In short
- Voice of the customer means everything a customer tells you, kept in one place.
- It usually stands in for four concrete things, and naming them shows which one is missing.
- Three things are commonly claimed for the phrase that it can't do.
- What people say isn't what people do. Somebody saying they would pay isn't a payment.
Voice of the customer is a collective noun for the things customers say about your product, not a method, a tool or a department, gathered somewhere they can be read together instead of one at a time.
The phrase is useful for exactly one thing, which is giving a budget line a name. It becomes a problem when it's used as though it were itself a source of evidence, because the evidence is always the underlying conversations, and those have to actually exist.
What does voice of the customer actually mean?
Everything a customer tells you, plus the discipline of putting it in one place.
The second half is the whole content of the term. No company lacks customer opinions. What companies lack is a place where the support conversation from March, the cancellation reason from May and the request from June can be read side by side by the person deciding what to build in July.
Where that place doesn't exist, the phrase describes an aspiration. Where it does exist, the phrase is unnecessary, because people just say the feedback.
There's one of those places open without an account.
Want the counting done for you? Twenty eight feedback tools, the price and the unit each one bills on, read on one day.
What is it usually standing in for?
Four concrete things, and it's worth naming them because the abstraction hides which one is missing.
| What somebody means by it | The concrete thing | Whether most companies have it |
|---|---|---|
| We should know what customers want | A counted list of requests | Rarely counted, often collected |
| We should know why people leave | A recorded reason per cancellation | Almost never |
| We should know what the product is bad at | Support conversations read in a batch rather than one at a time | The data exists, the reading does not |
| We should be able to prove a decision to the board | A number attached to a want | This is what the phrase is most often hired for |
The fourth row is the real one. Voice of the customer is most frequently invoked when somebody needs to justify a decision upwards, and a counted request list is what they actually need. That's a legitimate need, not something to dress up.
What the phrase cannot do
Three things, and they're the three most commonly claimed for it.
It can't tell you what to build. A counted list tells you what people asked for. What to build is a decision about cost, strategy and timing, and no amount of listening produces it.
It can't represent everybody. Every channel hears from the part of your audience that speaks. On 634 public Canny and Featurebase boards read on 3 October 2026, the median request held two votes, one of them its poster's. Two people are a real signal and not a referendum. The votes on all 634 boards are in one open file.
It can't survive being a project. A quarter of interviews produces a document. A standing channel produces a number that keeps moving, and only the second one is still useful in a year.
What do people get wrong about it?
Treating what people say as what people do. Somebody saying they would pay for a thing isn't a payment. Ask what they did last time rather than what they would do next time, and the answers get worse and more accurate at once.
Counting messages instead of people. Five emails from one engaged customer are one person's opinion, and they routinely outweigh fifty accounts that never write.
Collecting without closing. The single fastest way to kill a feedback channel is to never answer it. People file once, hear nothing, and stop, and the channel looks healthy right up until it's empty.
Buying a tool for the listening rather than for the counting. Listening is free and already happening. Counting the same want once, and being able to tell everybody who asked, is the part that's hard by hand.
How do you do it without buying anything?
Three habits, and the first two cost nothing.
Read a hundred support conversations in one sitting, and write one line each, the situation, not the request. Reading in a batch is the entire method, because a problem that's invisible one ticket at a time is obvious in a pile of forty.
Ask one question when somebody cancels, and record the answer verbatim. These are the clearest statements a product ever receives and almost nobody keeps them.
Then put the requests somewhere public where the next person can agree with one instead of writing it again. That's the only one of the three that needs software, and it's the one that turns opinions into counts.
What are good voice of customer survey questions?
Good voice of customer survey questions ask about a problem the person had, not about your product.
- What were you trying to do when you last got stuck?
- What do you do today instead, when this does not work?
- If we could fix only one thing for you this quarter, what would it be?
- What almost stopped you buying?
- Who else is affected when this goes wrong?
The second one is the most valuable and the least asked. A workaround somebody performs every week is the clearest description of an unmet need anybody ever gives you, and nobody keeps doing a manual chore for something they do not need.
Avoid anything shaped like would you use a feature that. People are agreeable and bad at predicting their own behaviour, and a yes there predicts nothing. The whole twelve question survey, with the reason for each one, is free to take.
Voice of the customer vs NPS
NPS is one instrument inside a voice of the customer programme. The programme is the whole practice.
Treating them as the same thing is the most common way this work ends up as a quarterly slide. The score gets presented, everybody agrees it should be higher, and the free text underneath is never opened.
The score tells you the temperature. The rest of the programme is the diagnosis, and reporting the first without doing the second is measuring rather than listening. What NPS can and cannot answer is a page of its own.
Voice of the customer vs market research
Market research asks people who are not your customers. This asks the people who are.
Market research sizes a market, tests a message, or finds out what a segment believes, and it usually involves people who have never heard of you. That is exactly its value when the question is whether to enter a market at all.
This is narrower and more actionable. Everybody in it has already chosen you, pays you or nearly did, and knows what the product actually does. For deciding what to build next, that population is worth more than a representative sample.
What goes in a voice of the customer report?
Three things, and the third is what makes anybody open the next one.
What changed, in a sentence, on whatever you measure.
What people asked for, ranked. Not a word cloud. A list with counts, in the customers' own phrasing where you can keep it.
What was done about the last report. A report that never refers to the previous one teaches its readers that nothing follows from it. One that opens with three requests from last quarter and the words these shipped is a report people will read. That is the entire trick and it costs nothing except doing the work.
What is different about this in business software?
The voice belongs to an account rather than a person, and the accounts are not the same size.
In a consumer product one customer is one vote and that is usually right. In business software one customer might be forty seats and a third of your revenue while another is one seat on the cheapest plan. Counting those the same way produces a ranked list that quietly serves your smallest customers first.
Most writing about this is written for the consumer case and never says so. The arithmetic that fixes it is to attach the revenue to the account, attach the account to the person, and let the weight travel with the vote. The mechanism is here.
What VoteFirst is, in these terms
The counted list, and the part that answers people.
Requests arrive from anybody without an account, nothing appears publicly until somebody approves it, and the same want is merged, not counted twice. Each request carries a plain count and a count weighted by what the people voting pay, which is the number the fourth row of the table above is really asking for.
Knowing who voted is what makes the closing automatic. When a request ships, the people who asked are told without anybody keeping a list. Everything that happened sits on one timeline.
It's worth stating what this isn't, because the phrase invites overclaiming. A request board is one of the four rows above.
It doesn't collect cancellation reasons, it doesn't read your support conversations, and it hears only from people who turn up. How the board is wired is written up here, and where voice of the customer data actually comes from has its own page.
14 days, a card at signup, then $9 or $39 a month.
