Churn Interviews: Which Questions Actually Change Your Roadmap?

Why do my churn interviews never produce anything I can act on?

Usually because you asked the wrong person the wrong question at the wrong time. Most churn interviews collect a polite reason that the customer invented on the spot, and a polite reason is not a cause. You end up with a folder of quotes and no decision, which feels like research and functions like theatre.

The fix is not more interviews. It is a narrower question set, a tighter sample, and a rule that says you do not run the interview at all unless you are willing to change something because of it. That last rule kills about half the interviews people plan, which is the point.

I have sat in on these conversations for clients often enough to know the failure is predictable. Someone asks "so why did you cancel?" and writes down whatever comes back. That answer is the beginning of the investigation, not the end of it.

What is a churn interview actually for?

It exists to find the moment the customer stopped getting value, which is almost never the moment they cancelled. Cancellation is an administrative event that happens weeks or months after the real decision. Your job in the interview is to walk backwards from the cancellation to that earlier moment and name it precisely.

That framing changes what counts as a good interview. You are not gathering opinions about your product. You are reconstructing a timeline. What did they try to do, what happened, what did they do instead, and when did they stop opening the tab.

If the output of an interview is an adjective, you failed. If the output is a date and a specific attempted task, you have something a product team can argue about. Adjectives cannot be built against.

How much churn is normal before I start panicking?

Recurly's published benchmarks give you a reference point. Its research page states that the Recurly network shows a median annual churn rate of 3.22% for SaaS businesses, using July 2026 data, with top-quartile performers coming in at 1.78% or below. That is the neighbourhood, not a target for your specific business.

Recurly also frames the bands usefully. It says 2% to 4% annual churn is the range where most well-run subscription businesses operate, and that above 5% annual churn is worth investigating regardless of vertical. So if you are sitting at 6%, you have a real problem rather than a benchmarking problem.

One arithmetic trap is worth naming, because founders fall into it constantly. Recurly points out that a 2% monthly churn rate translates to roughly 22% annual churn, not 24%, because of compounding. Small monthly numbers are much worse than they look when you annualise them properly.

Recurly is not the only place publishing this kind of data, and ChartMogul and Paddle both put out their own SaaS benchmark reports worth reading alongside it. Recurly is also honest about the limits of its own benchmarks. It says the most useful comparison is against businesses with a similar customer profile, not against an industry average that blends premium B2B SaaS with low-ARPC direct-to-consumer products. Take that seriously before you present any benchmark to a board.

Which churn is even worth interviewing about?

Not all of it, and this is where most teams waste their effort. Split your churn into voluntary and involuntary before you schedule a single call. Voluntary means someone decided to leave. Involuntary means a payment failed and nobody decided anything.

Interviewing involuntary churn teaches you nothing about your product, because there was no dissatisfaction to investigate. Recurly makes the practical point directly: if your SaaS business is running above 5% annual churn, involuntary churn is worth auditing first, since it is often fixable without any product changes. Go and read your failed payment records in Stripe before you interrogate your roadmap.

Within voluntary churn, prioritise accounts that looked like your best customers on the day they signed and were gone inside a year. Those are the expensive ones and the informative ones. A customer who was never a fit churning on schedule is not a mystery worth an hour of your time.

Which questions produce answers I can build on?

Ask about behaviour in the past, not intentions in the future. Five questions do most of the work: what were you doing before us, what made you go looking, what did you expect to be able to do, what actually happened the last time you tried, and what are you using now. Every one of those has a factual answer.

The most valuable of those five is the fourth. "Tell me about the last time you opened it" produces a story with a date, a task and an obstacle in it. "What could we do better" produces a feature request that tells you about their imagination rather than their experience.

The fifth question is the one people skip out of pride, and it is the sharpest. What someone switched to tells you what job they were really hiring you for, which is the whole premise of jobs-to-be-done. If half your churned accounts moved to Google Sheets, Excel, Notion or a shared Airtable base, you were never competing with the software you thought you were competing with.

Then ask one closing question and shut up: what would have had to be true for you to stay. It invites a specific counterfactual, and the answers cluster fast. This is the same discipline that makes win-loss interviews genuinely useful rather than a sales post-mortem ritual.

Which questions should I stop asking?

Stop asking for a satisfaction score. A number from someone who has already left is noise, because they have no incentive to calibrate it and you have nothing to compare it against. It also trains the conversation to be an evaluation rather than a story.

Stop asking whether they would recommend you. Stop asking them to rank a list of your features. Stop asking what they would pay. And stop asking whether they would come back if you built a particular thing, because the answer is yes and the answer means nothing.

Above all, stop offering the reason in the question. "Was it the price?" gets you price, every time, because price is the socially easy answer and you handed it to them. Price is the reason customers give when the value was unclear, and accepting it at face value is how roadmaps get built around discounts.

How do I get a churned customer to talk to me at all?

Ask fast and ask small. The window where someone still remembers the detail and does not yet resent the relationship is short, and it closes faster than most teams schedule around. Reach out within days of the cancellation, not at the end of the quarter when someone finally runs the report.

Make the ask specific and cheap. Fifteen minutes on Zoom or Google Meet, a named person, one sentence explaining that you are not trying to win them back. That last clause matters more than anything else in the message, because the reason people ignore these requests is that they expect a save attempt.

Then honour it. If you use the call to pitch, you have burned the channel for everyone who comes after you, and word travels in small markets. I would rather lose the account cleanly and keep the honest feedback loop.

Do not pay for the interview if you can avoid it. Paying selects for people who want the money, and in B2B the people worth hearing from will give you fifteen minutes because someone asked them properly.

How do I turn ten interviews into one decision?

Stop when the stories repeat, which usually happens sooner than people expect. Code each interview into a timeline and a blocked task rather than a theme, then group by the blocked task. Themes produce slide titles. Blocked tasks produce tickets.

Then apply one filter before anything reaches a roadmap. Does this blocked task appear in accounts you actually want more of? A loud problem inside a segment you are trying to leave is a distraction, and it is the most common way churn research pulls a product sideways.

Write the finding as a sentence with a consequence attached. Not "onboarding is confusing" but "accounts that did not connect a data source in week one churned, and nothing in our first fortnight of email asks them to." That version tells you what to change, and it is exactly the kind of gap that a properly sequenced onboarding email flow exists to close.

Some findings are marketing problems rather than product problems, and it is worth saying so out loud. If churned customers describe a product that does something you never claimed, your positioning sold the wrong thing, and no feature fixes that. Some of it is fixable with content aimed at retention rather than acquisition.

What should you do next?

Pull your last twelve months of cancellations out of HubSpot or Salesforce, split them into voluntary and involuntary, and count. If involuntary is a meaningful share, go fix payments first. If it is mostly voluntary, pick the six highest-value accounts that left inside a year and send six short interview requests this week.

Run them yourself if you are the founder. Ask the five behavioural questions, write down dates and tasks instead of adjectives, and stop when you hear the same story a third time. Then change one thing. One real change beats a research document nobody reads.

If you want help designing the question set, or you have a stack of churn interviews and no idea what they are telling you, get in touch. I am happy to read what you have and tell you honestly whether there is a finding in there.