Churn interviews: how to find out why customers leave
By Hey Juno · Published · Updated
Churn interviews are conversations with people who left, nearly left or stayed, designed to find out why customers leave in their own words. The reliable way to get past "it was too expensive" is to ask people to retell the last weeks before they left, not to explain their decision. This guide covers who to talk to, when, which questions work, how to read the answers and how to turn them into retention decisions.
It is written for teams commissioning customer research, including with Juno, the AI researcher from Hey Juno.
Why exit surveys and cancellation reasons fall short
Most teams already have some churn data: a cancellation form with a dropdown, a score, perhaps a free-text box. These tell you which box people ticked. They rarely tell you what happened.
"Too expensive" is the usual example. It is a socially easy answer. It is also a summary of a longer story that could be about price, about value, about a bill that arrived at a bad moment or about a feature that stopped working. A survey cannot ask what the person meant. An interview can.
A form also captures the reason someone gives at the moment of leaving, often in a hurry. A conversation soon afterwards lets people reconstruct what led there.
Who to talk to
Do not interview only the people who left. Three groups give you different information.
Recent leavers. They remember the decision and can name what changed. Aim for people who left recently enough to recall the details. What counts as recent depends on your product's rhythm, so choose a window that matches how often people use it.
At-risk customers. These are people whose use is fading, who downgraded or who have complained. They are still deciding, so they can describe doubts that leavers have already smoothed into a tidy story. Because they are still customers, be clear that you are asking to learn, not to sell.
People who stayed. Without them, you cannot tell which reasons are specific to leavers. If people who stayed describe the same frustrations, the frustration is not what separates the two groups. This comparison is where a lot of the learning sits.
When to reach them
Contact leavers soon after they go, while the sequence is still clear. Contact at-risk customers as the signals appear, not months later.
Make clear it is voluntary, that it will not affect their account and what you will do with their answers. See participant wellbeing in AI research and informed consent in customer interviews.
Questions that get past "it was too expensive"
The principle is simple: ask for events, not explanations. People are better at telling you what happened than why they acted, and a story holds details a summary drops.
The examples below are illustrative. They show the shape of the questions, not findings from a real study.
Start with the last straw. "Think about the day you decided to leave. What was happening?" Then follow with "What had happened just before that?" The point is the specific moment, not the general reason.
Walk back through the weeks. "When did you first think about leaving? What was going on then?" Churn is often a slow slide, and the first doubt usually differs from the final trigger.
Ask what they switched to. "What are you using now instead?" and "What made that look like a better option?" Leaving is also a choice for something else, even if that something is a spreadsheet or doing nothing. The alternative shows what job the product was being asked to do.
Ask what they had stopped expecting. "Was there something you used to expect from it that you no longer did?" This often surfaces quiet disappointments that never reach a complaint form: a report nobody opened, a feature that never arrived, a support reply that took too long.
Unpack the headline. If someone says "it was too expensive", do not argue and do not accept it. Ask "What were you comparing it with?" or "Tell me about the last time you looked at the bill." Often the real subject turns out to be value, not price.
Check what would have made a difference. "Looking back, was there a point where something could have gone differently?" Keep it open. Asking "Would a discount have kept you?" invites a polite yes.
Avoid questions that name the answer or offer a short list of reasons. Leading questions in research shows the common patterns and how to rewrite them.
How to read the answers
Read each conversation as a story first, then compare across them.
Separate the trigger from the cause. The last straw is what tipped the decision. The cause is what made the customer ready to tip. A missed invoice may be the trigger, while months of low use is the cause.
Look at what leavers and stayers share. A complaint raised by both groups is a weak explanation for churn. A complaint raised only by leavers deserves attention.
Keep the strength of evidence visible. Three people describing the same slow disengagement is a lead. Thirty is a pattern, though still not a count of how many customers feel it. Interviews describe why people act. A survey or your usage data tells you how many. See NPS and the why behind the score for one example of pairing them.
Treat any reason as provisional until you have seen it in more than one place. See checking AI research findings.
Recruiting leavers through invite links
Every Juno study has an invite link that participants open in their browser, by text or voice, with nothing to download. For churn work, this means you can send the link to former customers using the email address you already hold.
Juno also supports per-person invite links that carry context from your customer records and notes for the interviewer. That lets Juno start with what you already know, such as the plan someone was on, rather than asking them to repeat it.
Two cautions. First, people who agree to talk to a company they left are not typical leavers. They are often the more engaged or the more aggrieved. Say so when you report. Second, you can only reach people you have details for. If you also want to hear from people who left a competitor, a panel route may help. Hey Juno offers a managed panel service for that. See participant recruitment for AI interviews for the routes and trade-offs.
Turning findings into retention decisions
Aim for decisions, not a list of reasons. A useful output has three parts:
-
What happens. The sequence that leads from satisfied to gone, in the customers' words, with quotes you can trace back to a transcript.
-
Where it could change. The points in that sequence where something you control could have gone differently.
-
How sure you are. Which claims rest on many conversations and which on a few.
Then pick a small change and watch whether it moves what your own data shows. If leavers describe a confusing first month, test a different onboarding and look at retention afterwards. Interviews suggest where to look. Your numbers confirm whether it worked.
With Juno, each claim links back to the participant, the quote and the transcript moment, so you can see whether a recommendation rests on what people said or on an inference. For the wider approach, see voice of customer research.
Where this doesn't apply
Churn interviews are a weak fit when you have very few customers who have left, since patterns need a handful of conversations at least. They do not measure how big a problem is. They also cannot reach people who will not talk to you, who may have left for different reasons. And if the decision is already made, interviewing to confirm it will only confirm it.
Frequently asked questions
How many churn interviews do I need?
There is no fixed number. Themes can start to repeat after a modest number of interviews in a similar group, but you will need more if you want to compare segments or if leavers differ a lot. Start with what you can reach, look for patterns, and keep going until new interviews stop adding new reasons. See qualitative interviews at scale.
Should I offer an incentive to people who left?
Many teams do, because leavers have little reason to give their time. If you offer one, say so up front, and say it does not depend on what they tell you. Do not tie it to staying or returning.
Can I interview at-risk customers without upsetting them?
Yes, if you are honest about the purpose. Tell them you are trying to learn, not retain them, and that answers will not change their account. Do not turn the conversation into a save attempt.
Is an AI interviewer suitable for this?
It can be. Juno interviews real people by voice or text and adapts its follow-ups to what each person says. Check the findings the same way you would with any interviewer. See what taking part in an AI interview is like.