.webp)
.webp)
.svg)
TL;DR
- Your survey response rate is 7%. The other 93% are not silent. They call, ticket, review and complain every day. That is where the answer lives.
- NPS is broken for three reasons.
- The sample is skewed. Only your loudest fans and your angriest churners still respond. The mildly annoyed middle, the ones you actually lose, do not answer surveys. They just leave.
- The signal is late. By the time the score moves, the customer has already told their friends and the root cause is three weeks cold.
- The number is a proxy. It tells you something is wrong. It never tells you why, what it is costing, or what to fix.
- Meanwhile every call, ticket and review names the cause. In three real analyses:
- A retailer found a 13x spike in fan calls was not a heatwave story. It was a website showing stock as available when the shelf was empty. Named cause. Named fix.
- A payments business assumed its complaints were an engineering problem. 75.7% of CSAT failures traced to agent behaviour, not product. Different department. Different fix.
- A financial services firm found 47.6% of complaints in one theme described customer harm, and 16.1% carried a vulnerability flag. No survey box says "I am vulnerable." The free text said it the whole time.
- Keep your survey. Stop treating it as the centre of gravity. Read what your customers are already telling you.
Most customer experience teams I meet are quietly optimising something that is dying.
Survey response rates have been falling for a decade. A "good" rate now sits in single digits. Everyone in the room knows it. And yet the same teams keep sending more surveys, shortening them, chasing the responders who are left, all to defend a number fewer and fewer customers are willing to give.
I spend my days inside the raw feedback of large consumer brands, banks and payments businesses. Here is what I have learned: the problem is not that customers have gone quiet. The problem is that we built our whole operating model around the one channel where they are going quiet, and stopped listening to the ten where they are still shouting.
RECOGNISE THE ROOM
You already know your VoC program is outdated. Here is how it shows.
This is the operating model most large CX organisations are still running. Read it as a checklist.
THE SAMPLE IS BROKEN
The responders are not your customers
A shrinking survey sample is not just small. It is skewed. The people who still respond are the extremes: the delighted and the furious.
The vast middle, the customers who are mildly annoyed, quietly drifting, one bad experience from leaving, they do not answer. They just churn. So the survey does not only under-count. It points you at the wrong problems. You end up managing your loudest customers and losing your silent ones.
And a survey is always late. By the time the score moves, the damage is done, the customer has told their friends, and the root cause is three weeks cold. You are steering by the rear-view mirror, down a road most people already left.
THE SHIFT
They are not silent. You stopped reading.
Your customers generate more honest, specific, unprompted feedback every day than any survey programme could dream of collecting.
They call your contact centre. They open tickets. They complain. They leave reviews. They tell you exactly what is wrong, in their own words, at the moment it happens, for free. The job of a modern CX team is not to ask better questions. It is to read what customers are already telling you, at scale, and turn it into a decision.
And when you read it instead of scoring it, you do not get a summary. You get the answer to the question behind the question: why is this happening, what is it costing, and what do we do. Here is what that looks like. Three real analyses, anonymised.
1 · IT STARTS WITH A HUMAN HUNCH
A spike anyone could see, and the cause no one could
It is a Monday. A contact-centre lead at a national retailer is running at 200% of forecast with a leadership review in a few hours, and a hunch she does not trust enough to say out loud. So she asks the data:
"We’re at 200% to forecast. I wonder if it’s the weather? Can you help me build the narrative for the leadership meeting."
A real question, asked before facing the room
A dashboard would have shown her the spike. She could already see the spike. What she needed was the cause. The answer did not just agree with her, it tested the hunch, confirmed it, then found the mechanism she had not considered.
Fan and air-conditioning calls had jumped thirteenfold in a week. But the real find was the cause under the cause: the website showed stock as available in-store when the shelf was already empty. Customers drove in, found nothing, and called in frustrated. Not a weather story. A fixable data problem, named and quantified, ready for the room. She walked in with the answer and the fix, not just a spike on a slide.
2 · CAUSE · PAYMENTS
The cause is often the opposite of what you assumed
A payments business was certain its complaints were an engineering problem: broken flows, failed transactions. Reading the actual tickets turned the diagnosis inside out. Three quarters of the CSAT failures had nothing to do with the product.
