Cancel surveys: what to ask and what to skip
Growth without waiting on devExplainer6 min readMiniwall team
Ask one question: what's the main reason you're leaving? Give people four or five fixed answers, one of them "something else", and an optional box for their own words. Skip everything else. And never let the survey stand between a person and the cancel button.
That's an opinion from people who build screens for a living, not a research result. The rest of this post is the reasoning, so you can disagree with the right part.
One question is enough
Someone who has decided to cancel is done with your product for now. They'll give you about five seconds. A survey with a rating, a reason, a feature checklist and an email field spends all of that on the first screen and gets abandoned halfway, or answered at random to get past it.
The one question worth the five seconds is why. Not how satisfied they were, not whether they'd recommend you. A satisfaction score from a person who is leaving tells you they're leaving. The reason tells you what to do about the next person.
Fixed answers first, free text second
Free text alone sounds richer and works badly. Most people won't type a paragraph, and the ones who do are not a typical sample. Fixed answers get a response from nearly everyone and can be counted without reading anything.
Keep the list short and make the answers mutually exclusive, so a person can choose without thinking. This is the set we'd start from:
| Answer | What it often points to | What you might do with it |
|---|---|---|
| It's too expensive | The price, or a plan that doesn't match how they use you | A cheaper plan or a discount can be a fair reply here, and only here |
| It's missing something I need | A gap in the product | Route it to whoever decides the roadmap; reply to the person if they leave an address |
| I don't use it enough | Weak habit, or a seasonal need | A pause, or a reminder of what they did get done |
| I switched to something else | A decision already made | Ask nothing more. A "which one?" box is optional, never required |
| Something else | You missed a reason | Read every one of these. If a pattern shows up, it becomes a fixed answer |
Swipe sideways to see every column.
"Something else" earns its place. It's how you find out your list was wrong. Don't hide it, and don't make people type to use it.
Then an optional box for their own words
Below the choices, one text box, clearly optional, with a prompt like "Anything we could do better?" Some people will write two sentences that explain more than the whole table. Most will skip it, and that's fine. If it's required, you'll get "asdf" and a bit of resentment.
What to skip
- Ratings and scores. See above. They measure mood, not cause.
- Long forms. Every extra field costs you answers on the fields that matter.
- Required text. Forced typing produces junk and annoyance.
- Questions you already know the answer to. Plan, sign-up date, last login. You have those. Don't ask the person to type them back to you.
- Asking for an email or phone number. They're leaving. A request for more contact details reads as the start of a sales call.
- Several "are you sure?" steps in a row. One chance to reconsider is respectful. Three is a maze.
Don't make it a trap
A cancel survey is a courtesy to your future self, and only people who feel treated fairly will answer honestly. So:
- The way to cancel stays visible on the survey itself, with words that say what it does. "Cancel my subscription" next to "Keep my subscription" is clear. Two buttons that look alike, or a cancel link in pale grey, isn't.
- If the survey fails to load or the person closes it, they can still cancel. A broken survey must not become a broken cancellation.
- Answering should be a choice. If your survey requires an answer before the person can move on, test what happens to someone who doesn't pick one, and decide whether you're comfortable with it.
- The offer, if you make one, is one screen, and "no thanks, cancel" is just as easy to hit as "claim it".
Cancelling is also governed by rules that differ by place and by store. We haven't covered them here, and you should check the ones that apply to where you sell before you design the flow.
Don't answer every reason with a discount
The easiest follow-up to build is a coupon, shown no matter what the person picked. It's also the worst. Someone who says the product is missing a feature doesn't want 20% off a product that still misses the feature. Someone who switched to something else isn't coming back for a coupon. And if every cancellation produces a discount, people learn that clicking cancel is how you get one.
Match the reply to the reason. Price gets a conversation about price. "Don't use it enough" might get a pause instead of a goodbye. "Missing something" gets a note to your team. Some reasons get nothing but a polite thank-you, and that's a valid result.
Reading what comes back
Count the fixed answers first, then read the free text. If you have a few dozen answers, treat the shares as rough. Ten people picking price out of twenty-five is a hint, not a distribution, and the same warning applies here as in A/B testing with little traffic: small counts swing a lot.
Two numbers beside the reasons are worth having. How many people saw the survey, and how many of those answered. If only a fraction answer, the reasons you have come from the people who bother to answer, and they may not look like everyone who left.
If you do try an offer for one reason, decide beforehand what you'd count as it working, and read the result the way we describe in reading experiment results without fooling yourself.
When this doesn't fit
If cancelling happens on a screen you don't own, there's nowhere to show a survey, and a short email afterwards is the closest thing. If you have a handful of cancellations a month, skip the tooling and ask each person directly. A reply to one human is worth more than a chart of five.
