The usual mistake with a customer feedback questionnaire is not bad wording. It is asking about three different things in one form. A questionnaire that covers the product, the people who handled the enquiry and the delivery produces a single average that hides all three, and when the average moves nobody can say which of the three caused it.
Below are three questionnaires, written out in full, one for each subject. Each is short enough to answer on a phone. After them there are the wording rules that apply to all three, a note on what to compare the answers against, and the part that decides whether any of it produces a change.
Three subjects that fail independently
Product, service and delivery break for different reasons, get fixed by different people, and are judged on different timescales. A customer can be delighted by the handling of a problem that should never have occurred, and a customer can be perfectly served by a courteous team delivering something late.
Merging them is what produces the familiar pattern where satisfaction scores sit flat at four out of five for a year while complaints keep arriving. The flat score is three signals cancelling each other out.
Splitting them does not mean sending three questionnaires to the same person. It means picking the subject that matches the trigger. A delivery questionnaire goes out when something arrives. A service questionnaire goes out when a conversation ends. A product questionnaire goes out after the thing has been used for a while, which is a different moment entirely and the one most often got wrong.
Separate questionnaires also stay short, and short is what gets answered. Three focused forms of six questions each collect more usable material in total than one form of eighteen, because the eighteen question version is abandoned halfway through by the people with the most to say.
A product questionnaire
Sent after enough use for an opinion to exist. For software that usually means the end of the first month. For a physical product it means after several uses rather than after unboxing.
- What were you trying to do with it? (short open text)
- Does it do that? (Yes / Mostly / No)
- What did you have to work around? (open text, optional)
- How does it compare to whatever you used before? (Better / About the same / Worse / Nothing before)
- How is it to use day to day? (1 to 5, ends labelled "constant friction" and "no trouble")
- What is missing? (open text)
- Would you point a colleague at it? (0 to 10)
Question 3 is the one that earns the form. Workarounds are the most actionable answers a product questionnaire produces, because a workaround is a defect the customer has already diagnosed and solved on your behalf. Fifty answers to question 3 read in one sitting will produce a shorter and better prioritised list than any internal discussion.
Question 4 is the only comparative question in the set, and it belongs here rather than in the service or delivery forms. The answer "about the same" from someone who switched is more informative than a rating of four, because it says the switch has not yet paid off.
Question 7 needs volume to mean anything. Below roughly a hundred responses the figure moves around for reasons that have nothing to do with the product, so it is worth collecting and not worth reporting monthly at low volumes.
A service questionnaire
Sent within a day of an interaction ending. This is about the handling, not about the thing.
- What did you get in touch about? (short open text)
- Did it get sorted? (Yes / Partly / No)
- How many times did you have to ask? (Once / Twice / Three or more)
- How long did it take, against what you expected? (Faster / About right / Slower)
- How was the handling? (1 to 5, ends labelled "hard going" and "easy")
- Was there anything the reply missed? (open text, optional)
Question 3 is the sleeper in this set. A repeat count is a fact about what happened, not an opinion about how it felt, which makes it much harder to argue with in a review meeting. Two identical ratings of five mean different things if one of them took four emails to reach.
Question 4 asks about the expectation rather than the clock. Two days is fast for a complex request and slow for a password reset, and only the customer knows which one they thought they had. A stated service target on the website makes this question more useful still, because the answers then measure the promise rather than a vague feeling.
Question 2 deserves a specific habit. Every "Partly" and every "No" is an open case regardless of what the support system says, and treating them that way catches the issues that were closed administratively rather than actually resolved.
A delivery questionnaire
Sent within two days of arrival, while the packaging is still a memory.
- Did it arrive when you expected? (Early / On time / Late)
- Did it arrive in good condition? (Yes / Minor damage / Unusable)
- Was the packaging about right? (Too little / About right / Excessive)
- Was everything in the box? (Yes / No, and what was missing)
- How was the tracking and the notifications? (1 to 5)
- Anything else about how it arrived? (open text, optional)
This questionnaire has the highest ratio of actionable answers to questions asked, because every answer maps to a specific operational decision, and it is the one teams most often omit. Question 3 is the one that surprises people: excessive packaging generates as many remarks as insufficient packaging, and it costs money to produce.
Question 1 asks about the expectation rather than the promised date, deliberately. A parcel that arrived on the promised day still counts as late to a customer who was told something different at the point of order, and that gap is a fixable communication problem rather than a carrier problem.
Keep questions 2 and 4 as fixed options with a follow up text field rather than as one open question. Damage and missing items need to be counted by month and by carrier, and free text cannot be counted.