The survey said satisfaction was falling. The tickets said why, and named the department that owned the fix. Engineering was never going to solve a service problem.
3 · RISK · FINANCIAL SERVICES
The questions you did not know to ask
A financial-services client had one theme topping its complaints. The reflex is to hunt for a single broken system, but the complaints were spread almost evenly across back office, app, telephony and branch, which rules out a quick fix and points at something structural. Then the analysis surfaced a layer no survey question would ever reach for.
No customer ever fills in a survey box marked "I am vulnerable and you are failing me." It was there in the free text the whole time. In a regulated industry, reading it is the difference between catching an issue and explaining yourself to a regulator.
WHERE THIS LANDS HARDEST
Regulated, high-volume, multi-channel: the worst place to be survey-led
The bigger and more regulated the business, the more the silent majority costs you. Not in churn first, but in harm you find out about last.
Two rooms where I keep having the same conversation. In both, the conversations already contain the answers the VoC program cannot produce. These are the questions sitting unread in the free text.
THE REFRAME
This was never about the score
NPS and CSAT were always proxies. Cheap stand-ins for the questions the business actually cares about, defended so long we forgot they were stand-ins.
The specific, causal, act-on-it-tomorrow answer was sitting in the feedback customers were giving us anyway. We were just measuring around it.
The silent majority is not silent.
Keep your survey. It still has a place. But stop treating it as the centre of gravity. The centre of gravity is the signal your customers are already generating, in volume, whether or not you are listening.
Go and read what they are telling you or Book a demo
Send four weeks of conversations. We return what is breaking, what it is costing, and what to fix first, in 48 hours, drillable to every source conversation. Every analysis in this article started exactly this way.
A note on the numbers. Every figure here comes from a real SentiSum customer analysis and has been de-identified: no statistic is tied to a named company, and sector labels are deliberately broad. The pharmacy-retail and insurance questions above are exactly that, questions, drawn from the patterns we see in those industries’ conversations, not from client data. SentiSum turns unstructured customer feedback, across tickets, calls, reviews and complaints, into decisions for leading consumer and financial-services brands.
Frequently Asked Questions
Three reasons, in order of damage.
The sample is skewed. Only your loudest fans and your angriest churners respond. The mildly annoyed middle, the ones who quietly leave, do not fill in surveys.
The signal is late. By the time NPS moves, the customer has already told their friends. The root cause is three weeks cold.
The number is a proxy. It tells you something is wrong. It never tells you why, what it is costing, or what to fix. NPS was designed as a shortcut in 2003. The shortcut has become the destination.
As a trend line for the board deck, yes. As a decision-making tool for CX, no.
Keep tracking it. Stop treating it as the answer to what to fix next. The answer to that question is sitting in your tickets, calls and reviews, unread.
Nothing, and everything.
Do not chase a better score. Read the conversations your customers are already having with you. Every call, ticket, chat and review names the cause NPS only hints at. A specific store, a specific product, a specific agent behaviour, a specific policy.
The reframe is the point. Stop measuring around the answer. Start reading the answer.
Analysts sample. Machines read everything.
A team of three analysts can tag maybe 2% of your conversations in a good week. AI-native analysis reads 100% of them in an hour, tags every ticket, every call, every review by cause, sentiment, urgency and impact. Your analysts stop tagging and start acting on the patterns.
Same headcount. Different job. Bigger outcome.
Text analytics gives you word clouds and topic buckets. It answers what.
An AI-native reading layer names the cause, ranks it by impact, routes it to the team that owns the fix, and tracks it to resolution. It answers why, what it is costing, and what to do.
The test is simple. Ask your current tool: "which specific policy is driving the most compensation payouts this month, and which team owns the fix?" If you get a topic label back, you have text analytics. If you get an answer, you have a reading layer.
The audit takes four weeks of conversation data on your side and two days on ours. No integration.
You send historical data via file transfer. We return a document with the top failure patterns, the root causes, the commercial impact priced, and the fixes ranked. Every case study in the article started exactly this way.
If the audit shows nothing worth fixing, that is your win. You have proof your CX is healthy.
Explore Real Success Stories
Explore Success Stories
Talk to a Data Expert
