- The usual problem
- The score moves, and the comment box says "good product" or nothing at all.
- How to run it
- Keep the 0 to 10 question exactly as you ask it today. After the score, the interviewer asks about the reason, and the follow-up angle depends on the score: what would raise it for a passive, what went wrong for a detractor, what they would tell a colleague for a promoter.
- What to do with it
- Track the score as your trend line. Read themes and quotes to decide what to fix, and open the transcript behind any theme you plan to act on.
Guide · AI customer feedback
Find out why customers gave the score, not just the score.
An AI-adaptive feedback survey keeps your core questions fixed and adds a small number of follow-ups that respond to what each person actually said. This guide explains how that works in Candor, the three workflows it fits best, and the situations where you should use something else.
Last reviewed October 12, 2026
01 · The mechanism
Adaptive is not the same as branching
Branching logic picks the next question from a list you wrote in advance, based on which option someone selected. It is predictable and easy to audit, and it cannot react to a sentence it has never seen.
An adaptive follow-up is written in the moment from the respondent's own words. If someone says onboarding was "a bit rough", the next question can ask which step, instead of a generic "anything else?".
In Candor you write the anchor questions. The interviewer may add follow-ups between them, within a depth setting you choose, and it stops probing when an answer is already specific enough. The full rules are in How adaptive follow-ups work and on the methodology page.
| Static open text | Branching logic | Adaptive follow-ups | |
|---|---|---|---|
| Next question decided by | Nothing; one box | Selected option | The answer text |
| Every respondent sees the same wording | Yes | Within a path | Anchors yes, follow-ups no |
| Handles a vague answer | No | No | Can ask for specifics |
| Main risk | Thin answers | Unanticipated reasons | Wording varies; must be bounded |
02 · Workflows
Three recurring jobs it fits
- The usual problem
- Cancellation reasons collapse into "too expensive" and "not using it enough", which rarely say what to change.
- How to run it
- Ask after cancellation is complete, never as a gate in front of it. One follow-up can turn "too expensive" into which part of the price, compared with what, and when it stopped feeling worth it.
- What to do with it
- Group reasons by what you could change (onboarding, a missing integration, plan structure) rather than by the category label the customer picked.
Win/loss
Win/loss survey template →- The usual problem
- Sales notes record what the rep heard. Buyers rarely tell a rep the real reason they chose someone else.
- How to run it
- Send a short interview to recent wins and losses. Follow-ups ask what nearly changed the decision, who else was involved, and which alternative was considered.
- What to do with it
- Compare the reasons winners almost left with the reasons losers did leave. The overlap is usually the most useful finding.
03 · Examples
What a bounded follow-up looks like
These exchanges were written for this page to show the mechanism. They are not customer quotes, and they do not describe measured results.
- AskedHow likely are you to recommend us to a colleague?
- Answer7. It's fine, reporting is a bit clunky.
- AskedWhat's the clunkiest part of reporting for you right now?
- AnswerExporting. I rebuild the same chart in a spreadsheet every Monday because the export drops the date filter.
One follow-up, then the interviewer moves on: the second answer names a concrete, fixable cause, so further probing would add burden without adding much.
- AskedWhat was the main reason you cancelled?
- AnswerToo expensive.
- AskedWas it the price itself, or what you were getting for it at that point?
- AnswerWhat we got. Two of us stopped logging in after the integration with our CRM broke in March.
The category says price; the follow-up shows the actual lever is a broken integration. The respondent could have skipped the follow-up with no penalty.
04 · Honest limits
When to use something else
You need statistically comparable open-text answers. Follow-up wording differs by respondent. Keep your score or closed questions as the comparable measure and treat follow-ups as qualitative evidence.
The topic is sensitive or high stakes. Health, legal, safety, or anything distressing deserves a trained human interviewer who can read the room and change course.
You are still discovering what to ask. Early, open-ended discovery benefits from live conversation where a researcher can chase an unexpected thread for twenty minutes.
You only need a count. If the decision depends on how many, not why, a short closed survey is cheaper for respondents.
We have not yet published a controlled study of how adaptive follow-ups change answer quality. When we do, the protocol and limits will be published alongside the results.
05 · Next
Try it on a real question
Run your next NPS or churn survey with follow-ups
The Free plan includes every feature. Pricing is per completed conversation, and follow-ups are never metered.