The honest answer: it depends on the question you're asking. For directional product feedback, 30 to 50 responses can be enough. For a statistically significant market research study with a 95% confidence level and 5% margin of error, you need around 384 responses for any population over 100,000. That is the number most people quote, but it only makes sense in the right context. This guide shows you how to decide what is right for your survey, without turning it into a math exercise nobody wants to finish.
Key takeaways
- Survey sample size depends on the decision you need to make, not just the size of your audience.
- A 95% confidence level and 5% margin of error usually means about 384 responses for large populations.
- 30 to 50 responses can be enough for early product discovery, internal pulse checks, and qualitative theme spotting.
- If you want to compare segments, you need enough responses in each segment, not just enough responses overall.
- Response rate matters as much as sample size. If you need 400 completes and expect a 20% response rate, you need to invite about 2,000 people.
If you are collecting responses now, sign up for PollPe Survey Builder. PollPe's free tier includes unlimited responses, so the survey tool does not become the bottleneck after you have done the work of finding the right audience.
The two questions you have to answer before calculating anything
Before you open a sample size calculator, answer two questions.
1) What decision are you making?
A survey sample size only matters if it supports a decision.
Are you trying to:
- decide which onboarding problem to fix first,
- estimate market demand for a new feature,
- measure customer satisfaction across regions,
- compare pricing sensitivity by company size,
- or publish a statistically significant survey result to stakeholders?
Those are not the same task.
If you are doing early product discovery, you usually want patterns, not precision. If you are planning a launch and need to know whether 62% of your target buyers prefer option A or option B, you need a much tighter sample.
2) What is your tolerance for being wrong?
This is where confidence level survey math comes in.
Ask yourself:
- How much uncertainty can you live with?
- Do you need a broad directional read, or a number you can defend in a board deck?
- Do you need a national estimate, or just a signal from a defined customer list?
A startup founder, a product manager, and a UX researcher can all ask the same survey question and need different sample sizes.
Example:
- A founder testing an idea with 40 target users may only need enough signal to decide whether to proceed.
- A product team measuring feature preference across customers may need several hundred responses.
- A market researcher comparing age, role, and region may need thousands.
If you get these two questions wrong, the math is irrelevant. You can calculate the perfect sample for the wrong goal and still make a bad decision.
If your team is still deciding what to ask, our survey question types guide can help you choose between single select, multi select, matrix, rating, and open text questions.
The math, plain English
The standard formula most sample size calculators use is Cochran's formula:
n = z² × p × (1 - p) / e²
Here is what that means in plain English:
- n is the number of responses you need.
- z is the z-score for your confidence level.
- p is the share you expect to see for the thing you are measuring.
- e is your margin of error.
Most people do not know the true value of p, so they use 0.5. That is not random. It gives the largest sample size, which makes the estimate conservative.
What confidence level means
Confidence level tells you how often the result would fall within the margin of error if you repeated the same survey many times with random samples.
Common choices:
- 90% confidence means you are comfortable being wrong a little more often.
- 95% confidence is the standard for most business research.
- 99% confidence is stricter, but it costs more because you need a larger sample.
If you raise the confidence level, the required sample size goes up.
What margin of error means
Margin of error is the range around the result.
If a survey says 52% of respondents prefer feature A with a 5% margin of error, the true number is likely somewhere between 47% and 57%.
Smaller margin of error means more precision, but also a larger sample. A 3% margin of error needs far more responses than a 10% margin of error.
Why 384 is the default answer
The 384 number is not magic. It comes from the standard formula with these assumptions:
- 95% confidence level
- 5% margin of error
- p = 0.5, which is the most conservative estimate
- random sampling
For a very large population, the math looks like this:
- z = 1.96 for 95% confidence
- 1.96 × 1.96 = 3.84
- p × (1 - p) = 0.5 × 0.5 = 0.25
- 3.84 × 0.25 = 0.96
- margin of error 5% means 0.05 × 0.05 = 0.0025
- 0.96 ÷ 0.0025 = 384
So you round up to 385.
That is why people say "you need 384 responses." What they usually mean is, "for a large population, this is the sample size needed to estimate a proportion with about 5% error at 95% confidence."
A short worked example
Say you are launching a pricing survey for SMB buyers.
You want:
- 95% confidence
- 5% margin of error
- no prior guess for the answer, so p = 0.5
Step by step:
1. Take the z-score for 95% confidence, which is 1.96.
2. Square it. That gives 3.84.
3. Multiply by 0.5 × 0.5, which is 0.25.
4. That gives 0.96.
5. Divide by 0.05 squared, which is 0.0025.
6. You get 384.
That means you need about 384 completed responses if you want a clean estimate for a large population.
