A customer clicks cancel. A cancellation survey appears with five radio buttons and a free-text box, they pick “too expensive” because it is the shortest route to the door, and the answer lands in a dashboard nobody opens. This is the subscription version of the form: the questions a SaaS company shows when someone ends a paid plan. Not an HR exit interview, not an event registration refund.
Most of these forms are built to make the company feel better rather than to produce an answer anyone can act on. They ask at the worst possible moment, in the vaguest possible terms, and then treat the result as a diagnosis. Getting something usable out of the exit takes a different design.
What is a cancellation survey?
Cancellation survey: the short set of questions a subscription business shows a customer at the point they end a paid plan, or shortly after, to record why they left.
It has three honest jobs. It sorts leavers into countable buckets so you can watch the mix change over quarters. It captures language you did not anticipate, in the customer’s own words. And it separates a pause from a stop, which is the difference between a win-back email and a roadmap item.
It has one job it cannot do, and this is where most teams go wrong: it cannot tell you why churn happened. It tells you what the customer was willing to say on the way out.
Why does the reason people give not match the reason they left?
Because people reconstruct a plausible explanation rather than retrieving a real one. This is not cynicism about customers, it is one of the most replicated findings in psychology. In Telling More Than We Can Know, published in Psychological Review in May 1977, Richard Nisbett and Timothy Wilson reviewed the evidence and concluded that when people report on their own mental processes “they do not do so on the basis of any true introspection”. Instead, their reports rest on “a priori, implicit causal theories” about what would plausibly have caused the behaviour.
The same paper gives you the rule for when a self-report can be trusted. Accurate reports occur, Nisbett and Wilson write, “when influential stimuli are salient and are plausible causes of the responses they produce”. Read that as a filter on your own data.
A price increase three weeks ago is salient and is an obvious candidate cause, so “too expensive” from that cohort is probably close to true. A confusing first week six months ago is neither salient nor an obvious cause to the person answering, so it will never appear in your results, no matter how many people it quietly churned. The answers you collect are systematically biased toward recent, nameable events.
The practical consequence: never read a cancellation survey on its own. Put the stated reason next to what the account actually did. Seats that went unused from week two, a feature never activated, a support thread that went quiet. Where the story and the behaviour agree, you have a finding. Where they disagree, the behaviour is the better witness.
When should you ask, at cancellation or after?
Ask during the cancel flow to learn what triggered the decision, and follow up after the subscription ends to learn what shaped it. They are different questions and they need different moments.
| Moment | Captures well | Misses | Best used for |
|---|---|---|---|
| During the cancel flow | The trigger, the mood, and a countable reason from nearly everyone who leaves | Anything the customer has not consciously connected to the decision | Bucket counts you can trend month over month |
| A few days after the plan ends | Calmer, longer answers, and what they switched to | Volume, because the audience has already left and has no reason to reply | Verbatim language, competitor intelligence, win-back signals |
| Before anyone cancels | The slide, while it is still reversible | Nothing about the people who never respond to anything | Catching the drop early, which is the only version that saves revenue |
That third row is the one worth building toward. By the time the cancel button is clicked the decision is made, and a survey at that point is documentation rather than retention. Teams that track satisfaction in context and catch drops before churn get the same information while it can still change something.
Which questions produce an answer you can use
Keep the required part to one question and earn every question after it. Here is a set that works, with the reason each one is there.
- “Which of these is closest to why you are cancelling?” Single choice, six to eight options, plus “something else”. This is the only required question. Forced choice is what makes the result countable across months, and the count is the whole point of the exercise.
- A follow-up conditional on the bucket they picked. Someone who chose “missing a feature” gets “which one”. Someone who chose price gets “what would the right price have been”. A generic follow-up gets generic answers.
- “What were you trying to get done when you signed up?” Free text, optional. This catches the mismatch that never shows up in a reason list: the product worked fine and was bought for the wrong job.
- “What would have had to be true for you to stay?” Free text, optional. The counterfactual pulls better answers than “why did you leave”, because it asks for a specific change rather than a verdict, and it is the question most likely to name something you can build.
- “Is this a stop or a pause?” Yes or no. It costs one click and it routes the account: win-back sequence, or roadmap evidence.
