4  Customer Satisfaction

Customer satisfaction is the evaluative judgment a consumer forms after comparing the performance of a product or service against some standard of reference. It is at once the most surveyed construct in marketing practice and one of the most theoretically demanding: a satisfaction score is not a direct reading of an attitude, but the residual of a comparison process, and the standard against which performance is judged is itself shaped by the very experience being evaluated. This chapter treats satisfaction as two coupled objects. The first is a psychological construct: a post-purchase judgment built from affective and cognitive inputs and anchored to prior expectations. The second is a financial asset: a forward-looking signal that, when it moves, moves cash flows, equity returns, and the risk those returns carry. A serious account must connect the two, and the modern empirical literature, drawing heavily on the American Customer Satisfaction Index (ACSI), increasingly does.

The chapter proceeds from formation to consequence. We begin with how satisfaction is formed (i.e., the expectancy–disconfirmation paradigm and its affective and cognitive components) and how it is measured, including the modern move from survey scales to text-derived sentiment. We then formalize the disconfirmation model to make its identification problems explicit. From there, we turn to consequences: the satisfaction–profit chain that runs from cash flow growth and volatility to shareholder value; the asymmetric and sector-specific way equity markets price satisfaction; and the effect of satisfaction on the risk of returns rather than their level. We confront two findings that discipline naive optimism (i.e., the negative market-share–satisfaction relationship and the moderating role of financial leverage) and close with monopoly settings and the communication of uncertainty.

4.1 Formation: The Expectancy–Disconfirmation Paradigm

The dominant theory of how satisfaction arises is the expectancy–disconfirmation model. Its intuition is ordinary: a diner who expects a mediocre meal and receives a good one leaves delighted, while a diner who expects greatness and receives merely good leaves disappointed, even though the meals were identical. Satisfaction tracks not performance itself but performance relative to a pre-consumption expectation. The gap between the two is disconfirmation: positive when performance exceeds expectations, negative when it falls short, and zero when they coincide.

Let \(E\) denote a consumer’s pre-consumption expectation of performance along some attribute, \(P\) the perceived actual performance, and \(D = P - E\) the disconfirmation. The paradigm posits that satisfaction \(S\) rises in both the baseline expectation and the disconfirmation, \[S = \alpha + \beta_E\, E + \beta_D\, (P - E) + \varepsilon, \tag{4.1}\] with \(\beta_D > 0\) the central claim of the theory and \(\beta_E\) capturing the direct, assimilative pull of expectations on the final judgment. Equation Equation 4.1 makes the construct’s subtlety precise: because \(E\) enters both directly (through \(\beta_E\)) and as the benchmark inside the disconfirmation term, raising expectations has offsetting effects on satisfaction. It lifts the assimilative component while simultaneously raising the bar that performance must clear, a tension that any manager who has over-promised understands and that any estimator of Equation 4.1 must confront.

A field study of a novel restaurant experience by Oliver and Burke (1999) sharpens the dynamics behind this static equation. Pre-consumption expectations were manipulated with realistic reviews, the dining experience itself was recorded by protocol methods, and perceptions were captured immediately after. Four results recur in subsequent work. First, the manipulated expectation exerts an immediate effect that decays over the course of the experience, so the weight \(\beta_E\) is not constant but fades as direct evidence accumulates. Second, expectations shape perceived performance (i.e., forward assimilation) so \(E\) and \(P\) are not independent inputs but causally linked, which (as we note below) breaks the clean separation Equation 4.1 assumes. Third, expectations are revised retrospectively by what actually happened (i.e., backward assimilation), so the measured \(E\) depends on \(P\) when it is elicited after consumption. Fourth, the disconfirmation \(D\) remains the pivotal driver of \(S\), and the whole process operates in a dimension-specific way: disconfirmation on one attribute drives satisfaction with that attribute, not a diffuse global affect.

Disconfirmation. Satisfaction is the consumer’s response to the perceived discrepancy between prior expectations and the actual performance of a product as experienced after consumption.

4.1.1 Affect and Cognition

A consumer’s assessment of satisfaction draws on two channels that the disconfirmation equation collapses into a single \(S\). The cognitive channel is the deliberate, attribute-by-attribute comparison of performance against expectation (i.e., precisely the \(P - E\) computation of Equation 4.1). The affective channel is the emotional residue of consumption: the pleasure, delight, irritation, or regret experienced during use, which colors the judgment independently of any calculated discrepancy. The two are not substitutes. A technically flawless service encounter delivered coldly can produce low satisfaction despite zero cognitive disconfirmation, and a warm recovery from a service failure can produce high satisfaction despite a large negative one. Treating satisfaction as a purely cognitive ledger, therefore, mis-specifies its formation; the affective component is a separate input, and measurement instruments that capture only attribute ratings miss it.

4.2 Theoretical Foundations

Satisfaction is a comparison judgment, and the theories that ground it are theories of what gets compared, against what standard, and with what asymmetry. The organizing account is the expectancy-disconfirmation paradigm developed above: satisfaction tracks performance relative to a pre-consumption expectation, with the disconfirmation \(D = P - E\) as the pivotal driver (Oliver and Burke 1999). (Oliver’s original 1980 cognitive model of satisfaction established the paradigm; the field study cited here sharpens its dynamics.) Disconfirmation is itself a reference-dependence claim, which is why prospect theory is the second foundation (Kahneman and Tversky 1979; Tversky and Kahneman 1981). The expectation \(E\) is a reference point, performance is coded as a gain or a loss relative to it, and because losses loom larger than gains, negative disconfirmation depresses satisfaction more than equal positive disconfirmation raises it. This loss-averse asymmetry reappears downstream in how equity markets price satisfaction news, where dissatisfaction loads roughly 2.4 times as heavily as an equal satisfaction gain (Malshe, Colicev, and Mittal 2020).