For a smaller list, the needed sample drops a bit because you are sampling from a known finite group.
Sample size cheat sheet
Below is a practical cheat sheet for common population sizes. The numbers are approximate and assume a random sample and p = 0.5.
| Population size | Needed for 95% confidence, 5% margin of error | Needed for 90% confidence, 10% margin of error |
|---|---|---|
| 50 | 45 | 29 |
| 100 | 80 | 41 |
| 500 | 218 | 59 |
| 1,000 | 278 | 63 |
| 5,000 | 357 | 67 |
| 100,000+ | 384 | 68 |
A few things to remember:
- Once your population gets large, the sample size levels off.
- You do not need 10,000 responses just because you have 10,000 customers.
- If your goal is a directional read, a smaller margin of error may be fine.
- If your population is tiny, use the finite population version of the calculation, not the large-population shortcut.
If you need a quick check while building a study, a sample size calculator will give you the same result under the hood. The calculator is useful, but only after you know what kind of answer you need.
When 30 responses is enough
There are plenty of situations where 30 responses is enough, and forcing a bigger sample only slows you down.
Early product discovery
If you are trying to understand:
- why trial users drop off,
- what problem they are hiring your product to solve,
- which features confuse new users,
- or what language customers use to describe pain points,
then 30 to 50 responses can be enough to spot patterns.
You are not trying to estimate the exact share of the market. You are trying to learn what matters.
This is where PollPe's AI survey creation with Aria helps. You can draft a fast survey, test a few question variations, and get to the next round of learning without spending a day setting up forms. That matters most when you are still shaping the problem, not proving a thesis.
Qualitative coding saturation
If your open text answers start repeating the same themes, you may have enough responses for the moment.
That does not mean the survey is finished forever. It means you have enough to see recurring themes such as:
- price sensitivity,
- missing trust signals,
- unclear onboarding,
- or feature confusion.
If you are analyzing themes from sentiment or open text, our customer sentiment analysis guide is a good companion piece.
Internal employee pulse checks
For internal feedback, a 30 to 50 response sample can work when:
- the team is small,
- the goal is to understand a pain point quickly,
- and you plan to follow up with interviews or a workshop.
A weekly team pulse survey does not need the same rigor as a market sizing study.
When speed matters more than precision
If you are deciding whether to move forward, delay, or test again, the value is in speed.
A rough signal now is better than a perfect number next month.
That is also why PollPe's free tier matters. It includes unlimited responses, so you can keep collecting feedback while you refine the question set. You do not need to stop because the plan says you have used up your monthly quota. That is a real issue with some tools that cap free plans at 10 responses a month.
If you want to start testing fast, create your survey in PollPe Survey Builder and keep iterating until the pattern is clear.
When 384 isn't enough
A sample of 384 can be fine for a top line answer, but it can fall short in several common B2B cases.
Segment-level analysis
If you want to compare:
- founders vs product managers,
- enterprise buyers vs SMB buyers,
- India vs global,
- or new users vs power users,
then the total sample matters less than the sample inside each group.
A total sample of 384 might sound solid until you realize only 41 of those responses came from a segment you care about.
That is too small to compare with confidence.
Rare events
If only a small share of users has the experience you are measuring, you need a larger sample.
Examples:
- churned customers,
- users who saw a specific feature,
- buyers who rejected a pricing tier,
- or people from a niche role like procurement or IT security.
If the thing you want to measure is rare, the base size shrinks fast.
Pricing and conversion tests
Pricing surveys often need more than a simple 384 because the analysis can get split by:
- company size,
- buying role,
- geography,
- and willingness to pay bands.
The same is true for conversion questions. If you are testing a landing page, one feature variant, or a message angle, your low-base cells can become too small to trust.
Low-base cells
A low-base cell is a subgroup with too few responses to analyze.
Example:
- 300 responses total
- 8% are from finance leaders
- that gives you 24 finance responses
You can read those answers, but you should not overclaim from them.
If your analysis depends on comparing low-base groups, you need more total responses, or you need to oversample the important group.
Calculating sample size for subgroups
If you want to analyze five segments, do not assume one sample covers them all.
A common mistake is thinking, "We need 400 responses, and we have five segments, so 400 is enough."
It is not.
If each segment needs a usable base, then each segment needs its own minimum.
For example, if you need around 384 responses in each of five segments, you need about:
384 × 5 = 1,920 responses
That is the clean version.
In practice, segment sizes are uneven, so you may need even more total responses if one group is hard to reach. A segment that makes up only 10% of your audience is going to need oversampling if you want to compare it fairly.
This matters a lot in B2B surveys because audience splits are rarely balanced. You may have:
- 60% end users,
- 25% managers,
- 10% decision makers,
- 5% admins or operators.
