Almost every CX programme commits 3 deadly sins of analysis: sampling, silence and sentiment. Most teams commit all 3. One of them usually costs more than the other two, and which one depends on your business, not on best practice.
They’re easy to live with for years, because none of them throws up an error message. A sampled analysis reads like an analysis. A silent customer reads like a satisfied one. Three sentiment buckets read like a finding.
Sampling. You read a fraction of your feedback and assume it speaks for the rest.
Silence. The customers who score you 7 or 8 complain about nothing and quietly stop buying.
Sentiment. Three blunt buckets standing in for the emotion (the thing that predicts what customers do next).
Below, we take each sin in turn: how it hides, the kind of business where it’s usually the expensive one, and the first thing to check in your own data. You won’t need a new tool, and all 3 checks fit in an afternoon.
Sin 1. Sampling: the fraction you actually read
How it hides. Almost nobody reads all of their open-text feedback, and almost nobody says so out loud. The analysis gets done on whatever someone could get through: the most recent few hundred responses, the ones attached to the lowest scores, or whatever was in the export when the deck was due. Doing this manually runs to about 77 hours for every 700 comments. That’s 2 working weeks for one quarter of one channel.
A sample is riskier than it feels, because the useful comments are the minority. Gartner found that 83.6% of standard survey verbatims provide low insight value: vague, and short on context. The comments that would change your priorities are exactly the ones a partial read is most likely to miss.
Where it costs most. A few hundred accounts, each worth a lot: enterprise software, wholesale, commercial insurance, business banking, anything with a named account manager. When one customer is a real share of revenue, an unread comment isn’t a rounding error. A sample that skips the 3 accounts up for renewal has skipped the only part of the file that mattered this quarter.
The first check. Find the denominator. How many open-text responses arrived last quarter, across every channel, and how many did a person actually read? Write the ratio down. Then ask the harder question: were the ones you read chosen on purpose, or were they the first few hundred rows of the export? If it’s the second, you don’t have a sample. You have a sort order.
Sin 2. Silence: the customers who never make a list
How it hides. Only three out of ten customers are giving direct feedback (Qualtrics, 2026 Consumer Experience Trends). The same research found that 30% of consumers now stay silent after a bad experience, an all-time high since 2021.
The quiet group is also the biggest one. Passives, the customers who score you 7 or 8, average around 40% of a customer base across industries (Retently and CustomerGauge, 2024/2025 NPS Industry Benchmarks). Repurchase rates for Passive customers are 50% lower than Promoters (Bain, The Economics of Loyalty). And satisfied isn’t the same as safe: 60% to 80% of customer defectors said they were “Satisfied” or “Very Satisfied” on their last survey (Bain, The Ultimate Question 2.0).
It isn’t feedback fatigue, either. Gartner found that 81% of consumers say they provide customer feedback when asked to do so at least some of the time. Just 14% of consumers said it is only negative experiences that lead them to provide feedback. Customers are willing to talk. The moments you use to ask just aren’t catching the ones who are about to leave.
Where it costs most. A very large base, lower value each: retail, utilities, telecoms, mass-market banking, insurance, subscriptions. Nobody leaves loudly enough to trigger anything, so the loss arrives as thousands of small, quiet departures. The first place it shows up is a revenue line, months later.
The first check. Take everyone who churned in the last 12 months and ask how many of them gave you any feedback at all in their final 6 months. Not how they scored. Whether they spoke. If most of them were silent in the window where you could still have acted, your early warning is listening to a group that doesn’t include the people leaving. There’s a fuller version of this exercise in Silent Leavers: The Churn You Do Not Hear.
Sin 3. Sentiment: 3 buckets doing the work of 24 signals
How it hides. This is the sin teams are least likely to count, because it looks like a solved problem. The text comes back tagged positive, negative or neutral, the percentages go on a slide, and the slide looks like analysis.
But 3 categories can’t hold what customers write. Take a comment you’ll recognise: “it was fine, took a while, the third time I called someone finally sorted it”. That’s resignation, low-grade frustration and fading confidence, scored a 7 because the problem did get fixed in the end. A 3-bucket model files it as neutral, right next to a contented customer with nothing to report. Intensity goes missing too. “I’m a bit annoyed” and “I’ve had enough of this” are both negative. One is a service ticket. The other is a cancellation waiting to happen.
