Market Research & Surveys

A respondent's tenth straight agree-scale question doesn't get the same careful consideration as the first — acquiescence and satisficing both quietly substitute effort-minimizing shortcuts for genuine judgment.

A navigation menu organized around "Reporting," "Admin," and "Integrations" often reflects which internal team owns which feature, not how a customer actually thinks about accomplishing their own goals within the product.

The internal logic behind a product's navigation and the logic actual users bring to finding things within it are frequently different — card sorting exists specifically to surface that gap before launch.

Surveying every customer within a handful of selected regional offices is considerably cheaper than surveying an equal number of customers scattered randomly across every office nationwide — and the two approaches don't produce equally precise results.

Asked directly, customers say every feature matters. Forced to choose between realistic product bundles, the same customers reveal a genuine, usable set of trade-off priorities they couldn't articulate any other way.

Most research proposals get priced against a brief that's actually three different possible projects wearing one description. Here's how to price the one you're actually going to deliver.

The same script, read by two different interviewers, can produce meaningfully different response patterns on sensitive topics — a documented effect that has nothing to do with either interviewer doing anything wrong.

A panel member who's answered the same brand-tracking survey a dozen times isn't a neutral, unchanged observer anymore — the repeated measurement has become part of what's being measured.

Some brand and product perceptions live below the level of conscious, articulable belief — projective techniques work around this by asking an indirect question that doesn't require the respondent to already have a ready-made verbal answer.

The identical question, asked first versus asked fifteenth, can produce meaningfully different answer distributions — order isn't a neutral container for content.

A survey filled via quotas to match known population demographics can look statistically representative on paper while still carrying selection risks random sampling was specifically designed to avoid.

Asking a customer to recall how satisfied they were six months ago produces an answer contaminated by everything that's happened to them since — recall isn't retrieval, it's reconstruction.

The errors a team catches by having multiple people touch a study don't disappear when you work alone — they just go uncaught unless you replace that check with something explicit.

A research agency proud of a 2,000-respondent sample can still be delivering a worse answer than a smaller, more carefully recruited study, because sample size and sample bias are entirely separate problems.

Scope creep in independent research is usually discussed as a budgeting and boundaries problem. It's just as often a quiet, direct threat to whether the study can actually answer its original question.

People systematically over-report virtuous behavior and under-report undesirable behavior on surveys, even when told their answers are anonymous — and the bias is predictable enough to design around.

A consumer who says sustainability is their top purchase driver and then buys the cheaper, less sustainable option isn't lying in the survey — stated and revealed preference measure genuinely different things.

Leading language, double-barreled questions, and unbalanced scales don't feel biased when you write them yourself. Here's how to catch them before the study goes live.

Two survey questions describing the exact same policy or product feature, one framed around gains and one around losses, can produce meaningfully different response patterns from the identical underlying facts.

A customer asked how recently they made a purchase will often place it closer to the present than it actually happened — a systematic, well-documented memory distortion, not a random guess in either direction.

A brand can achieve strong aided recognition, being reliably recognized once named, while showing weak unaided recall, rarely coming to mind first without any prompting at all — two genuinely different competitive positions.

A team that runs usability testing on a prototype and treats the results as validation of the underlying concept is answering a question nobody asked.

The same set of open-ended customer comments, coded independently by two analysts without a shared framework, can produce meaningfully different theme summaries — a coding process problem, not a data problem.

A focus group where participants quickly, smoothly reach apparent agreement isn't necessarily revealing genuine, representative consensus — it may be revealing the same groupthink dynamics a too-homogeneous group is specifically prone to.

The convenience of an always-available online survey panel comes from a specific selection process — people who chose to join a panel for rewards — that isn't the same thing as a random cross-section of the population.

Asking a respondent to describe what kind of person a brand would be if it were human can surface genuine underlying associations that direct questions about brand attributes often fail to draw out as clearly.

Straight-lining through grid questions, rushing past a survey's minimum realistic completion time, and failing embedded attention checks are all specific, detectable signs a respondent wasn't actually engaging with the questions.

Self-reported survey data on socially sensitive behaviors systematically diverges from actual behavior in a predictable direction, toward whatever answer appears more socially desirable, regardless of the respondent's actual conduct.

A shopper who notices they're being observed, even subtly, may browse more carefully, linger longer, or make different purchase decisions than they would during an entirely unobserved, ordinary shopping trip.