
Feedback analysis
Voice of customer analysis with AI: a weekly routine
Voice of customer analysis with AI, as a weekly routine: group what customers wrote by reason, read their own words, and pick one thing to fix.
Another week of reviews, survey comments and support messages has piled up, and nobody has read them. Voice of customer analysis with AI turns that pile into one weekly habit. Gather what customers wrote, group it by reason, read their own words under each group, and pick one thing to fix.
A complaint you read this week can still be fixed before your next app update goes out.
Key takeaways
- Voice of customer analysis is three jobs: find what customers need, sort those needs into groups, and decide which come first.
- Done by hand every week, it means reading, grouping, counting and finding quotes. That is why most teams stop.
- Group what customers wrote by reason, most common first, and read the customers' own words under each group before you act.
- Most common is not the same as most important. The count is a reason to look. You still decide.
What is voice of customer analysis?
It means reading what customers wrote and working out what they need from you.
A 1993 research paper split it into three jobs (Griffin and Hauser, "The Voice of the Customer"). Find what customers need. Sort those needs into groups. Decide which come first. Back then the words came from interviews, and people read the notes.
The three jobs are the same today. Now customers write to you without being asked, in reviews, survey comments and support messages.
Why does a weekly voice of customer program stall?
Because the feedback lives in many places, and reading it by hand is slow.
- App Store reviews. In App Store Connect (Apple's site for app makers), pick your app and click Ratings and Reviews in the sidebar. You can filter reviews by app version and by star rating, and pick a country from the menu at the top (Apple's help page).
- Google Play reviews. Play Console (Google's site for app makers) saves your reviews as one CSV file per month. A CSV is a plain spreadsheet file, one review per row. Google posts new data to these files within 3 to 7 days, so the newest few days can be missing (Play Console Help). The Google Play guide has the download steps.
- Survey comments and support messages. Export the answers people typed, from whatever tool holds them, as a CSV file.
Then comes the slow part. You read everything. You sort it into groups of similar complaints. You count each group. You find a customer's own words for each one. Nobody has time to do that every week, so the plan to read feedback stays on paper.
What should you ask every week?
Ask the same three questions, in the same words, every week. That is what lets you compare one week with the next.
- What did customers complain about most this week?
- What showed up this week that was not there last week?
- What are customers asking for that we do not have?
Then open the top two groups, read the customers' words, pick one thing to fix, and write one line: "This week: charged twice for one order. Sent to the team." Next week you can see if it went away. Those lines are your weekly voice of customer report.
How does Metricsense do this every week?
Each step below starts with the problem, then shows what Metricsense does about it.
- Your feedback sits in different places. Metricsense connects to the App Store and Google Play, so reviews arrive without a download. Anything else, like survey comments, comes in as a CSV file.
- Nobody has time to sort a pile into groups. You describe what to find in plain words, such as "What did customers complain about most this week?" Metricsense calls a saved question like this an AI Insight. There is no tagging and no model training.
- You cannot tell if three complaints are one problem. Metricsense reads reviews, tickets and surveys together and groups them by reason, most common first. A problem is counted once, wherever it was raised.
- A group you cannot trace is a guess. Each group links to the exact customer words behind it, so you can read them before you act.
- You cannot tell if last week's complaint is still there. Trend tracking shows whether a group is growing or fading over time, so question 2 answers itself.
- You cannot watch it every day. Alerts by email or Slack tell you when a problem comes up again. You can also send a finding to Jira or GitHub Issues, so it reaches the people who fix it.
Here is what comes back for question 1. Illustrative data, not actual customer data.
| Group, most common first | New this week? | One customer sentence under it |
|---|---|---|
| Charged twice for one order | Yes | "I bought it once and my card shows two charges." |
| Promo code will not apply | No, it was here last week too | "My promo code says invalid when I try to pay." |
| Too many notifications | No | "Five alerts before lunch is too many." |
| Wants to save a card | No | "Let me save my card so I do not type it every time." |
What you learn from this: the double charge is new and at the top, so it is this week's pick. The promo code complaint is still waiting from last week.
How do you check the answer?
Whatever made the groups, never send your team one you have not opened. Three checks, a few quotes each:
- Do the quotes say the same thing? If one is about something else, the group is too wide. Ask a narrower question.
- Is there a real sentence under every group? A group with no quote is an opinion.
- Does the date fit? A complaint marked new should have quotes from this week, not from last month.
The guide on how to analyze customer feedback with AI covers checking in more depth.
What will voice of customer analysis not tell you?
It reads what people chose to write. The customers who wrote are not all your customers. So the answer tells you what the writers said, not what everyone thinks.
Most common is also not the same as most important. That same 1993 paper asked whether a need that comes up often also matters more. So treat the count as a reason to look, not as the decision. And it cannot tell you why a customer left without writing a word.
Try it on this week's feedback
Connect your app store or upload a CSV, and ask the first question. Start free with Metricsense.


