Key Takeaways
- Churn interviews surface the "why" behind cancellation that product analytics cannot — the competitive switch, the specific friction, the moment expectations broke.
- Traditional qualitative interviews don't scale: recruiting, scheduling, transcribing, and coding take weeks. For most product teams, this means churn research happens too rarely and too late.
- Sonderly is an AI customer interview tool that conducts open-ended qualitative interviews asynchronously, then returns thematic analysis across all respondents — with quotes linked to source transcripts.
- A churn study can be live in under an hour. The output identifies dominant themes, their prevalence across the respondent set, and minority signals that survey response options would never surface.
Why Analytics Alone Don't Explain Churn
Sonderly is an AI customer interview tool designed for product teams who need to understand not just what users did, but why they left. This guide walks through how to use it to run a churn study — from designing the interview guide to reading the analysis output — and what the resulting insights look like in practice.
Product analytics can tell you when a user churned, which features they last touched, and how their engagement trended before cancellation. What they cannot tell you is what was happening in a customer's head during that period: the competitive offer that landed at the right moment, the specific workflow failure that finally outweighed the switching cost, or the quiet decision made in a team meeting that never registered in an event log.
Churn is usually a decision made over time and shaped by factors that exist largely outside the product itself. A user who cancels after six months may have started looking at alternatives three months earlier — following a frustrating onboarding experience for a new colleague, a pricing conversation with their CFO, or a competitor demo that changed their frame of reference. None of that appears in a usage dashboard.
The only reliable way to surface this kind of motivation is to ask — in open-ended terms that let respondents describe what actually happened, not select from a list of options a researcher predicted in advance.
The Scaling Problem with Traditional Interviews
Qualitative interviewing is the right method for understanding churn. The problem has always been operational.
Running live interviews at meaningful scale requires recruiting participants who have recently churned (which requires CRM infrastructure and timely outreach), scheduling individual sessions (which requires coordination across calendars), transcribing recordings (which is either slow when done manually or costly when outsourced), and then coding themes across a full transcript set — identifying patterns, grouping similar sentiments, tagging contradictions. For an experienced researcher working alone, a twenty-interview study easily represents several weeks of work across the full pipeline.
The result is predictable: churn interviews happen infrequently, in small batches, and often after a gap long enough that the freshness of experience has faded. Findings from eight conversations conducted two months after a churn event are less actionable than findings from thirty conducted within the week of cancellation.
This is the scaling constraint that AI-powered customer interviews are designed to remove. When the interview is conducted asynchronously by an AI, a churned user receives a link, completes the session at a time convenient for them, and their responses flow directly into analysis — no scheduling, no transcription step, no manual coding. The research pipeline shrinks from weeks to hours.
What an AI-Powered Customer Interview Actually Looks Like
From the respondent's perspective, an AI-powered interview is a conversational exchange — text or voice — that follows a research guide the product team has designed. The AI interviewer asks opening questions, listens to responses, and probes with follow-up questions based on what the respondent says. It is not a fixed survey cycling through a predetermined list; it follows the logic of a conversation, asking clarifying questions, exploring tangents the respondent raises, and covering the topics the researcher specified.
From the researcher's perspective, the interface is the interview guide and the analysis output. You design the protocol — the opening questions, the probing logic, the topics to explore and avoid — inside Sonderly. Once the study is live, Sonderly generates a shareable interview link. You distribute that link through whatever channel reaches your churned users: a cancellation-flow prompt, a win-back email sequence, or a direct CRM outreach. Each respondent who clicks the link completes an independent session. Sonderly captures the transcript, analyzes it alongside all other responses, and surfaces themes across the full set.
The researcher retains access to every transcript. Every theme in the analysis is linked back to the specific quotes and sessions that generated it, so findings can be verified at the granular level rather than taken on faith.
How to Set Up a Churn Study in Sonderly
A well-structured churn study has three components: respondent criteria, the interview guide, and distribution.
Step 1: Define your respondent criteria.
For a churn study, the target population is users who have cancelled or lapsed within a recent window — typically the past 30 to 90 days, depending on your product's natural usage cycle. The closer the outreach to the churn event, the more accurate the recall and the more useful the findings. Sonderly operates on a bring-your-own-audience model: you identify and contact respondents through your CRM, email platform, or cancellation flow, and Sonderly provides the interview link.