If the smallest group is the one you care about most, plan for that group first.
If you are running a pan India study, language also affects subgroup quality. A single English survey can suppress completion rates in some audiences. Running the same survey in Hindi, Telugu, and Tamil can raise response quality for region specific studies and reduce the chance that your sample is skewed toward English first users.
For larger studies, PollPe's Enterprise plan includes a built in respondent panel. That is useful when you need specific demographics or job roles and cannot wait for your own list to fill the quota.
Response rate vs sample size: how many invites do you actually need to send
Sample size is the number of completed responses you want.
Response rate is the share of invited people who actually finish the survey.
Those are not the same.
Use this simple formula:
Invites needed = target completes ÷ expected response rate
Quick example
Say you need 384 completed responses.
If your expected response rate is:
- 20%, you need to invite about 1,920 people
- 15%, you need to invite about 2,560 people
- 10%, you need to invite about 3,840 people
- 8%, you need to invite about 4,800 people
This is why improving response rate is one of the cheapest ways to improve survey quality. Better targeting, shorter surveys, and clearer wording can cut the number of invites you need.
If you want practical ways to raise completions, see our guide on how to improve survey response rates.
A few common ways to improve completion:
- keep the survey short,
- lead with one clear purpose,
- remove vague or double barrel questions,
- use simple language,
- and send reminders.
That is also where question design matters. A bad question can suppress completion even if your sample list is strong. Use the right question type for the job, and check the survey question types guide before you send the next draft.
Common mistakes
Confusing sample size with response count
People often say, "We sent 1,000 surveys," as if that is the same as 1,000 responses.
It is not.
The number that matters is completed responses. That is what sample size refers to.
Using the global formula when the population is tiny
The 384 answer works for large populations. If your list has 120 people, you do not need to pretend it is a nation-sized survey.
For small lists, use the finite population version of the calculation.
Ignoring non-response bias
If only one type of person answers your survey, the sample can be biased even if it is large enough on paper.
For example:
- only power users respond,
- only happy customers respond,
- only English speakers respond,
- or only people from one region respond.
A bigger sample does not fix that problem by itself.
Treating one survey as proof of everything
A survey is a tool, not a verdict.
If the result matters a lot, pair the survey with interviews, product data, support tickets, or sales notes. That is especially useful in B2B, where buying committees are complex and one respondent may not represent the whole account.
Using the wrong target audience
You can have a perfect sample size and still get a useless study if the audience is wrong.
If you want to understand enterprise buyers, do not survey a general consumer list and hope the answers line up.
FAQ
Is 100 responses enough?
Sometimes, yes.
If you are doing early discovery, a team pulse, or a rough customer read, 100 can be more than enough. If you need a statistically significant survey for a large audience at 95% confidence and 5% margin of error, 100 is not enough.
The right answer depends on the decision you are making.
What confidence level should I use?
For most B2B survey work, 95% is the standard choice.
Use 90% if you want a lighter, faster read and the stakes are lower. Use 99% if the result is high stakes and you can afford a larger sample.
Does sample size matter for B2B?
Yes, but the context matters more than in broad consumer research.
B2B audiences are smaller, more segmented, and more expensive to reach. That means you often care more about the quality of the list than the abstract population size. A survey of 60 very relevant buyers can be more useful than 600 random respondents.
Can I weight a small sample?
You can weight a sample, but weighting does not create data that was never collected.
If one segment has too few responses, weighting may help the report look more balanced, but it does not fix weak evidence. Use weights carefully and do not treat them as a shortcut around a small base.
How do I increase sample size cheaply?
Start with the obvious moves:
- shorten the survey,
- send to a better matched list,
- time reminders well,
- offer a clear reason to respond,
- and make the survey available in the right languages.
For pan India audiences, Hindi, Telugu, and Tamil versions can improve completion. For niche audiences, a respondent panel can be faster than trying to grow your own list from scratch.
Also, do not let your tool cap you early. PollPe's free tier includes unlimited responses, so you can keep gathering data until you reach the sample you actually need.
Bottom line
The right survey sample size depends on what you need to decide.
If you are still exploring a problem, 30 to 50 responses may be enough to show a pattern. If you need a statistically significant survey with a 95% confidence level and 5% margin of error, plan for about 384 completed responses for a large population. If you need segment-level insight, rare event data, or account-by-account comparisons, the total sample will need to be much larger.
The math is simple. The hard part is picking the right question, the right audience, and the right level of certainty.
If you want to run the next survey without fighting tool limits, start free in PollPe Survey Builder. If you already know you need a bigger study or a respondent panel, check PollPe pricing and choose the plan that fits the sample you need.
