Customers are leaving. How do you find out whether the product is the problem?
Three customers cancel. One mentions a difficult setup. A feature request appears in the backlog, and the team starts discussing a redesign. Before committing, ask a useful question: what evidence connects that difficulty to the decision to leave?
Customer churn means losing customers under a defined rule, such as a subscription ending without renewal. It describes an outcome. It does not explain the reason.
This guide shows how to compare customers who stayed and left, keep missing evidence visible and choose a next step. You do not need a predictive model to begin. You need a question the available evidence can help answer.
Look beyond the customers who stayed.
In his 28 October 2026 Mind the Product article, Jay Nathan highlights the risk of learning only from retained customers. Looking only at successful users can hide difficulties experienced by people who leave. This guide includes a separate, small comparison exercise to investigate competing explanations.
Start by defining the group and decision window. For example, compare accounts on the same plan that reached their first renewal in the same quarter. A customer leaving after three years may be in a different situation than someone who abandons setup.
Keep the unit consistent. An account can have several users, so low activity from one person doesn’t automatically mean an inactive customer. Record what “left”, “stayed” and “not yet decided” mean before comparing them.
A worked example: six accounts, several explanations
Illustrative example: a small stock-planning service investigates accounts on its starter plan. The table covers their first 30 days of setup and the decision at their first 90-day renewal. The records and explanations are constructed for this exercise.
Importing a stock list is one setup step. The team initially suspects import difficulty explains cancellations. It combines event records with follow-up conversations, keeping the source of each observation clear.
| Account and setup | Follow-up evidence |
|---|---|
| A / Left Import failed twice. | Customer reported giving up during setup. |
| B / Left Import completed. | Customer reported a budget reduction. |
| C / Left Import completed. | Customer needed offline use, which the service does not offer. |
| D / Stayed Import failed twice. | Customer used a weekly manual workaround and reported extra effort. |
| E / Stayed Import completed. | Customer described using the stock plan in its weekly process. |
| F / Pending Import events missing. | Renewal has not happened; no interview yet. |
A gives the team a specific setup problem to investigate. D is the counterexample: the same recorded failure appears in an account that stayed. Retention does not make D’s extra effort acceptable, and import failure alone does not explain every cancellation.
B and C point towards different questions about affordability and fit. Check their reported reasons in context, rather than automatically treating them as a discount or an offline-feature commitment.
This small, selected set can’t show how common a cause is across the customer base or prove causation. Its value is identifying what to investigate next. Do not turn it into a percentage-based forecast.
Turn the comparison into useful research.
1. Keep observations separate from explanations
Purpose: Make each claim traceable.
Write down the source and date beside each note. “Two failed imports recorded” is an observation. “The import experience caused cancellation” is an explanation that needs more evidence. A customer’s account of their reason is valuable, but still represents their perspective.
GOV.UK’s guidance on analysing research distinguishes what researchers saw or heard from findings and proposed actions. Keep those layers separate in your working notes. If two sources disagree, preserve the disagreement and investigate it.
2. Ask about the decision without supplying the answer
Purpose: Understand what happened in the customer’s context.
Invite customers through your agreed research process, explain the purpose and obtain appropriate consent. Include a contrasting retained account where possible. People who reply may differ from those who do not, so record recruitment gaps.
GOV.UK’s in-depth interview guidance recommends open, neutral questions and concrete examples. Try these prompts:
- “Talk me through the last time you tried to complete this task.”
- “What happened next, and how did you deal with it?”
- “When did you start considering another option?”
- “What else influenced the decision to continue or leave?”
Avoid asking whether your proposed redesign would have made them stay. That puts your explanation into the question and asks the person to predict an alternative past.
3. Choose an action that matches the evidence
Purpose: Address a supported problem without promising to fix churn.
For A and D, inspect the import failure and observe the recovery task. If the failure is reproducible, a small repair or clearer recovery path may be justified because it helps people complete setup. Define that outcome before claiming it will improve renewal.
For B, check whether a budget change reflects a wider pattern before changing pricing. For C, check whether offline work fits the intended market and product direction. For F, check the event coverage and wait for a valid decision window. These are different investigations.
| Finding to explore | Next check |
|---|---|
| Import failure appears in A and D. | Observe import and recovery; confirm the failure mechanism. |
| B reported a budget reduction. | Understand what changed and compare similar accounts. |
| C needs offline use. | Check the intended audience and the underlying task. |
| F has missing events and no renewal outcome. | Check tracking coverage; keep the decision pending. |
Use an evidence-to-decision note.
For one proposed improvement, write a short note using the fields below. Give someone responsibility for the next check and agree when the team will revisit the decision.
Begin with one recent cancellation and one relevant account that stayed. You are looking for a sharper research question, not a representative churn rate. Expand the comparison deliberately as you learn which differences matter.
A useful outcome might be: “The import recovery is difficult, including for a customer who renewed. Test that recovery before deciding the scope of a repair.” That gives the team a practical decision without pretending we’ve settled the reason for every cancellation.
The account records, comparisons and decision note are original teaching material.
You May Also Like
Unleashing Creative Innovation: “Design the Thought”
Your form timed out. Help users check before they submit again
