Voice of customer (VoC) research is the practice of collecting what customers say, in their own words, about their problems, their doubts and why they bought, then using that language to decide what to change and test on your site.
For CRO, its main sources are customer interviews, surveys and review mining.
VoC tells you what to test and how to phrase it; it doesn't tell you what will win. The A/B test does that.
The term comes from product development. In 1993, Abbie Griffin and John Hauser published "The Voice of the Customer" in Marketing Science.
It treated VoC as the part of Quality Function Deployment that identifies customer needs, structures them and sets their priorities.
CRO teams use the same idea with a narrower goal: find the objections and phrases that belong on the page.
Customer interviews: what they are good for in CRO
A customer interview is a one-on-one conversation, usually 30 to 60 minutes, with someone who recently bought, signed up, cancelled or nearly bought.
You're trying to rebuild their decision: what triggered the search, what they compared, what almost stopped them and what tipped them over.
Interviews are strong on motive and weak on behavior.
Nielsen Norman Group's guide to user interviews notes that interviews rely on self-reported data and that "human memory is flawed".
People also shade answers to look good. So use interviews to learn why people act and which words they use, and use analytics, session recordings and usability testing to see what they actually do.
That split matters for testing. An interview might tell you that buyers worried a product was a knock-off.
It can't tell you whether a line that answers that worry lifts revenue.
Only a test can, and the answer is sometimes no.
Who to interview
- Recent buyers (within a few weeks): they still remember the decision and the alternatives.
- Near-misses: trial users who didn't convert, or shoppers who abandoned a cart and agreed to talk. They hold the objections your page failed to answer.
- Churned customers for subscription products: their reasons often point to promises the site made that the product didn't keep.
Skip people who have never bought from you or anyone in the category.
They will happily speculate, and speculation is the data you least need.
How many interviews
There is no fixed number, but the usual rule is to stop when new interviews stop producing new needs or objections.
The University of Cambridge Institute for Manufacturing's guide to customer interviews puts it as "90-95% of 'needs' can be revealed from 20-30 interviews", with the time to stop being "when no new needs are being found".
For a single product or funnel, many teams hit that point sooner.
Treat the number as a ceiling to check against, not a target.
Voice of customer research: the sources and what each one gives you
Interviews are the deepest source but the slowest.
Most CRO programs combine them with cheaper sources that reach more people. Here is how they compare.
| Source | What it's best at | Main weakness |
|---|---|---|
| Customer interviews | The full decision story, the reasons behind it and follow-up questions | Slow; small samples; memory and politeness bias |
| Post-purchase surveys | Hundreds of short answers on what nearly stopped people buying | Only buyers answer; no follow-up questions |
| On-site or exit surveys | Objections from people who haven't bought yet | Low response rates; short, shallow answers |
| Your own reviews | Outcomes and benefits in customers' words | Skews toward happy customers if you request reviews |
| Competitor and category reviews | Complaints you can position against | Not your customers; needs filtering |
| Support tickets and chat logs | Real confusion points, with how often they come up | Mostly post-purchase problems |
For writing survey questions, see our guide to survey questions.
For how this work fits next to analytics, see qualitative vs quantitative research.
What VoC-led tests look like in our database
The most common pattern in our results is simple: take the top objection from surveys or reviews and answer it where people decide.
Oliver Kenyon ran a test for a store selling wild bird food, where 70% of collection page visitors left without opening a product.
Three lines went above the grid, all taken from post-purchase surveys: fresh packing, aflatoxin testing on peanuts and free door delivery on sacks.
The result was $44,890 a month ($0.66 per session).
Kasper Dolk's team saw three questions keep coming up in customer surveys for a phone accessory: "is it durable?", "is it the official X?" and "does it protect my phone camera?".
Adding benefit bullets in the product hero that answered them was worth an estimated $763,679.14 in extra profit per year.
The same idea works with proof instead of claims.
Oliver Kenyon added a single review block below the mobile add-to-cart button, chosen to answer objections pulled from post-purchase surveys: +$63,735 a month and +$0.43 revenue per session.
@oliverkenyon@oliverkenyon
SplitBase
@kasper_dolk
Clean CommitLarger research projects show up too, but as bundles.
