A retention survey asks an active customer whether the product is still working for them, at a moment chosen because something in their behaviour suggested it might not be. It is a targeted question to a specific account, not a periodic pulse sent to everyone.
That distinction is the whole method. A survey sent on a calendar tells you the average mood. A survey sent because an account's usage halved three weeks after renewal tells you something you can act on this week.
Retention Surveys Against Everything Else
Most teams already run one or two survey types and assume they are covered. They measure different things.
| Survey | Asked of | Answers | Best at | Blind to |
|---|---|---|---|---|
| Retention survey | Active accounts showing a risk signal | Is this still working for you | Catching a specific account before it decides | Accounts with no visible signal |
| NPS | Everyone, periodically | Would you recommend us | Trend over time | Why the number moved |
| CSAT | After a specific interaction | Was that any good | Rating one touchpoint | Anything outside that touchpoint |
| CES | After a task | How hard was that | Friction in a single flow | Value, price, strategy |
| Onboarding survey | New accounts at a fixed point | Did you get set up | First-run friction | Anything after week four |
| Cancellation survey | Accounts already leaving | Why are you going | Reasons, cheaply | Prevention, since the decision is made |
The cancellation survey is the one worth dwelling on. It is the most common thing teams have, and it is the least useful for retention, because by the time it fires the account has evaluated an alternative, made a case internally and picked a date. Its answers are also the most rationalised: people give a reason that is easy to say, not the one that made them start looking. Read what cancellation flow data actually shows as diagnosis of causes, not as an early-warning system.
What Should Trigger One
Send on behaviour, not on a schedule. Each of these is a reason to ask a specific account a specific question:
| Trigger | What it may mean | Ask about |
|---|---|---|
| Usage drops against the account's own baseline | Workflow moved elsewhere, or a champion left | Whether the job changed |
| Onboarding stalls before the activating action | Setup blocked, or the value was not obvious | What stopped them |
| A key feature is never adopted | Wrong fit, or they never found it | What they use instead |
| Repeated support contact on one theme | A problem they are working around daily | The workaround and its cost |
| Downgrade, or seats removed | Value no longer matches the tier | Which part stopped being worth it |
| Silence past a normal interval | Disengagement, which is the quietest signal | Whether they still have the problem |
| Feature adoption trending down across a cohort | A change removed something people relied on | What got harder |
| An explicitly negative piece of feedback | An open, unresolved frustration | Whether it has been addressed |
Two rules make triggers work. Compare an account against its own baseline, since normal usage for a five-seat team and a five-hundred-seat team share no scale. And cap the frequency per account, or the customer already having a bad month gets surveyed three times about it.
A Question Library
Ask about one thing at a time. These are grouped by what you are trying to learn; take two or three from one group rather than assembling all of them into a form.
Value: is this still doing the job
- What are you using [product] for at the moment?
- Has what you needed from it changed since you started?
- Which part of your week does it actually save?
Friction: what is costing them effort
- Was there anything you tried to do recently that took longer than it should have?
- Is there a step you find yourself repeating or working around?
- What did you do the last time [common task] did not work the way you expected?
Missing capability: what is not there
- Is there anything you currently do outside [product] that you would rather do inside it?
- What made you look elsewhere most recently, if anything?
Replacement risk: is something else in the picture
- Are you using another tool alongside this one? What for?
- If your team stopped using [product] next quarter, what would you move to?
Retention: what would keep them
- What would have to be true in six months for this to still be worth it?
- What is the one thing that would make the biggest difference to you?
Two to three questions in-app, five to seven by email where you have more of someone's attention. A short survey a customer finishes beats a thorough one they abandon.
Write the Question So the Answer Is Usable
Most survey data is weak because the question did the answering.
| Instead of | Ask | Why |
|---|---|---|
| Do you love the new dashboard? | What did you use the dashboard for last week? | The first invites agreement. The second returns a behaviour. |
| How much has [product] improved your workflow? | Has anything about your workflow changed since you started using it? | The first presumes improvement and gets it. |
| Would you like faster reporting? | What do you do while a report is running? | Everyone wants faster. The second tells you whether it matters. |
| Are you happy with our support? | What happened the last time you contacted support? | An adjective versus an account of events. |
| Why did you stop using the export? | Are you still exporting? If so, how? | The first assumes a fact they may not share. |
| Rate the value for money | What would you compare this to when deciding whether to renew? | The first gets a number; the second gets the comparison being made. |
The pattern: ask what happened, not how they feel about what happened. A recalled action is evidence. An agreement is politeness.