- “Can we email you in a month to ask how it is going?” Yes or no. This is how you get the honest version later, once the decision has stopped being defended.
Question five and six are the ones most teams leave out, and they are the two that decide what happens next. For wording on the open questions, our product survey questions bank has 100 of them sorted by what each one measures.
Which question type tells you what
Every type buys you something and costs you something. Mixing them on purpose beats defaulting to a radio list.
| Question type | What it actually tells you | Where it misleads |
|---|---|---|
| Single choice, fixed list | The share of leavers per bucket, comparable month to month | Every reason outside the list gets forced into the nearest wrong box |
| Multiple choice | That reasons stack, which most of them do | Nothing is ranked, so you cannot tell the trigger from the background |
| Free text only | Language you did not anticipate, in their words | Most people skip it, so you hear the angriest and the fondest |
| Rating scale at the exit | Intensity, and a number that trends | A score on the way out is a mood, not a diagnosis |
| Conditional follow-up | Real detail behind a bucket they already committed to | It can only ever explain the reason they were willing to state |
| Stop or pause | Whether to spend on win-back or on roadmap | People overstate how likely they are to return |
Our post on customer satisfaction survey questions goes deeper on wording and scale choice if you are building the question bank rather than just the exit form.
Keep it short enough to be finished
Length is the single biggest lever on whether you get data at all. SurveyMonkey analysed 100,000 surveys, sampling 2,000 for each question count from one to fifty, and found that the sharpest rise in abandonment happens inside the first fifteen questions rather than at some far-off limit. Drop-off is front-loaded, so the fifth question costs you more respondents than the thirty-fifth.
At the exit, that curve is steeper still, because the person answering has already decided you are no longer worth their time. Baremetrics, writing from their own subscription analytics practice, recommends a single required question for the cancellation reason with any additional detail collected by email afterwards. That is the right shape: one required question, everything else optional and skippable in one click.
A cancellation survey must never block the exit
A cancellation survey is never a condition of cancelling, and the legal position on this is sharper than most teams assume.
The FTC’s amended Negative Option Rule, the one the press called click to cancel, was vacated by the Eighth Circuit on 8 July 2025 in Custom Communications, Inc. v. Federal Trade Commission, No. 24-3137. The court struck it on procedure, not on principle: the Commission had skipped the preliminary regulatory analysis the FTC Act requires. The opinion is explicit that this was a process failure, saying “we certainly do not endorse the use of unfair and deceptive practices in negative option marketing” before concluding “we grant the petitions for review and vacate the Rule”.
So the federal rule is gone and the reasoning behind it is not. State law did not move at all. California’s Automatic Renewal Law requires that a customer who signed up online can end the subscription “exclusively online, at will, and without engaging any further steps”, and where a business shows a retention offer during cancellation it must keep a click to cancel link continuously and proximately displayed throughout. If you sell to consumers in California, a survey that sits in front of the exit is not a growth tactic, it is exposure.
Design for it directly. The survey sits beside the cancellation confirmation, not in front of it. Every question is skippable. The cancel button stays visible and active on the same screen the whole time. You will lose some responses. You will also stop collecting the kind of answer a trapped person gives, which was never worth having.
What to do with the answers
Answers that stay in a survey tool are a reporting exercise. The two moves that make them matter are routing and repetition.
Routing means each response leaves the survey with the account attached and lands where the owning team already works: the churn reason on the account record, the feature request in your feedback backlog, the pricing complaint in front of whoever owns pricing. Sleekplan surveys can trigger by URL, segment, plan, action, or anything in your CRM, so a cancellation-shaped audience is a targeting rule rather than a separate tool, and responses can be piped onward by webhook when they come in.
Repetition means treating a single answer as an anecdote and the same answer forty times as evidence. One person saying the reporting was too thin is noise. Forty of them across a quarter, concentrated in accounts on one plan, is a roadmap item with a revenue number attached. That is the only form in which cancellation data has ever changed a product.
Start with three questions
Cut whatever you have to one required single-choice reason, one conditional follow-up, and the stop-or-pause toggle. Ship that this week. Read the buckets next to usage data rather than on their own, and watch the mix rather than any single month’s winner.
Then build the earlier version, the one that asks before the cancel button gets clicked. The exit survey tells you how you lost. Only the earlier one gives you a chance to not lose.