Voice customer feedback: the why behind the score.
NPS tells you a customer is unhappy. It does not tell you why. A 60-second voice note does: in their words, with their actual tone, without a call on your calendar.
The short answer
What a score cannot tell you
A 7 from a churning customer and a 7 from a loyal one who hit a bug look identical on a dashboard. Same row, same number, same color. You only know which is which when you ask. The score tells you there is a signal somewhere. It does not tell you what the signal is.
NPS and CSAT are still useful, as trend lines, as board metrics, as a quick pulse across a large customer base. That is what they were designed for. What they were not designed for is explaining why the number moved. A score going from 42 to 37 in a quarter is a fact. The reason is a conversation. Most teams skip that conversation because it takes time and calendar space they do not have.
A text follow-up form asks “why did you give us a 7?” and gets a typed, edited, sanitized answer, or nothing. Most customers close the tab. The ones who do fill it out tend to write what they think sounds polite. A voice link at that same moment gets the actual reason. Speaking is lower friction than writing for most people, and the answer comes out before the internal editor kicks in. Sixty seconds of voice tells you what ten survey questions cannot.
When to send a voice link
These five moments produce the most honest answers. In each case, the question is specific to what just happened.
After a product trial ends
Most post-trial surveys ask whether someone plans to subscribe. That is the wrong question. Ask: “What almost made you sign up?” The almost is where your product loses the sale. A churned trial user has nothing to lose by being honest, and voice is fast enough that they usually are. Three answers from people who did not convert will teach you more than fifty satisfaction scores from people who did.
After a support ticket closes
The standard follow-up is a CSAT star rating. It tells you whether the customer is happy with support, not whether the underlying problem is actually resolved. Those are different things. A voice link with one question, “Did we actually fix the problem for you?”, tells you which it was. Sent within an hour of the ticket closing, before the customer moves on, reply rates consistently hit 40-60%.
Post-churn
Exit interviews are almost never done because they feel awkward to schedule and the customer has already moved on. A voice link removes the friction. “What would have made you stay?” is a 60-second ask, not a 30-minute call. Some customers will not respond. The ones who do are often surprisingly direct, and a few will name something specific and fixable. One of those answers can pay for months of product roadmap.
After a big feature ships
Usage data tells you whether customers tried the feature. It does not tell you what they thought of it. Send a link to the first 20 people who used it and ask: “What is your first reaction to [feature name]?” You will get a mix of genuine enthusiasm, confusion about where to find things, and edge cases your QA missed. All of it is faster to collect via voice than by waiting for the pattern to show up in support tickets three weeks later.
Regular pulse check
Once a month, send a link to a rotating group of 15 to 20 active customers with one question: “What is the one thing you would change about [product] right now?” No context, no preamble. The answers self-organize around whatever is actually frustrating people at that moment. It takes 10 minutes to read the summaries. Over time you get a running picture of what your product feels like to use, something no dashboard metric gives you.
How it works for your customers
You create a Magic Link in HeySpeak with one question attached. The link goes into whatever message you already send: email, Slack, your support tool. The customer clicks it, sees your question, taps record on their phone or laptop, speaks for up to 60 seconds, and hits send. No app download. No account creation. No form to fill out.
HeySpeak transcribes every response automatically and generates a short AI summary. You can read all summaries in a few minutes on your dashboard and click through to the audio or full transcript for anything that stands out. If a customer prefers not to speak, they can book a time on your calendar instead. The link gives them both options.
All processing runs on Mistral AI, which is 100% EU-hosted. Audio is stored in a private bucket and never publicly accessible. Responses are yours. There is no data shared with third-party ad networks or analytics platforms.
Keep going
Two more pages on getting real signal from customers.
Common questions
Can I use this alongside NPS, or does it replace it?
What question should I ask customers for feedback?
How do I make sure customers actually respond?
What do I do with the voice responses?
Is customer voice data stored securely?
Hear from your customers this week.
Five free responses to start. No credit card, setup in under a minute.
Create your first link