Turning an Answer Into an Action
A response with nobody assigned to it is a survey you should not have sent.
| What the answer shows | Read it as | Do this | Owned by |
|---|---|---|---|
| Cannot state what it is used for | Value never landed | Re-run activation on the actual job, not a feature tour | Customer success |
| Names a workaround they perform regularly | Friction with a known cost | Cost the workaround, then decide whether it is a fix or a roadmap item | Product |
| Names a capability that is genuinely absent | Fit gap | Log it as evidence with the account attached; do not promise it | Product |
| Mentions another tool doing part of the job | Displacement in progress | Find out which part, and whether it is coming back | Customer success |
| Repeats a problem already reported | Loop never closed | Answer the original report first. Nothing else will land until you do | Support |
| Positive, specific, and detailed | Reference or expansion | Ask for the detail in their words while it is fresh | Customer success |
| No response at all, repeatedly | Disengagement | Treat non-response as a signal in its own right, not as missing data | Customer success |
That last row is the one teams skip. A silent account is not a neutral account, and silent users rarely become vocal before they leave.
Segmenting the Responses
The same sentence means different things from different accounts. Split at minimum by:
- Plan and tier. Complaints concentrated at one tier are usually a packaging problem rather than a product one.
- Tenure. Month-one confusion is onboarding. Month-eighteen confusion is a change you made.
- Team size. Problems that only appear above a certain seat count are collaboration problems.
- Lifecycle stage. Onboarding, steady state, pre-renewal and post-expansion produce different answers to the same question.
- Usage level relative to their own baseline, not to the average.
- Feature adoption. Accounts using one capability and accounts using six are different products in practice.
Without segmentation the aggregate reads as mild dissatisfaction everywhere, which is not actionable and is usually not what is happening.
What a Response Is and Is Not
A retention survey answer is a signal, not a forecast. Accounts that answer bleakly renew. Accounts that answer warmly leave, for budget cuts, a reorganisation, a champion moving on, or an acquisition that came with someone else's tooling. None of that is visible in a survey response, and none of it is your product.
What the survey genuinely buys you is time and specificity: an identified account, a named problem, and a window in which doing something about it is still possible. Treat a low score as a reason to have a conversation, not as a probability.
Two related traps. Surveying only the accounts you already worry about will confirm that you were right to worry, because you never asked anyone else. And a survey with no visible follow-up teaches customers that answering achieves nothing, which quietly removes the channel.
Running It Without a Survey Platform
The instrument matters less than what happens next. At low volume this works as a personal email from a real person, which reliably outperforms a form.
What has to exist is the back end: every response landing in the same record as your other feedback, attached to the account, so a theme can be counted across surveys, support and requests rather than living in a survey tool's own reporting. Open-text answers are the valuable part and the part most likely to be stranded.
FlagUp does not run surveys. What it holds is that record: responses and requests in one place with the account attached, duplicates grouped so a theme is counted once, and a roadmap and changelog that tell the people who raised something what happened. See what each plan includes.
Frequently Asked Questions
What is a retention survey?
A short survey sent to an active customer, triggered by something in their behaviour, asking whether the product is still doing the job they need. It differs from NPS by being targeted and from a cancellation survey by arriving while the outcome is still open.
How is a retention survey different from NPS?
NPS is periodic, universal and gives you a trend. A retention survey is triggered, targeted at one account and gives you a reason. Teams often run both, and the mistake is expecting NPS to explain why the number moved.
When should you send one?
On a behavioural trigger: usage falling against that account's own baseline, stalled onboarding, a key feature never adopted, repeated support contact on one theme, a downgrade, or an unresolved negative comment. Cap how often any one account is asked.
How many questions should it have?
Two or three in-app, five to seven by email. Completion matters more than coverage, and one open-text question about a recent action usually returns more than four rating scales.
Can survey responses predict churn?
No. They surface accounts worth a conversation and tell you what to talk about. Plenty of churn has causes a survey cannot see, including budget, reorganisation and a champion leaving.
What do you do about accounts that never respond?
Treat repeated non-response as a signal rather than missing data, and reach out through a channel that does not require them to fill anything in. Disengaged accounts are the least likely to answer and the most worth asking.
FlagUp helps teams keep survey answers, requests and support signals in one place, attached to the account that raised them. Start free.
Related articles
- What is Customer Health Scoring? Definition, Examples, and Tools
- Why SaaS Users Churn, and What the Data Can Tell You
- 7 Churn Signals Hiding in Your Cancellation Flow Data
- How to Use Customer Success Tactics to Retain SaaS Users
- How to Reduce SaaS Churn Using a Structured Feedback Loop
- What Silent Users Are Actually Telling You About Churn