The right survey sample size is the smallest number of completed responses that gives you enough confidence to make a decision. It depends on how precise you need to be (margin of error), how confident you want to be (confidence level), and whether your population is finite.
These are the most common choices in business surveys. If you’re unsure, start with the simplest: 95% confidence and ±5% margin of error.
| Confidence level | Margin of error | Typical required responses* | When to use |
|---|---|---|---|
| 90% | ±10% | ~68 | Very directional feedback; early discovery. |
| 95% | ±10% | ~97 | Quick pulse checks; lightweight reporting. |
| 95% | ±5% | ~385 | Standard “confidence + precision” baseline. |
| 99% | ±5% | ~664 | High-stakes reporting; larger buffers needed. |
*These are conservative ballpark estimates for large populations using p = 0.5 (the most conservative case). If you know the expected proportion, required sample can be smaller.
Precision costs responses. Cutting margin of error from ±10% to ±5% generally increases required responses by about 4×.
Most business surveys estimate a proportion (e.g., % satisfied, % who recommend, % who experienced an issue), so the “proportion” sample size formula is the most common.
The margin of error gets smaller when you increase sample size, but it shrinks at a diminishing rate (square root relationship).
E = z × √( p × (1 − p) ÷ n )
Rearranging the formula gives the common sample size equation for proportions:
n = (z² × p × (1 − p)) ÷ E²
Conservative: p = 0.5 → n = (z² × 0.25) ÷ E²
Example: 95% confidence (z = 1.96), ±5% margin (E = 0.05) → n ≈ 384.16, so round up to 385.
If your population is small and you are sampling a meaningful fraction of it, you can reduce required sample size with a finite population adjustment.
n_adj = n ÷ ( 1 + (n − 1) ÷ N )
Where N = population size
Practical tip: FPC tends to matter when your sample would be around 10% or more of the population. If you’re sampling 1–2% of a big customer base, it usually changes very little.
The most common planning mistake is stopping at “we need 385 responses” and forgetting that response rate might be 10–20%.
Invites needed = Target responses ÷ Expected response rate
Example: If you need 385 responses and expect 15% response rate, invite about 2,567 people (385 ÷ 0.15).
Statistics helps with precision, but real surveys fail in predictable ways: low response rate, biased samples, and uneven segment coverage.
If you want to compare regions, product tiers, or customer cohorts, you need enough completed responses in each segment to make a decision.
A survey with 15% response rate can still be useful — but only if respondents are reasonably representative. If only your happiest (or angriest) customers respond, results skew.
Fix: sample across the customer base, avoid over-surveying power users, and include multiple channels if needed.
If you’re just trying to find the top 3 pain points, you often don’t need 385 responses — you need clear patterns.
Fix: pair a smaller survey with qualitative follow-up (interviews) or open-text analysis.