Voice of customer analysis: a worked example, from 87 comments to four findings

Voice of customer analysis is the step between collecting what customers say and deciding something because of it. Most guides stop at a diagram of that step. Here we do it on real data and show the numbers at each stage.

The data is one public discussion: “Are meal kits dying?” in r/mealkits. A customer of ten years opened it with It’s gotten awful, and 87 comments followed. That is about the size of a batch of survey comments or a week of support tickets, which makes it a good practice set.

If you want to see what the raw material looks like first, we sorted real quotes into nine kinds in Voice of customer examples. This post is about what to do with them.

1Write down one question

“What do customers think?” is not a question you can finish answering. Pick one decision the analysis should inform and write it as a question before you read anything. For this thread we used:

Why are long-time meal kit customers unhappy, and where are they going instead?

The question decides what you tag and what you ignore. This thread also contains recipe tips, a debate about sodium, and an argument about who owns which company. All of that is interesting; none of it answers the question, so it gets set aside rather than summarised.

2Count people, not comments

The thread has 87 comments. They were written by 60 different people. Four comments were deleted or removed and one is a moderator notice, so they don’t count as customer voices at all.

The gap matters more than it looks. The most active participant wrote 5 of the 87 comments; the next wrote 4. If you count comments, one persistent person can look like a trend. So the first thing to build is a list of people, with every comment attached to the person who wrote it.

Check who is speaking, too. One comment in this thread recommends a meat delivery brand, from an account whose name matches that brand. It may be helpful, but it is not a customer, and it should not be counted as one.

3Tag what each person says

Read every comment and tag what it says about your question. Keep the tags plain and close to the customers’ words; you can merge them later. For this question, five tags covered almost everything relevant:

• Quality went down: smaller portions, worse produce, cheaper ingredients.
• Unreliable delivery: late boxes, missing or damaged ingredients.
• Price: too expensive, confusing pricing, premium upcharges.
• Doing it themselves: reusing recipe cards, meal-planning apps, back to the grocery store.
• Still happy: people who use a service and like it, or say the decline is overstated.

One comment can carry several tags. A person who writes that quality fell, that they are switching, and that they would pay more for the old quality gets all three. The tags are a reading, so write down the rule for each one; someone else should be able to tag the same comment the same way.

4Count each theme by people

Now count, for each tag, how many different people said it. This is the table that came out of the thread:

Two things changed when we counted people instead of comments.

Price shrank. Price comes up again and again in the thread, but much of that is a handful of people replying back and forth. Five people complained about price, and between them they wrote 14 of the thread’s comments, so price feels bigger than it is when you read through. Counted by people, it is the smallest theme.

The biggest theme was not a complaint. The question we started with was about unhappiness, but the largest group is people describing what they do instead: saving the recipe cards, using a meal-planning app, going back to the store. For a meal kit company, that is the competitor that matters most, and it doesn’t show up in a list of complaints.

5Put the counterpoints next to it

Twelve people in a thread titled “Are meal kits dying?” said they are happy with a service. That is not noise to filter out; it limits what the complaints can mean.

The clearest exchange in the thread is between two long-time customers. One says Portions are going down, produce quality is going down, service is going down, and ability to actually get the product is going down. The other answers: The meat has stayed the same size since 2019. and My experience is mine and your is yours. (link). The same person also warns, earlier, that dissatisfied people are far more likely to post complaints (link).

Both can be true. Read closely, most complaints name a specific service and a specific recent period, and several of the happy customers use a different service. So the finding is not “meal kits got worse”; it is narrower, and more useful.

6Write findings you can check

A finding is a sentence someone can act on, with the evidence attached: how many people, the strongest quote, and who disagrees. Here is what this thread supports:

1. Unhappy long-time customers are not only switching brands; many are leaving meal kits for recipes, apps and the grocery store.

   Support15 of 60 people

   Counterpoint12 people are still happy with a service

2. The main complaint is that quality fell while prices held, not that prices rose.

   Support11 people on quality; 5 on price

   CounterpointA regular customer says portions and produce are unchanged for them

3. Some of the unhappy customers would pay more to get the old quality back.

   Support2 people, including the person who started the thread

   CounterpointOnly two people; a lead to test, not a result

4. Late or incomplete boxes break the reason people buy: convenience.

   Support6 people

   CounterpointOne person found changing the delivery day fixed it

Then check every quote word for word against the source. Paraphrases drift: “I’d pay more to have the quality back” (link) easily becomes “customers want premium options” in a slide, and that is not what was said. If a quote can’t be found in the source exactly, it doesn’t go in the report.

Last, say what the analysis is. This is one thread, 60 people, in a community of enthusiasts, mostly written in August 2026. It is good evidence about how unhappy regulars talk and where they go. It is not a market share estimate, and the counts are not percentages of all customers.

The same steps on a spreadsheet

None of this needs special software for one thread. A spreadsheet with these columns is enough:

Count unique people per tag with a pivot table, and read the “against” rows before you write anything. It gets slow fast, though: the meal kit review thread we used in our examples post has 325 comments, and a real project usually needs several threads.

That is the part Valtaris automates. You export a thread with our Chrome extension, and the report does steps 2 to 6: comments grouped by person, themes counted by unique people, counterpoints kept beside each finding, and every quote checked word for word against the source, with links. For more on reading a single thread critically, see How to read a Reddit thread without fooling yourself.

About the data. All quotes are verbatim from public Reddit comments in one thread and link to their source; we don’t show usernames. The thread was captured on 27 September 2026 with 87 comments; scores and replies may have changed since. Theme counts reflect our tagging, done by hand with the rules above; a different reader may draw some lines differently. Valtaris is not affiliated with Reddit or any company mentioned.