Surveys & Quantitative Research

First PublishedMay 24, 2026Last UpdatedMay 25, 2026ByAtif Alam

The decision this page enables: is the pattern you saw in 8 interviews real at the scale of your whole customer base?

What it is

A survey is a structured questionnaire delivered to a defined sample of your audience. It exists to do one thing well: take a hypothesis you already have (usually from discovery interviews or voice of customer) and tell you how widely it holds.

Surveys are bad at discovering what to ask about — the question wording locks in your assumptions. Use qualitative research to figure out the what; use surveys to size and rank the how much.

Why it matters

The ICP moves from a hypothesis to a defensible claim when you can show “62% of users at companies 5-15 reported the same pain.”
The Workbook → KPI baseline benefits from NPS / CSAT / CES as standard, comparable metrics across periods.
Pricing decisions benefit from a Van Westendorp price-sensitivity survey before committing to tier prices.

When to survey vs. when to keep interviewing

Situation	Use…
You don’t yet know what to ask about	Interviews (qualitative)
You have 1-2 sharp hypotheses to validate at scale	Survey (quantitative)
You want to rank options from a known list	Survey
You want to discover new dimensions or motivations	Interviews
You need a metric to track over time	Survey (NPS/CSAT/CES on a cadence)

Surveys are wasted effort before you’ve done interviews. They produce false confidence — beautifully visualized — about questions you didn’t really understand.

How to design a survey that works

A practical 5-step process:

Write the hypothesis first. One sentence. “I believe X is true for Y% of segment Z, because of A.” If you can’t write that, you’re not ready to survey.
Pick the smallest sample that answers the hypothesis. See sample-size priors below — don’t oversample.
Keep it under 10 questions, ≤4 minutes to complete. Completion rate falls off a cliff past ~5 minutes; drop-off after question 8 is heavy on every platform.
Lead with closed questions, end with one open question. Closed = multiple-choice / scale; comparable and countable. One open “anything else?” at the end captures the surprise.
Pre-test with 3-5 humans first. They will misread your wording in ways you can’t predict. Fix the wording before the survey goes live.

Question types in order of usefulness

Multiple choice (single-select) — cleanest to analyze; ideal for ranking options.
Multiple choice (multi-select) — useful but harder to compare; cap the number of selectable options.
Likert scales (1-5 or 1-7) — for sentiment / agreement. Use an odd number so respondents can pick neutral; force-choice (even) only if you’re tracking polarization deliberately.
NPS-style (0-10) — see standard metrics below.
Open text — reserve for the last question, and the one optional “why?” follow-up after the most important closed question.

Sample size priors (rough rules, not statistics)

n ≥ 30 to make a basic point about a single segment (“most of our power users want X”).
n ≥ 100 to cross-tabulate two segments reliably (“startups vs. enterprise”); you need ~30 per cell.
n ≥ 400 to compare changes over time with confidence (a roughly ±5% margin of error around a binary outcome at 95% confidence).
Response rates: email to existing customers, 15-30%; in-app modal, 5-15%; cold/external survey, 1-5%. Budget recruits accordingly.

These are approximations. If you’re making a decision that costs >$50K to reverse, get a real statistician to size it.

The three standard metrics worth running

Metric	Question	Scale	What it measures
NPS (Net Promoter Score)	“How likely are you to recommend [product] to a colleague?“	0-10; subtract % detractors (0-6) from % promoters (9-10)	Loyalty / advocacy. Useful as a trend, not an absolute.
CSAT (Customer Satisfaction)	“How satisfied were you with [interaction]?“	1-5; % responding 4-5	Episodic satisfaction (per ticket, per onboarding step).
CES (Customer Effort Score)	“How easy was it to [complete this task]?“	1-7; lower = easier	Friction during a task. Predicts churn better than CSAT.

Don’t over-fit on the absolute number — NPS especially varies wildly by industry. Use them as trend lines: is the number moving up or down quarter over quarter?

Templates

General “validate-a-pain” survey (8 questions)

1. Which best describes your role?
   ( ) Engineer  ( ) Engineering manager  ( ) Founder/CTO  ( ) Other

2. What size is your team?
   ( ) 1-5  ( ) 6-15  ( ) 16-50  ( ) 50+

3. How often does [the pain] happen to you?
   ( ) Daily  ( ) Weekly  ( ) Monthly  ( ) Rarely  ( ) Never

4. When it does happen, how disruptive is it?  (1 = mild, 5 = blocks work)
   ( ) 1  ( ) 2  ( ) 3  ( ) 4  ( ) 5

5. What do you currently do to work around it?  (select all)
   [ ] Manual workaround  [ ] Built our own internal tool
   [ ] Use [competitor A]  [ ] Use [competitor B]  [ ] Live with it

6. If this problem disappeared, how much time/money would you save per month?
   ( ) <$50 / <2h  ( ) $50-500 / 2-10h  ( ) $500-2000 / 10-40h  ( ) >$2000 / >40h

7. How likely are you to switch to a tool that solved this?  (0-10)
   [____]

8. Anything else we should know?  (optional, open text)

NPS micro-survey (1 question + 1 follow-up)

On a scale of 0 to 10, how likely are you to recommend [product] to a friend
or colleague?
[ 0  1  2  3  4  5  6  7  8  9  10 ]

What's the main reason for your score?
[ open text ]

Metrics to track

Response rate — see priors above; falling response rate over time is the early warning that you’re surveying too often.
Completion rate — fraction who finish the survey vs. start it. Healthy: >70%. Below that, your survey is too long or the early questions are off-putting.
Sample size per segment — for any segmented cut, hit n≥30 before drawing a conclusion.
NPS / CSAT / CES trend — month-over-month or quarter-over-quarter; ignore absolute level, watch direction.
Open-text theme growth — count of distinct themes in the optional comment; growing tells you new pains are emerging.

Examples

SaaS workspace (B2B through-line)

After 12 discovery interviews surfaced “glue-breaking” as the dominant pain, ran an 8-question in-app survey to 800 active users. n=212 completions (26% response rate). Key results:

78% reported glue-breaking happens weekly or more often.
64% rated it severity 4 or 5.
41% reported time savings of >10h/month if it disappeared.

This converted the qualitative hypothesis into a defensible pricing claim (“>10h/month for a majority of accounts”), which then anchored the $29/seat tier.

Consumer fitness app (B2C contrast)

After 14 discovery interviews, ran a 6-question email survey to a 5,000-user mailing list. n=183 completions (3.7% response rate — typical for B2C cold list). Key result:

The “loss of motivation by week 3” pattern held at 70% of respondents.
Cross-tab showed it was 81% for users 25-35 and 58% for users 36-45 — clear age skew worth a product decision.

Cross-tab usefulness required the larger sample; with n<60 the segment cut would have been noise.

Common pitfalls

Surveying before interviewing. You don’t know what to ask. The survey will measure something — but probably not what matters.
Leading questions. “How frustrating is it when…” is a leading question. “How often does X happen?” is not.
Overlong surveys. Past 10 questions or 5 minutes, completion collapses and your sample skews to the most patient respondents.
Reading too much into small samples. A 12-person survey is not data; it’s a slightly more structured interview. Honor the n≥30 floor per segment.
Treating NPS as a target. NPS is a thermometer. Setting it as a goal corrupts the question (sales teams gaming the “0-6 detractor” cutoff is a classic anti-pattern).
Ignoring the open-text answers. The free-text “anything else?” usually contains the next thing to interview about. Tag and cluster it like you would interview notes.