Surveys
A survey is a structured set of questions put to customers or prospects at scale: satisfaction scores, purchase intent, cancellation reasons, open-ended feedback. Surveys became the default qualitative tool because they are cheap to run and produce numbers that fit in a report. The limitation is built into the format: a survey records what a respondent chooses to state, in the framing the question hands them, and the stated reason for a decision is frequently not the reason the decision was made. The size of the gap has been measured outside marketing entirely: a University of Chicago study of location data from more than two million cellphones found that while about 22% of Americans report attending religious services weekly, roughly 5% actually do (Pope, "Religious Worship Attendance in America: Evidence from Cellphone Data," 2024). People misreport even behavior they have no commercial reason to misreport.
How it actually works
The standard version is a quarterly satisfaction survey and an exit form on cancellation. The results get averaged into scores, the scores go into a deck, and the open-ended responses, which hold most of the value, go largely unread because they do not aggregate.
The gap the format creates is between manifest and latent sentiment. When a customer cancels and writes "I can't afford it," that is the quick answer that ends the interaction. It is manifest sentiment: the stated reason. The latent reason is often that they already found a competitor with a better answer to their problem, and price is the simpler thing to write down. A survey that stops at the stated reason records a rising count of price objections while the actual cause, a competitive gap, goes unmeasured. Acting on the manifest layer then produces a discount program for customers who were not leaving over price. The discount does damage twice: it prices against a stated reason that was not the real one, so the churn driver stays in place, and it resets the reference price for the retained base, compressing margin on every future purchase from customers who would have paid full price. Stated intent has the same gap looking forward. As our founder puts it about Blockbuster: "I'm sure if I was asked in the Summer of 2008 if I'd planned to rent a movie from Blockbuster in the next 12 months, I would have said yes." By the next year he and his friends were browsing Netflix ([the three fits](/insights/s-curve-of-growth#the-three-fits)).
Design decides how much of this a survey can recover. Leading questions produce the answers the author expected: ask "how much do you love the new feature" and the scale itself argues for a positive reply. Better instruments ask about behavior rather than opinion, ask what the customer did the last time the problem came up instead of what they would do, and pair closed scores with open-ended prompts that get read one by one. Even then, a survey works best as one input among several, checked against interviews, reviews, and what the purchase record shows people did.
In practice
In the client program that took acquisition cost from $52 to $13.50, survey responses were one input, weighed against reviews, community discussion, and behavior. No single instrument carried the conclusion; the stated answers were checked against what customers did.
Where we come in
We design survey instruments against the known failure modes: behavior-based questions, no leading frames, open-ended responses read in full, and every result checked against interviews and the purchase record before it changes a decision. At Old School Labs, where our founder served as CMO, that practice supported a turnaround that lifted revenue 87% in ten months.
Start a Revenue Health Pre-Assessment →See it in action
Related terms
- Manifest vs latent sentiment
- The stated reason versus the reason underneath it. "I can't afford it" on an exit survey is manifest; the competitor the customer already chose is latent.
- Leading question
- A question whose framing pushes the respondent toward an answer. It inflates the result in the direction the author expected.
- Response bias
- The set of distortions between what respondents experience and what they report: social desirability, recall error, and the tendency of the angriest and happiest customers to answer most.
- Open-ended response
- A free-text answer in the customer's own words. It resists averaging, so it often goes unread, and it holds the detail the closed scores compress away.
- Net promoter score (NPS)
- A single loyalty score from one survey question. It compresses everything customers could report into one number, which makes it easy to track and limited as a diagnostic.
- Stated vs revealed preference
- What people report they would choose versus what their purchases show they chose. When the two conflict, the purchase record is the more reliable account.