Every CX dashboard carries an invisible footnote: these results describe the customers who chose to answer. For many programs that is a small share of the customer base, and the people who stay silent are not a random sample of the rest.
This article examines non-response bias, one of the least discussed stall points in the VoC execution gap. It is a methodological problem rather than a technology problem, and it can be addressed with established survey-research practice.
The Short Answer
Non-response bias occurs when the customers who answer a survey differ systematically from those who do not, on the very measures the survey is trying to estimate. A low response rate does not automatically cause bias, and a high response rate does not guarantee its absence. What matters is whether the reasons customers respond or stay silent are related to their experience. When they are, scores can look healthier, or worse, than reality, and priorities drift toward the concerns of the most vocal.
Why the Response Rate Is Not the Real Problem
It is intuitive to treat response rate as a quality indicator. Survey methodologists have shown that intuition is unreliable.
Robert Groves' review of household surveys, published in Public Opinion Quarterly in 2006, concluded that response rates are a weak predictor of nonresponse bias. Bias depends on how strongly the likelihood of responding is correlated with each specific measure (Groves, 2006). A meta-analysis by Groves and Emilia Peytcheva across 59 studies reached the same conclusion: most of the variation in bias occurred between different measures within the same survey, not between surveys with different response rates (Groves & Peytcheva, 2008).
These studies come from household and public-opinion research, not customer experience. The principle transfers directly, though: a CX survey can be unbiased on one question and badly biased on another, depending on who is motivated to answer.
Who Stays Silent, and Why It Matters
Customer research suggests that silence is common and that it is not neutral.
Qualtrics XM Institute's 2024 consumer feedback study, based on 28,400 consumers in 26 countries, found that after a very poor experience only 34% sent feedback directly to the company, while 21% told no one at all. The share giving direct feedback after a bad experience had fallen by 7.2 percentage points since 2021 (XM Institute, February 2024).
Survey participation in service is also low. A Gartner study of 89 customer service and support leaders reported an average survey response rate of 23%, with 38% of leaders reporting single-digit response rates (Gartner, September 2022). McKinsey estimated that a typical CX survey samples about 7% of a company's customers, and only 13% of the CX leaders it surveyed were fully confident their measurement gave a representative view of their customer base (McKinsey, February 2021).
Both studies are a few years old and drawn from specific populations, so the exact figures should not be generalized. The direction is consistent, however: most customers' experiences never reach the survey.
Evidence That Silence Is Not Neutral
One of the clearest illustrations comes from a study of more than 166,000 live-chat sessions in a large electronics manufacturer's US customer service operation. Only 16.2% of customers rated their chat. Researchers Kunwoo Park, Meeyoung Cha and Eunhee Rhim trained models on the chat transcripts of customers who did rate, then inferred satisfaction for those who did not.
The limitations matter. The study covers one company and one channel, and the non-raters' satisfaction is model-inferred and was validated against only a small human-coded sample. It does not prove that silent customers are generally unhappy. It does show that, at least in some settings, the survey picture can be markedly more positive than the underlying experience.
Channel Bias: What You Hear Depends on Where You Ask
Non-response interacts with channel choice. The same XM Institute study found that website surveys skew toward positive experiences while emails skew toward negative ones: consumers were 12 percentage points more likely to complete a website survey after a very good experience than after a very bad one, and 12 points more likely to email a company after a bad experience than after a good one.
Forrester's 2026 predictions add a structural concern: it expects response rates from generic website feedback collection to "sink more due to the decline in traffic on B2B and B2C websites" (Forrester, October 2025). A channel mix designed several years ago may now be hearing from a different, and narrower, group of customers.
Gartner's research note It's Not Survey Fatigue: The Real Reason Customers Don't Reply (Gartner, August 2025) signals that the causes of low participation are worth examining rather than assuming.
Detecting and Reducing Non-Response Bias
Addressing blind spots in your feedback data requires two disciplined routines: diagnosing where representation breaks down, and widening how signals are captured.
What Not to Do
Some common tactics raise response rates while increasing bias. Sending surveys only after successful interactions, asking agents to request high scores, or offering incentives that appeal to particular segments can all make the respondent group less representative. The goal is not more responses. It is responses that describe the customer base.
Conclusion: Treat Silence as Data
A program that reports only what respondents say is reporting on a self-selected group. The most mature programs treat the gap between respondents and the full customer base as something to measure, explain and narrow, using comparison, follow-up and operational evidence.
The next article in this series moves from who is heard to who acts: the question of who owns the fix once a problem is visible.
Frequently Asked Questions