Wording rules for all three
| Instead of | Ask | Why |
|---|---|---|
| How satisfied are you overall? | Did it do what you needed? | An outcome can be acted on, a feeling cannot |
| Was the service fast and helpful? | Two separate questions | A fast unhelpful reply cannot answer the combined version |
| How much did you enjoy it? | How was it? | The first sets a floor under the answers |
| Rate the professionalism, friendliness and expertise of the team | How was the handling? | Long matrices get the first row repeated down the block |
| Any other comments? | What would have made this better? | The specific version gets answered several times more often |
| Please rate your experience | Name the specific thing being rated | Nobody knows what experience refers to |
Two further rules do not fit the table. Keep one scale shape per questionnaire, because mixing three point and five point scales in one form makes the results harder to read than the extra nuance is worth. And label the ends of any rating scale in plain words rather than with numbers alone, since respondents otherwise guess at what the middle means and the guesses differ.
What to compare the answers against
A number with nothing beside it is not a finding. Three comparisons make the same answers useful, and none of them requires extra questions.
| Compare | Against | What it shows |
|---|---|---|
| This month | The same month last year | Real change, with seasonality removed |
| One team or carrier or region | The others, same questions | Where the variation actually sits |
| People who answered | People who did not | Whether the sample is the loudest quarter of customers |
The third comparison is the one that gets skipped and it matters most at low response rates. If forty of two hundred people answered, the finding is about those forty. Checking whether they differ from the rest in an obvious way, such as being concentrated among recent customers or among one product line, takes minutes and prevents a decision being made on a skewed sample.
One more habit is worth building in. Record the wording of every questionnaire alongside its results, with the date it changed. Rewording a question resets its trend line, and a comparison across a rewording is a comparison of two different questions. Teams that do not keep this record end up explaining a step change in the numbers that was caused by their own edit.
The half that decides whether this works
A questionnaire is a few hours of drafting. What happens to the answers is the ongoing cost, and it is where most customer feedback programmes quietly stop.
The pattern is consistent. Responses arrive as notification emails, are read once, and settle into an inbox with no shared record of which ones were answered and by whom. Two people either both reply to the same customer or neither does. Three months later the questionnaire is still collecting and nobody can say what changed because of it.
Three properties prevent that, and they are properties of where the responses land rather than of the questionnaire. Each response needs an owner, so a reply is one person's job. It needs a status, so that read, replied and closed are distinguishable at a glance. And the reply needs to be sendable from the same screen as the answers, because switching applications is the friction that turns a two minute reply into a task for later.
A spreadsheet export can carry the first two with discipline and cannot do the third. Tools built around managing responses after they arrive hold owner, status and history on the response itself, and the situations where that pays off are the recurring ones: three questionnaires running continuously, read by a small team, where the cost of manual bookkeeping compounds every week.
Volume is worth checking before committing to a tool. Customer feedback arrives in bursts, quiet for a quarter and busy after a launch or an incident, which fits badly with pricing metered by monthly responses. Typeform's Basic plan includes 100 responses a month and one user at 39 USD a month, with 1,000 responses and three users on Plus at 79 USD. Tally offers unlimited forms and submissions on its free tier under fair usage guidelines, with a paid Pro plan at 24 USD a month. SurveyMonkey's team plans are billed per user with a minimum of three, and responses past the allowance are charged individually. Those are each company's published prices as of September 2026, and the choice follows from which number moves in your busiest month.
What to change first
Split whatever single questionnaire exists now into the product, service and delivery versions, and attach each one to its own trigger rather than to a schedule. Then add the repeat count question to the service version, because it produces the one number in this whole set that nobody can wave away. If the answers currently land in a shared inbox, Halict shows what handling one of them looks like instead.
Q1. How many questions should a customer feedback questionnaire have?
Six or seven, with at most two open text fields and those at the end. Longer questionnaires are abandoned partway through by the people with the most to say, which biases the results towards the indifferent. Three short questionnaires on separate subjects collect more usable material than one long one covering all three.
Q2. Should product, service and delivery be in the same questionnaire?
No. They fail for different reasons, get fixed by different people, and are judged at different moments, so merging them produces one average that hides all three. Send each one on its own trigger: after a conversation ends, after a delivery arrives, and after the product has been used for a few weeks.
Q3. What is the single most useful question to add?
For service, how many times the customer had to ask. It is a fact about what happened rather than an opinion about how it felt, so it survives a review meeting intact, and it distinguishes two identical high ratings where one took four emails to earn. For a product, asking what the customer had to work around.
Q4. How long after a purchase should a customer feedback questionnaire be sent?
It depends on the subject. Delivery questions go out within two days of arrival, service questions within a day of the conversation ending, and product questions after enough use for an opinion to exist, which is usually a few weeks. Asking at the moment of purchase collects expectation rather than experience.
Q5. How many responses are needed before the numbers mean anything?
For a recommendation question scored from 0 to 10, roughly a hundred, since below that the figure moves for reasons unrelated to the product. Outcome questions such as whether something got sorted are useful from the first few dozen, because each answer maps to a specific case rather than to an average.