Your voice of customer research is already sitting in your inbox
Yevhen Kurapov, co-founder · Updated Oct 23, 2026
Voice of customer means finding out what customers want, expect and dislike, in their words. A small business doesn't need a program for it: the questions people send by chat, DM and email already say it. Keep them, count them by topic, and spend 20 minutes a week turning the top one into a fix.
Hannah runs a loose-leaf tea shop online with her sister, and for two years one email turned up most weeks: how many cups does a 2-ounce pouch make? She'd answer it between packing orders and forget about it. One Sunday she searched her inbox and counted more than forty. She put "about 25 cups" on every product page, under the price, and the question nearly stopped.
Big companies call this voice of customer and run whole programs for it, with surveys, dashboards and someone whose job title has "insights" in it. A shop of two can get most of the value from the questions customers already send by chat, DM and email, as long as somebody keeps them. Usually nobody does. They sit in DMs nobody scrolls back through, on someone's phone, or in the memory of whoever was on shift that day.
Most people who leave don't tell you
A Qualtrics XM Institute survey of 20,000 consumers in 14 countries, published in October 2025, found that only 3 in 10 explain why they leave, and 30% tell nobody at all and just switch. In a separate Qualtrics survey from the third quarter of 2025, 47% of bad experiences led the customer to cut spending, and online retail was among the worst, at 58%. Those are shares of bad experiences rather than of customers.
Zendesk, which sells support software, asked about 5,100 consumers in 22 countries for its 2025 report, and 63% said they'd switch to a competitor after a single bad experience. That's what people said they would do; nobody watched what they actually did.
Questions behave differently. A person who asks "does this have caffeine?" or "will it arrive by Friday?" is pointing at the exact spot where your page or your process left them unsure, and they're still deciding. The ones who left without a word may well have had the same doubt.
Five ways to listen, compared
| Method | What it costs you | What it misses |
|---|---|---|
| Ask: a one-question survey after delivery, a review request, or "What almost stopped you buying today?" at checkout | Free; ten minutes to set up | People who didn't buy, and anyone not in the mood to type |
| Read reviews, yours and competitors' three-star ones (the most honest kind) | Free, half an hour | Everything before the purchase; people who left never write one |
| A notebook by the till or a shared sheet where anyone writes down a question they hear twice | Free | Whatever happens after the first busy week, when it usually stops |
| Tags in your inbox: Gmail labels or helpdesk tags such as "shipping", "sizing", "wholesale", "complaint" | Free in Gmail | Messages spread across Instagram, WhatsApp and the website |
| Reading chat transcripts | Your time; the routine below caps it | Everyone who didn't write: only people who chose to ask something show up |
If I could do only one of these, it would be the checkout question. It's specific, it costs nothing, and it catches people at the moment they almost didn't buy.
Twenty minutes on Monday
Before the inbox, with a timer.
- Five minutes on last week's reviews, yours and one competitor's. Copy any sentence that names a specific problem into a running list.
- Ten minutes on last week's conversations. Not all of them. Start with the ones your answers didn't cover: chats handed to a person, questions answered with "I don't have that information", anything about delivery or returns. Add each repeated question to the list with a tally mark.
- Five minutes to fix one thing. Take the question with the most tally marks and answer it where people will see it before they have to ask: a line on the product page, an FAQ entry, a sentence in your shipping policy, a new file for your assistant. The FAQ page template has forty answers you can adapt.
Keep the list. After a month you'll see which questions shrank once you fixed them and which keep coming back. The stubborn ones are usually a product or policy problem rather than a wording one.
Where ChatPal shortens step two
Say ChatPal answers Hannah's website chat and DMs. On Monday she opens Chats → All and last week's conversations are in one list instead of three apps. Search finds a customer by name, email or phone, not by what they wrote, so this is for reading, not for hunting a phrase. Under Analytics → Conversation Insights she sees how many conversations fell into each category she named ("caffeine question", "wholesale") and the share of replies that found nothing in her files. When that share climbs, people are asking about things she hasn't written down. The "Uncovered" filter shows the conversations that matched none of her categories, which is where a new topic shows up first. There's no sentiment score and no ready-made list of questions; she reads the conversations herself.
A chat from Hannah's shop
Hannah's shop added its product pages and contact page; the product pages list the roasted hojicha as low caffeine and suggest it for evenings. She set up two categories: "caffeine question", counted only, and "wholesale", with the instruction to say the shop sells to cafés from 2 lb per blend and that Hannah sends the price list. She set lead capture to collect a name and email instead of a phone number.
A made-up example, not a recorded chat. The replies follow the rules the assistant works by.
Client: Hi! Does the roasted hojicha have caffeine? I mostly drink tea after dinner.
Assistant: It's listed as low in caffeine, and we suggest it as an evening tea.
Client: Nice. Also, I run a small café in Bend. Do you do wholesale? We'd go through maybe 5 lb a month.
Assistant: Yes, we sell to cafés from 2 lb per blend, and Hannah sends the wholesale price list herself. Could you leave your name and email so she can send it?
Client: Sam Ortiz, [email protected]
Assistant: Thanks, Sam. Hannah will send you the wholesale price list.
On Monday that chat adds one to each of Hannah's categories and puts Sam under Leads as a café in Bend wanting about 5 lb a month. If "caffeine question" keeps growing, the fix is the same one as the cups: a line on every product page. To check answers like these before customers see them, there's how to test an AI chatbot.