Survey sample size: how much is sufficient?
The question is not how many people are enough, but what conclusions you want to draw and what level of risk you are willing to accept.
There is no single correct number for every case. For quantitative surveys of large customer files, a common convention is around 385 respondents at a 95 percent confidence level with a margin of error of plus or minus 5 percent, but this number only holds if the sample is randomly and correctly selected. If you need to understand reasons rather than measure ratios, conducting in-depth interviews with about 8 to 12 people from each customer group is sufficient for decision-making.
- needs figures to present externally (fundraising, board)
- needs to understand why customers think that way through qualitative interviews
- a small and homogeneous customer base, reduce sample size according to the formula
- gather enough numbers but all from loyal customers, not representative
- combine multiple segments into a single template
The question about sample size is often reversed: people ask "how many people are enough" when the correct question should be "enough to conclude what". Sample size does not exist independently of the objective. A number suitable for one type of conclusion may be completely meaningless for another.
Two types of conclusions, two different standards.
Before thinking about numbers, you need to determine what you want to know.
- Quantitative, measuring ratios: "What percentage of customers remember the brand name?", "What is the average satisfaction score?" Such questions require a large, randomly selected sample, as the results will be scrutinized closely.
- Qualitative, understanding reasons: "Why do customers leave?", "What makes them trust?" Such questions require depth, not breadth. A few dozen interviews with the right people conducted correctly are more valuable than hundreds of surveys asking the wrong questions.
Misunderstanding the goal at this stage is the most costly mistake: needing to understand the reason for gathering a large sample, needing numbers but only asking a few familiar people.
Convention 385 and its accompanying conditions
The number 385 frequently appears in survey design documents. It comes from the Cochran formula, applied to very large customer files with a 95 percent confidence level and a margin of error of plus or minus 5 percent, assuming an estimated proportion of 0.5 (the case of highest dispersion).
However, this number only holds true when two conditions are met simultaneously: the sample is randomly selected and accurately represents the target segment. Lacking either, 385 responses are merely an illusion of accuracy.
The important reality: if your client base is small and known in advance, the necessary number decreases significantly. With a base of 1,000 people, according to finite population adjustments from the same formula, you only need about 278 responses for the same level of confidence. There is no need to strive for 385 at all costs if the total base is not large enough.
385 responses gathered from loyal customers are not as reliable as about 150 responses with a proper template.
Practical experience, Sinh Vũ
For qualitative interviews: continue until saturation.
In qualitative research, the standard is not a fixed number but rather saturation, which is the point at which subsequent interviews no longer reveal new themes or observations.
Many studies note that this point is usually reached with about 8 to 12 in-depth interviews per customer group, provided that the interviewees are correctly selected and the questioning is sufficiently deep. If you have multiple customer segments, each segment should be treated as a separate group, not lumped together.
Factors that really determine the number
- Homogeneity of the customer base: The more similar customers are in behavior and attitude, the smaller the sample size needed. A heterogeneous group requires more or must be segmented.
- Number of segments requiring separate conclusions: Each segment needs its own sufficient template. Consolidating everything into one summary table and then breaking it down later creates numbers that are not credible for any group.
- Risk level of the decision: The larger the decision, the less reversible it is, the higher the safety margin needs to be. A small packaging change requires less data than a complete brand repositioning decision.
- Purpose of using the results: Data for internal decision-making has different standards than data for external publication, fundraising, or presentations to committees.
Common mistake when setting sample sizes
Sufficient quantity but the wrong people: Surveying 400 people, but most are loyal customers, former employees, or acquaintances. The results may seem stable but are not representative of actual customers. The quality of the sample frame is more important than the quantity of feedback.
Sufficient total but lacking in each group: Collecting enough 385 responses but grouping all segments together leads to no group having enough samples to draw separate conclusions. When analyzed by group, the numbers break down and become unusable.
How Sinh Vũ finalizes the sample size in a project
In the brand audit process, the sample size is agreed upon in the first week when defining the scope, not after data collection has begun. The reason: this number is directly linked to the type of conclusions each research layer needs to draw.
For layers that need to measure perception proportionally, use surveys with a sample size appropriate for the customer segment. For layers that need to understand the reasons behind behaviors, lean towards in-depth interviews until saturation. Sinh Vũ's scope map clearly states: how many samples, who to ask, what conclusions to draw, before any questions are sent out. This ensures that each finding later stands on evidence, not the feelings of the person conducting it.
Decision checklist
Topic: What sample size is sufficient to make informed decisions. Sinh Vũ Handbook, sinhvu.com
Select each item you find appropriate, then print or save as PDF to take with you.
If you have marked most of the signs above, this is the time to discuss in more detail. Sinh Vũ can help you review and propose a direction.
References
SoPact, Qualtrics: explaining the Cochran formula and adjusting for finite populations. NCBI: summarizing research on saturation points in qualitative interviews. Practical experience from Sinh Vũ's brand audit projects.
Frequently asked questions
I only have 50 responses. Is that usable?
Suitable if you clearly state its limitations. Fifty responses are enough to detect initial trends or guide the next round of in-depth interviews, but not enough to announce percentages as if representing the entire market. The issue is not the number 50, but whether you are honest with the readers about that limitation.
My customer segment is only about 1,000 people. Do I still need 385?
No. When the customer base is limited and known, the required sample size decreases. With a group of 1,000 people, according to the finite population adjustment from Cochran's formula, you only need about 278 responses for the same level of reliability and error. What matters is still the quality of the sample, meaning choosing randomly in the right way, not just gathering the easiest people to ask.
How many in-depth interviews are enough?
There is no hard number, but the principle is to continue until saturation, meaning when subsequent interviews no longer reveal new topics or discoveries. Many studies note that this point is often reached within 8 to 12 interviews for each customer group. If you have multiple segments, each segment needs its own sufficient sample, not combined.