That’s the difference between sentiment and emotion. Sentiment tells you a customer is unhappy. Emotion tells you why, how much, and what it’s costing. Adoreboard reads feedback across 24 emotional signals: eight base emotions at three levels of intensity, grounded in Plutchik’s model of emotion. Those signals build a Trust Score, and Trust moves before customer behaviour does.
Where it costs most. Anywhere the decision to leave builds slowly rather than being triggered: renewals, memberships, subscriptions, long contracts, advice relationships. There’s no single bad moment to catch, just a drift that lives in the intensity of the language. In Forrester’s 2026 US Customer Experience Index, of the three dimensions, emotional quality had the largest gain, increasing by 1.4 percentage points year over year.
The first check. Pull 20 comments your current tool classed as neutral and read them yourself. Count how many are genuinely indifferent. In our experience the neutral pile is where the resigned customers sit, and a team that reads 20 of them stops arguing about this sin pretty quickly.
Which sin is costing you most
A company with a few hundred high-value accounts and one with millions of customers rarely lose money in the same place. The shape of your customer base decides it.
| The shape of your base | Usually the expensive sin | Why | The first check |
|---|---|---|---|
| A few hundred accounts, high value each | Sampling | One unread comment can carry a real share of revenue | The denominator: comments received against comments read |
| A very large base, lower value each | Silence | The loss is thousands of quiet departures, none big enough to escalate | How many of last year’s churned customers spoke at all |
| Renewal or subscription, any size | Sentiment | The decision to leave builds over months, in language 3 buckets can’t separate | Read 20 comments your tool called neutral |
Most programmes commit all 3, so this is about the order you fix them in. And sentiment sits underneath the other two: fixing coverage means you read everything, and fixing measurement means what you read is worth something. Do it the other way round and you’ve read everything through a lens that can’t tell annoyance from departure.
Run the 3 checks and you’ll have one sentence to take to your exec team, and it’ll land harder than any benchmark because it’s about your own customers: we read this fraction of what our customers told us, this share of the people who left never told us anything, and this many of the comments we call neutral came from people who’d given up on us. To turn that into a money argument, ranking your themes by volume and then by the revenue of the customers who raised them, see What NPS Cannot Tell You.
What fixing all 3 looks like
Each sin has its own fix, and the fixes need to be one system, or they undo each other.
Read everything, not a sample, and not just the surveys. Adoreboard’s Deep Semantic Clustering technique analyses customer feedback 923 times faster than manual human analysis, reducing processing time per comment from 30 seconds to 0.03 seconds. That’s what turns full coverage from a staffing problem into a monthly routine.
Hear intensity, through 24 emotional signals rather than 3 buckets, building into a Trust Score. That’s what separates the customer who’s annoyed from the customer who’s decided. Then put a money figure on it. Customer Revenue Impact takes those signals and the revenue of the customers behind them, and gives you revenue-ranked priorities: every issue in your feedback, ordered by what it’s costing, not by how often it’s mentioned.
It holds up under testing. In a validation study of more than 20,000 customer records from a major North American telecommunications provider, emotion-based measurement predicted churn with 86% accuracy, landing within 1% of actual historical churn.
3 sins, and the one costing you most is answerable this week with data you already hold. Worth knowing before you set your next CX priorities.
Curious what this looks like on your own data? Book a 15-minute scoping call: adoreboard.com/demo
Source: Gartner, Increase the Insight Value of Customer Verbatims, Customer Experience Research Team, 19 December 2022, ID G00705811.
Source: Gartner, How to Boost Voice-of-the-Customer Survey Response Rates, Augie Ray, Deborah Alvord, Christopher Sladdin, 26 April 2024, ID G00797253.
GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally. Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.
Source: Emotion Leads CX Quality Rebound In The US, Forrester, 3 August 2026.
Net Promoter®, NPS®, NPS Prism®, and the NPS-related emoticons are registered trademarks of Bain & Company, Inc., NICE Systems, Inc., and Fred Reichheld. Net Promoter ScoreSM and Net Promoter SystemSM are service marks of Bain & Company, Inc., NICE Systems, Inc., and Fred Reichheld.
