. Here is how to size a survey correctly, avoid the common mistakes, and actually recruit the number you need.
Sample size comes down to four inputs and one formula. For a general-population survey, 384 completed responses give you a 95 percent confidence level with a 5 percent margin of error. Change the confidence, the margin of error, or the size of a small audience, and that number moves. This guide walks through why sample size matters, the four inputs that drive it, the formula with a worked example, rules of thumb for qualitative work and for segmented studies, and how to actually recruit the number you calculated.
Why sample size matters
You survey a sample because you cannot ask everyone. Sample size is the bridge between the handful of people who answered and the whole audience you want to describe. Get it right and your results are precise enough to act on. Get it wrong in either direction and you waste money or make bad calls.
Too small, and your findings are noisy. A 60 percent preference from 40 people could easily be 50 percent or 70 percent in reality, which is not a basis for a roadmap or a pricing decision. Too large, and you spend budget and fielding time on precision you will never use. The goal is not the biggest possible sample. It is the smallest sample that answers your question with the confidence you need.
Two ideas do most of the work here. Precision is how tightly your estimate pins down the truth, expressed as the margin of error. Confidence is how often that estimate would land inside its stated range if you ran the survey many times. Everything in the calculation is a trade-off between those two things and the cost of collecting more responses. If you are new to the wider discipline, the basics of market research put sampling in context alongside the other decisions a study involves.
The four inputs
Every sample-size calculation needs four things. Decide each one before you touch the formula.
Population size
This is the total number of people you want to draw conclusions about: all your active users, every enterprise buyer in a category, adults in a country. You rarely need the exact figure. Once the population is in the tens of thousands, the required sample barely changes, so an estimate is fine. Population only matters a lot when your audience is small, which we cover below.
Confidence level
Confidence level is the probability that your result falls within the margin of error, if the survey were repeated many times. The convention across research is 95 percent, which corresponds to a z-score of 1.96. If a decision carries real risk, you might move to 99 percent (z of 2.576). Lower-stakes, directional reads sometimes use 90 percent (z of 1.645). The Pew Research Center methods guide is a good reference for how confidence is applied in practice.
Margin of error
Margin of error is how far your sample result may sit from the true value, expressed as plus or minus a percentage. A result of 45 percent with a 5 percent margin means the real figure is likely between 40 and 50 percent. Five percent is the common default. Tighter margins, such as 3 percent, demand substantially more responses. Wider margins, such as 10 percent, need far fewer and suit quick directional checks.
Response distribution
This is the proportion of people you expect to answer a given way. When you do not know it in advance, use 0.5, or 50 percent. That value produces the largest required sample, so it is the safe, conservative choice. If you have strong prior data (say you know roughly 80 percent behave one way), you can use that to reduce the sample, but 0.5 is the standard starting point.
The formula
For a large or unknown population, use the infinite-population formula:
n = (z? ? p ? (1 ? p)) / e?
Where n is the sample size, z is the z-score for your confidence level, p is the response distribution, and e is the margin of error as a decimal.
When your population is small, apply the finite population correction to avoid over-recruiting:
n_adjusted = n / (1 + ((n ? 1) / N))
Where N is the population size. The larger N is, the less this correction changes anything, which is why big audiences all converge on roughly the same number.
A worked example
Suppose you want to survey product managers about a new feature, at 95 percent confidence and a 5 percent margin of error, with an unknown distribution.
- z = 1.96, so z? = 3.8416
- p = 0.5, so p ? (1 ? p) = 0.25
- e = 0.05, so e? = 0.0025
n = (3.8416 ? 0.25) / 0.0025 = 0.9604 / 0.0025 = 384.16, which rounds up to 385.
Now say your total population is only 2,000 product managers. Apply the correction:
n_adjusted = 384 / (1 + ((384 ? 1) / 2000)) = 384 / 1.1915 = 322.
The small audience cut the requirement from 385 to 322. If that population were 200,000 instead, the correction would leave you at 384, essentially unchanged.
Confidence and margin at a glance
The table below shows required completed responses for a large or unknown population, using a distribution of 0.5. Use it as a quick reference before running the formula yourself.
| Confidence level | Margin of error | Required sample size |
|---|---|---|
| 90% (z = 1.645) | 5% | 271 |
| 95% (z = 1.96) | 10% | 96 |
| 95% (z = 1.96) | 5% | 384 |
| 95% (z = 1.96) | 3% | 1,067 |
| 95% (z = 1.96) | 2% | 2,401 |
| 99% (z = 2.576) | 5% | 664 |
| 99% (z = 2.576) | 3% | 1,843 |
Two patterns stand out. Halving the margin of error roughly quadruples the sample. Moving from 95 to 99 percent confidence adds a large chunk of extra responses. Precision is expensive, so buy only as much as your decision needs.
Rules of thumb for qualitative versus quantitative
The formula above is for quantitative surveys where you want to estimate a proportion or a metric. Qualitative research plays by different rules.
Qualitative studies (interviews, usability sessions, open-ended discovery) aim for saturation, the point where new participants stop revealing new themes. There is no margin of error involved. As a practical guide:
- Usability testing: 5 to 8 participants per distinct user type surfaces most major issues.
