survey-designlisted
Install: claude install-skill teodstr/garage
# Survey design
A survey can only measure what interviews have already uncovered, and a badly written one measures the questionnaire instead of the market. This skill builds short, neutral surveys and reads the results with the right caution.
## When to use
- After `customer-interviews`, to measure how common a problem, a habit or a preference is.
- To compare segments or track satisfaction over time.
- Not to discover new problems, and not to forecast sales: "would you buy it?" answers overstate real demand.
## Inputs
1. The decision the survey informs.
2. The target population and how to reach it.
3. Constraints: budget, deadline, survey tool.
## Method
1. **One decision, a few numbers**: write the decision and the three to five numbers needed to make it. Every question feeds one of them; cut the rest.
2. **Define the population and the sample**: who exactly, how they will be reached, and the bias of that channel (a post on the founder's social network reaches the founder's network, not the market).
3. **Compute the margin of error** for a proportion, and say that it only holds for a random sample: margin = 1.96 × √(p × (1 − p) / n). With p = 0.5, it is about ±9.8 points for 100 answers and ±4.9 points for 400.
4. **Write neutral questions**: one idea per question, no loaded words, no jargon, a "don't know" or "not applicable" option when it fits, balanced scales (as many positive as negative points), a precise time frame ("in the last 30 days").
5. **Order the quest