SplitBase's research for NeuroMD found the paid-traffic pages used brand language rather than customer language.
The new landing page with customer-voice copy, trust signals and objection answers got +55.3% conversions, but it was a whole new page, so you can't credit the copy alone.
Applica rebuilt Five Minute Journal's app onboarding on Jobs-to-be-Done research and saw +20% ARPU (p = 0.04), again with several changes at once.
Customer interview questions that produce usable answers
Good interview questions ask about a specific past event.
Nielsen Norman Group recommends prompts like "Tell me about the last time you..." because remembering one incident brings back real details.
Questions about the future ("Would you buy this?") get guesses, and guesses are usually too kind.
Questions for recent buyers
- Take me back to when you first started looking for something like this. What was going on?
- What were you using or doing before? What was wrong with it?
- What else did you look at? Why didn't you pick those?
- Was there anything that almost stopped you from buying? What settled it?
- What did you expect to happen after you bought? Has it?
- How would you describe this to a friend who has the same problem?
Question 4 feeds your objection list. Question 6 gives you headline and bullet copy in customers' own words.
Questions for near-misses and churned customers
- What were you hoping this would do for you?
- Walk me through the moment you decided not to go ahead (or to cancel).
- What would have needed to be different?
- What did you do instead?
Questions to avoid
- Leading questions: "Did you find the checkout easy?" invites a yes. Ask "How did checkout go?"
- Design opinions: "Do you like the new homepage?" Customers aren't designers, and liking isn't buying. Test it instead.
- Hypothetical futures: "Would you pay $20 more for X?" Stated intent is a poor guide to what people pay.
- Two questions in one: "Was the price and the shipping clear?" You won't know which part they answered.
Record the call (with permission) and get a transcript.
The exact wording is the point. A note that says "worried about quality" is far less useful than the customer's phrase "I've bought cheap ones that cracked in a week."
Review mining: turning reviews into test ideas and copy
Review mining means reading large numbers of customer reviews, yours, competitors' and adjacent products', and pulling out repeated problems, objections, outcomes and phrases.
Joanna Wiebe of Copyhackers popularized the method for copywriting with Amazon review mining.
She suggests focusing on 2- to 4-star reviews, where people tend to be more critical and more specific.
A simple review-mining process
- Collect 100 to 300 reviews: your own, two or three competitors', and reviews of products customers switch from.
- Tag each useful line as a problem, an objection, a desired outcome or a reason they chose the product.
- Count the tags. A theme that appears in a fifth of reviews deserves a place on the page; a single vivid quote does not.
- Keep the wording. Copy phrases verbatim into a spreadsheet next to their tag.
- Turn the top themes into hypotheses: "Buyers worry about X; answering X near the add-to-cart button will lift conversion."
Customer words can go straight into headlines. Laura Roeder Social Media replaced a homepage newsletter headline with a line from customer feedback, "Yours is the only newsletter that I actually read."
Sign-ups rose 24.31% over 12 days, with a 97% chance to beat the original.
More examples are in our headline and copy tests and review tests.
When VoC-based changes don't win
Grounding a test in customer research raises your odds, but it doesn't guarantee a lift.
Clean Commit built FAQs from verified reviews on collection pages, grouped by fit, durability and washing.
Conversion was +8.4% and revenue per visitor +33.2%, but with 10,412 visitors over 33 days the probability to beat control was only 83.4%.
The agency itself called it a directional signal, not a win.
A few reasons VoC ideas fall flat:
- The objection was real but rare. One loud reviewer isn't a pattern. Count before you build.
- It was answered in the wrong place. A shipping worry answered in the footer is as good as unanswered. Put the answer next to the decision.
- The page got longer. Every added block pushes something else down. Sometimes the cost is bigger than the gain.
- Buyers' reasons differ from non-buyers' reasons. Post-purchase surveys only hear from people who got past their doubts. Balance them with exit surveys or near-miss interviews.
Treat every VoC finding as a hypothesis and run it as a test.
Our customer research hub covers the other methods, and you can browse all A/B test results to see how similar ideas performed elsewhere.