How many participants do I need
Overview
The right sample size depends on your research method, and on whether you're trying to uncover usability issues or measure something with statistical confidence. This article gives specific starting points for the methods available in Decode.
User Research
- Prototype Testing — 5 to 8 participants per user persona is usually enough to uncover the majority of usability issues
- Task research (live website or app) — 5 to 8 participants for qualitative feedback on usability; 30+ participants if you need quantitative metrics like time-on-task
- A/B & Preference Testing (A/B Test) — 1,000+ users per variant, since you're measuring a small percentage difference and need enough volume to detect it reliably
- A/B & Preference Testing (Preference Test) — 20 to 50 participants is usually enough to get a clear, reliable winner between options
Consumer Insights
- Ad Testing — 100 to 150 respondents per ad concept, to reliably measure recall, appeal, and purchase intent
- Concept Testing — 100 to 200 respondents per target segment, to validate whether a concept has real market viability
- NPS — 100 to 400 respondents; larger samples reduce the volatility that's inherent in how NPS is calculated
- Conjoint — 200 to 300 respondents per market segment, to accurately calculate how participants trade off different features against each other
Shopper & packaging
- Shopper testing — 30 to 50 participants for qualitative eye-tracking and walkthroughs; 150+ participants if you need quantitative virtual shelf results
- Packaging design testing — 100 to 150 respondents, to evaluate findability, standout appeal, and brand alignment
Diary Study and in-context research
- Diary Study — 15 to 25 participants; expect a higher drop-out rate than usual over the course of a longer study
- In-context social testing — 8 to 12 participants; this kind of observational data is dense and takes longer to analyse per participant, so smaller samples are typical
Interview methods
- Live Interviews — 15 to 30 participants per segment, to reach the point where new interviews stop revealing new themes
- AI Interviews (AI Moderator) — 50 to 100 participants; AI moderation scales more easily, so you can treat the resulting qualitative data more like a quantitative survey than you could with live interviews alone
Add a buffer
Always add 10% to 30% on top of your recruitment goal, to account for drop-offs, no-shows, or unusable responses.
What determines your final number
The benchmarks above are starting points. Your actual sample size should also account for:
- Subgroup comparisons — make sure each group has enough participants to support the comparison
- Number of concepts — if participants are split across multiple concepts, make sure each one gets enough participants for analysis
- Audience diversity — you may need more participants if your target audience contains substantially different segments
Example: If you're evaluating three concepts using a monadic design, each participant only sees one concept, so make sure each concept individually receives enough participants, not just the study as a whole.
Rule of thumb
Start with the benchmark for your method, then adjust upward for the number of groups, concepts, or segments you need to analyse separately.
What to do next