It is worth segmenting respondents if your user base has meaningfully different profiles — for example, individual users versus team administrators, or users who churned during a trial versus those who cancelled after a paid subscription. Thematic analysis across a mixed population can obscure segment-specific patterns.
Step 2: Build the interview guide.
This is where the quality of your study is largely determined. A churn interview guide should open with a broad, non-leading question that gives the respondent room to describe the situation in their own terms. A useful opener might be: "Can you walk me through what led up to your decision to cancel?" or "What was happening on your end in the weeks before you decided to stop using the product?"
Follow-up probing should cover, at minimum:
- What the respondent was trying to accomplish with the product at that point
- Whether they evaluated or switched to an alternative — and if so, what that alternative offered
- What specifically felt missing, frustrating, or insufficient
- Whether there was a specific moment or event that tipped the decision
The most important discipline here is avoiding leading questions. Asking "Was pricing a factor?" is far less useful than letting the respondent describe their own factors and then asking what role pricing played if they raise it. The goal is to hear what actually happened, not to confirm a hypothesis.
Sonderly's interface lets you configure the opening questions and the probing logic — specifying which topics the AI should explore when they surface and which to avoid or handle with care (for example, questions that require product knowledge the AI may not have). This gives the researcher methodology control without a live interviewer in every session.
Step 3: Distribute the link.
Once the guide is configured and the study is live, Sonderly generates a shareable interview link. You can embed this in a cancellation confirmation, include it in a win-back email sent a few days post-cancellation, or trigger it through your CRM as part of an automated churn workflow. Each respondent who clicks the link completes an independent, asynchronous session. There is no scheduling required and no coordination overhead.
A study can be live and collecting responses in under an hour from the time you begin setup.
What the Thematic Analysis Output Returns
As responses accumulate, Sonderly's analysis builds across the transcript set. The output organizes findings into themes — groupings of similar sentiments, motivations, or experiences that appear across multiple respondents. Each theme shows its prevalence across the respondent set, includes representative quotes drawn directly from transcripts, and links to the underlying sessions for verification.
To illustrate what this kind of output looks like in practice, consider a hypothetical study of churned users from a project management tool. After collecting responses from 40 participants, the analysis might return something like the following:
Theme: Integration gaps with adjacent tools — The most commonly stated reason for cancellation was friction between the product and the rest of the respondents' tool stack, particularly around task tracking and team communication tools. This surfaced as both a day-to-day inconvenience (maintaining parallel records in two systems) and a switching trigger when a competitor with native integrations became visible. Representative quote: "We just spent too much time updating things in two places. Once we realized [the competitor] connected directly to [our ticketing system], it was an easy call."
Theme: Pricing tolerance varied by team size — Respondents from smaller teams were more likely to describe pricing as a factor, but further probing consistently revealed that the pricing concern was secondary to a capability gap: they described cancelling because the feature set didn't justify the spend at their scale, not because the price was abstractly too high.
Theme: A shared onboarding friction point — A subset of respondents — seven out of 40, unprompted — described frustration during initial setup, specifically around importing existing project data. This was not the stated reason for cancellation in any of these cases. But it appeared as a consistent early negative experience that shaped their overall relationship with the product. None of these respondents were in the same customer cohort or industry vertical.
That third theme — a minority signal, appearing in fewer than 20% of responses, cited unprompted and across otherwise unrelated respondents — is the kind of finding a well-designed survey is structurally unable to surface. It would not appear in a dropdown. No one would have thought to include it as a response option. An AI interview captures it because respondents describe their experience in their own words, and the analysis identifies the pattern across sessions.
(Note: The example above is illustrative of the type of output Sonderly's analysis produces. It is not drawn from a specific published study.)
The Kind of Finding a Product Team Acts on
The integration gap theme in the example above is directly actionable. It names a specific capability gap, identifies the competitor capturing the defecting users, and points to the moment the decision became easy. A product manager can take this into a roadmap conversation with specificity: not "users want better integrations" (a vague direction) but "users who evaluate [Competitor] after experiencing the double-entry problem tend not to come back" (a tractable problem with a competitive frame).
