Surveys and quantitative signals

The shortest lesson in the module, on purpose

Most marketing advice tells you to survey your customers. This lesson's first job is to tell you when not to.

A survey answers exactly one kind of question: how many? Interviews and review mining discover what is going on — the pains, the triggers, the words. A survey measures how many people it's going on for. That's genuinely valuable: before you bet your two channels and your homepage message on three conversations, knowing whether a finding is common or a fluke is worth ten careful minutes.

But the order is fixed. Interviews discover; surveys count. A survey can only answer questions you thought to ask, so if you run it before you've discovered anything, you will measure your own assumptions to two decimal places and call it evidence. That is worse than no survey, because it arrives dressed as proof.

When a survey is not worth running

Three honest disqualifiers. If any applies to you this week, draft the survey (the exercise below) and don't field it yet. That is a pass, not a failure.

You have nothing to count yet. No interviews, no mined quotes, no candidate finding. Go back a lesson.

You have no audience to send it to. Take Herzog Physio. Tomas has treated roughly 2,000 patients over the years — and can email none of them, because nobody ever collected an address. His Instagram has 340 followers, last posted to in March. His realistic survey is a card at reception, drip-feeding from ~45 new patients a month; at plausible completion rates he'd wait a month or more to reach a usable sample. For Tomas, this month, two patient interviews teach more, faster, and the survey waits until the email list he's building from the front desk exists. If your reachable audience is under about a hundred people, you are in Tomas's position.

The decision wouldn't change. If 30% and 70% would both lead you to the same action, skip the measuring.

Ostara's survey — what n=63 can and can't say

Mateus is in the opposite position: an owned list of 2,300 people who bought or subscribed, with a 31% open rate on the rare sends. After two interviews and his review mining, he has real things to count. His five questions:

1. Who was your most recent order for? — myself / a gift / both (single choice — a fact)
2. How did you first come across Ostara? — Instagram / a friend / a blog or article / search / can't remember (single choice)
3. What nearly stopped you ordering? (open text — the highest-value open question in e-commerce)
4. How confident were you that your order would arrive intact? (1–5 scale, both ends labelled)
5. Where do you look for gift ideas? — gift guides & blogs / Instagram / Pinterest / search / in shops (multiple choice)

Sixty-three people respond. Notice what the design did: every question quantifies something the interviews and mined reviews had already surfaced — the gift occasion, the discovery route, the breakage worry. Nothing in it fishes for new topics; the one open question catches whatever the closed ones missed. The results feed directly into the synthesis you'll do in Lesson 2.6.

Now the honesty. Sixty-three responses from a 2,300-person list is the engaged slice of an already-loyal audience — the people who open a quarterly email are not a fair sample of buyers, let alone of the market. Unhappy customers and one-time buyers are underrepresented by construction. So Mateus reports every figure with its n attached, treats the results as directional, and looks for agreement with the other methods rather than certainty from this one. Twelve responses would deserve even more caution: at that size, one person is eight percentage points. There is no magic threshold that turns a survey into proof; there is only stating your n and refusing to claim more than it supports.

The four ways to poison a question

Each error, with a broken and a fixed version from Ostara.

Leading. ❌ "How much do you love the new Sal collection?" — the verdict is pre-installed. ✅ "How would you rate the Sal collection?" on a balanced 1–5 scale.

Double-barrelled. ❌ "Was delivery fast and the packaging beautiful?" — two questions, one answer box; a yes tells you nothing about which half. ✅ Split it in two, or keep only the one you'll act on.

Hypothetical. ❌ "Would you pay €120 for a four-piece dinner set?" — the same polite fiction Lesson 2.4 banned, now at scale. ✅ "What did you last spend on tableware, roughly?" Past behaviour, even in a survey.

Unbalanced scale. ❌ Good / Very good / Excellent / Outstanding — a scale with no way to complain manufactures satisfaction. ✅ Equal room on both sides, worst to best, ends labelled.

Formats are simple: single choice for facts, 1–5 scales for attitudes, at most one open-text question. Five to seven questions, answerable in under three minutes — every extra minute costs you respondents, and the people who abandon are not random, so long surveys don't just shrink your sample, they bend it.

GDPR in two rules, and three free tools

Rule one: anonymous by default. Collect no personal data unless you truly need it — an anonymous survey is the simplest compliant design there is. Surveying your existing customers about your own service is generally legitimate-interest territory; state who you are and what it's for.

Rule two: never bundle consent. "Submit (you'll also receive our newsletter)" is non-compliant. Adding respondents to any marketing list requires a separate, unticked, explicit opt-in. Keep the survey and the list apart.

Tools, all free at this scale: Google Forms (free, unlimited responses), Tally (generous free tier, EU-hosted option), LimeSurvey (open source, self-hostable, for the strictest data-residency needs). The tool is the least important decision on this page.

Draft yours now — five questions, one stated purpose, and an honest call on whether it fields this week or waits.

---