- In-depth interviews: saturation commonly arrives between 12 and 20 participants for a focused question.
- Concept or discovery work: 15 to 30 sessions across your key segments is typical.
If you are running qualitative work, size by user type and theme, not by statistics. The complete guide to user research covers how to match method to sample, and the market research methodology guide does the same on the quantitative side.
For quantitative surveys, the formula holds, but one rule of thumb overrides all the math: never let any group you plan to report on fall below a usable base. That leads directly to segments.
Sizing for segments
The single most common sample-size mistake is calculating for the total and forgetting the breakdowns. If you compute 384 for your overall audience, then split results by region, company size, and role, some of those cells will have 20 or 30 responses, far too few to trust.
Plan the sample around your smallest reporting segment. A practical target is 100 to 300 completes per subgroup you intend to analyze on its own, with about 200 as a solid middle ground for most B2B work. Do the arithmetic in reverse: decide the segments first, set a per-segment target, then multiply.
| Reporting plan | Per-segment target | Total completes needed |
|---|---|---|
| Overall only | n/a | ~384 |
| 3 segments | 200 | 600 |
| 5 segments | 200 | 1,000 |
| 8 segments | 150 | 1,200 |
A study you thought needed 384 responses can easily need 1,000 or more once you commit to reading it by segment. This is exactly why a customer segmentation study has to be scoped before fieldwork, not after. Deciding your cut points late is the fastest way to end up with segments too thin to report.
How to actually hit the target
A calculated sample size is a recruiting target, and this is where most surveys slip. Two effects work against you.
First, not everyone you invite completes the survey. If you expect a 30 percent completion rate and need 400 completes, you have to reach roughly 1,333 qualified people. Always plan invitations around your expected response rate, not the final number.
Second, and harder in B2B, is finding people who genuinely qualify. Hitting 200 completes among verified enterprise security buyers, or among product managers at Series B companies, is a sourcing problem long before it is a statistics problem. General consumer panels rarely contain enough of these people, and unverified respondents introduce fraud and misrepresentation that quietly corrupt your base. Good participant recruitment is what makes the calculated number achievable rather than aspirational, and it is doubly true when you recruit B2B research participants with narrow professional criteria.
This is the problem CleverX is built to solve. It is a B2B research platform with 8M+ verified professionals plus a B2C reach, spanning 150+ countries, with typical delivery in about 2 to 5 days on a pay-as-you-go basis. Because every participant’s professional identity is verified, you can set precise screening criteria and still fill every segment, including the small, hard-to-reach cells that break most studies. Hitting a sample-size target across every segment is fundamentally a recruiting problem, and a verified panel is what solves it. If you run repeated or higher-stakes B2B studies, the discipline of effective B2B market research starts with knowing you can actually field the sample you designed.
Ready to field your next study? Recruit verified participants on CleverX and hit your sample target across every segment.
A quick checklist before you field
- Define the population you want to describe.
- Set your confidence level, usually 95 percent.
- Choose a margin of error your decision can live with.
- Use 0.5 for distribution unless you have solid prior data.
- Run the formula, then apply the finite population correction if your audience is small.
- List every segment you will report on and set a per-segment base.
- Inflate the invitation count to account for your expected completion rate.
Do these seven things and your survey will be both precise enough to trust and lean enough to afford.
Frequently asked questions
What is a good sample size for a survey?
For most general-population surveys, 384 completed responses gives you a 95 percent confidence level with a 5 percent margin of error. If you need tighter precision or you plan to break results down by segment, you will need more. The right number always depends on your confidence level, margin of error, and how you plan to analyze the data, not on a single magic figure.
How do I calculate sample size for an unknown population?
When the population is very large or unknown, use the infinite-population formula: n equals z squared times p times one minus p, divided by e squared. With a 95 percent confidence level (z of 1.96), a distribution of 0.5, and a 5 percent margin of error, this returns about 384. You do not need the exact population size once it is in the tens of thousands, because the required sample barely changes.
Does a bigger population need a bigger sample?
Not proportionally. A survey of a 10,000-person audience and a 10-million-person audience need almost the same sample size for the same precision. Population size only matters meaningfully when your audience is small, roughly under 20,000, where the finite population correction reduces the number you need.
What sample size do I need to segment my results?
Plan the sample around your smallest reporting segment, not the total. Each subgroup you want to analyze on its own needs its own adequate base, often 100 to 300 completes for reliable reading. If you have five segments and want about 200 responses in each, you are recruiting for 1,000 completes overall, not 384.
How many people do I need for qualitative research?
Qualitative work follows saturation, not statistical formulas. Most usability studies surface the majority of issues with 5 to 8 participants per user type, and interview or discovery studies commonly reach saturation between 12 and 20 participants. These numbers describe depth of insight, not margin of error, so do not apply the survey formula to them.
What is the difference between margin of error and confidence level?
Confidence level is how often the true value would fall inside your range if you repeated the survey many times, usually set at 95 percent. Margin of error is how wide that range is around your result, for example plus or minus 5 percent. Raising confidence or narrowing the margin both increase the sample size you need.