The onboarding friction theme is equally actionable — and more surprising. It points to a specific, reproducible step in the early user journey, suggests the failure mode is not immediately visible in event data (since none of these users dropped off during onboarding), and indicates that the damage accumulates over time. The fix is findable. The cause is clearly stated.
The distinction between this and a survey result is not just depth — it is the quality of the context that makes a finding driveable. A survey might report that 40% of churned users rated "product fit" below expectations. A qualitative interview tells you which specific moment generated that rating, what the user was trying to accomplish, what they said when they found they couldn't, and what they did next.
That context is what turns a research finding into a conversation with an engineering lead. It is what makes a product decision feel grounded rather than directionally correct but uncertain. And it is what scales when you run the same study across enough respondents to know which themes are dominant, which are segment-specific, and which appear only in a minority of cases but consistently enough to act on.
Frequently Asked Questions
What is an AI customer interview tool?
An AI customer interview tool is a platform that conducts structured, open-ended qualitative interviews with users or customers without requiring a live human interviewer in each session. The AI follows an interview guide designed by the researcher, asks probing follow-up questions based on what respondents say, and returns analysis across the full set of responses. Sonderly is an AI customer interview platform purpose-built for product research, UX research, and market research use cases.
How does automated user research analysis work?
Sonderly's automated user research analysis works by processing transcripts from all respondents in a study and identifying recurring themes — groupings of similar sentiments, motivations, or experiences across the respondent set. Each theme is linked to specific quotes and source transcripts, so findings can be verified at the session level. The researcher retains full access to all transcript data; the analysis is a structured view over that data, not a replacement for it.
How many respondents do I need for a churn study?
Qualitative research does not require the same sample sizes as quantitative surveys, because the goal is thematic depth rather than statistical representativeness. In qualitative methodology, the relevant concept is thematic saturation — the point at which new responses stop surfacing new themes. Research on saturation (Guest, Bunce, and Johnson, 2006, in Field Methods) found that dominant themes in a homogeneous population often emerge within the first 12 responses, with saturation occurring well before 30. A churn study with 20 to 40 respondents is typically sufficient to identify the dominant patterns, and an AI interview tool makes collecting at that scale practical without significant time or coordination overhead.
How is an AI customer interview different from a survey?
Surveys capture what respondents choose to express within the constraints of pre-set questions and response options. Qualitative interviews capture what respondents say in their own words — including information the researcher did not anticipate asking about. Churn drivers in particular often involve context that would not fit in a survey field: the conversation at a team meeting that shifted the decision, the specific workflow failure that broke trust, the competitor demo that reframed expectations. AI interviews provide the depth of qualitative conversation with the scale that asynchronous, automated delivery enables.
Can I use Sonderly with my existing CRM or email workflow?
Sonderly operates on a bring-your-own-audience model. You identify and contact respondents through your existing tools — email platform, CRM, or cancellation flow — and Sonderly provides the interview link for each study. You control outreach timing, messaging, and segmentation; Sonderly handles the interview and analysis. See Sonderly's documentation for current information on integrations and automation support.
Is the transcript data accessible after the study?
Yes. Every transcript from every session is accessible inside Sonderly. The thematic analysis output links each theme and each representative quote directly to the source transcript. Researchers can drill into any session, read the full conversation, and verify findings at the granular level.
Conclusion
Churn interviews are one of the highest-leverage qualitative research activities a product team can run. They surface the motivations, competitive dynamics, and specific friction points that analytics cannot capture — and they do so in a form that is specific enough to act on.
The barrier has historically been operational: recruiting, scheduling, transcribing, and coding at any meaningful scale takes weeks. For most product teams, that means churn research happens too rarely and too late to influence the decisions it should.
AI-powered customer interviewing, as Sonderly enables it, removes that barrier. A study can be live and collecting responses in under an hour. Respondents complete the interview on their own schedule. The analysis surfaces themes across the full respondent set, with quotes and source transcripts accessible for verification. The result is qualitative insight at a speed and scale that makes it practical to run a churn study in the window when the findings are most valuable — close to cancellation, when recall is accurate and the experience is still fresh.
If your team has questions about churn that the dashboard can't answer, that is the right starting point. Define the respondent criteria, build the interview guide, and let the conversation surface what the data can't.
Request a free trial with Sonderly →