Two motivational and informational theories refine the comparison. Regulatory focus (Higgins 1997) holds that a promotion-focused consumer evaluates an experience against ideals and the presence or absence of gains, while a prevention-focused consumer evaluates against oughts and the presence or absence of losses; the same objective performance therefore yields different satisfaction depending on which standard the consumer brings, and on whether the offering’s framing fits that orientation. The affect-as-information account (Schwarz and Clore’s work on feelings as information) grounds the affective channel introduced above: consumers read their momentary feelings during consumption as data about how satisfied they are, which is why a warm recovery can lift satisfaction despite zero cognitive disconfirmation, and a cold but flawless encounter can depress it. This is the theoretical reason the affective input cannot be reduced to the cognitive ledger of Equation 4.1. The asymmetry between the two valences is itself asymmetric across outcomes: Mittal, Ross, and Baldasare (1998) show that negative and positive information about a firm are weighted very differently depending on whether the judgment concerns the product or the relationship, so a single loss-aversion coefficient understates the structure.

A further complication is that satisfaction depends on who is held responsible for the outcome, not only on the outcome. Botti and McGill (2011) show that a consumer who chose an option and got a poor result is less satisfied than one who received the same result from someone else’s choice, because personal causality converts a bad draw into a bad decision. This matters directly for the design of assortments, recommendation systems, and self-service: shifting the locus of choice to the customer shifts the locus of blame with it.

Two further theories bound the construct. Equity (fairness) theory (Adams’s work on inequity, extended to consumption by Oliver and Swan) holds that satisfaction depends not only on one’s own outcome-to-input ratio but on its comparison to the seller’s and to other customers’, so a consumer who feels fairly treated can be satisfied with a modest outcome and one who feels exploited dissatisfied with a good one; this is the channel through which perceived price fairness and service-recovery justice enter satisfaction, and it links to the fairness constraint on pricing in Chapter 20. Finally, adaptation-level dynamics explain why satisfaction is unstable over time: the reference point itself migrates toward recent experience, so a sustained improvement is re-encoded as the new normal and ceases to delight, the hedonic-adaptation logic behind the decay of the expectation effect documented in Oliver and Burke (1999) and behind the difficulty of maintaining satisfaction through incremental gains. Together, these theories explain why the construct is the residual of a comparison rather than a direct attitude reading, and why its measurement and its financial consequences (the subjects of the remaining sections) inherit the reference-dependence and asymmetry built into its formation.

4.3 Measurement

Historically, satisfaction has been measured by survey: a respondent rates overall satisfaction, expectations, and performance on Likert or semantic-differential scales, and these ratings are aggregated—at the national level by indices such as the ACSI, whose structure, purpose, and estimated cross-industry findings are set out in Fornell et al. (1996), and which anchors most of the marketing–finance evidence discussed below. Survey measurement is well understood, but it is slow, sparse, and reactive: it samples a fraction of customers at fixed intervals and asks them to introspect, which invites recall and social-desirability bias.

The frontier alternative reads satisfaction directly from what customers write. Sentiment analysis treats a corpus of customer-generated text—reviews, support transcripts, social posts—as a high-frequency, census-scale signal of satisfaction, classifying the polarity (and increasingly the intensity and target) of expressed affect (S. Kumar and Zymbler 2019; Wei et al. 2020). The appeal is operational as much as scientific: a sentiment pipeline observes the affective channel of Chapter 4 in near real time, covers customers who would never return a survey, and scales to firms and markets that no survey panel reaches.

The two measurement regimes trade off along several axes, summarized in Table 4.1. Neither dominates: survey scales remain the validated anchor against which text measures are benchmarked, while sentiment supplies coverage and timeliness that surveys structurally cannot.

Table 4.1: Survey versus text-based measurement of satisfaction
Property Survey scales (e.g., ACSI) Text sentiment
Sampling Panel, intermittent Census of writers, continuous
Construct captured Cognitive + elicited affect Expressed affect, in context
Latency Weeks to quarters Near real time
Principal bias Recall, social desirability Self-selection of who writes
Validation status Field standard Benchmarked to surveys

A worked illustration makes the text measure concrete. A minimal lexicon sentiment score classifies each document by the net count of positive and negative tokens it contains, normalized by length—a transparent baseline that more elaborate supervised classifiers refine but do not conceptually replace.

Code
set.seed(42)

# Minimal, dependency-free lexicon sentiment score in [-1, 1].
positive <- c("great", "excellent", "love", "happy", "satisfied",
              "recommend", "good", "fast", "reliable", "delighted")
negative <- c("bad", "terrible", "hate", "slow", "broken",
              "disappointed", "refund", "rude", "late", "useless")

sentiment_score <- function(texts) {
  clean  <- gsub("[^a-z ]", " ", tolower(texts))
  tokens <- strsplit(trimws(clean), "\\s+")
  vapply(tokens, function(w) {
    p <- sum(w %in% positive); n <- sum(w %in% negative)
    if (p + n == 0) 0 else (p - n) / (p + n)
  }, numeric(1))
}

reviews <- c(
  "Great service and a reliable product, I would recommend it.",
  "Terrible experience, the item was broken and the staff were rude.",
  "It arrived; nothing remarkable to report about the purchase."
)
data.frame(review = reviews, score = sentiment_score(reviews))
#>                                                              review score
#> 1       Great service and a reliable product, I would recommend it.     1
#> 2 Terrible experience, the item was broken and the staff were rude.    -1
#> 3      It arrived; nothing remarkable to report about the purchase.     0

4.4 Estimation and Identification

The static disconfirmation model Equation 4.1 looks like an ordinary regression of satisfaction on expectations and a performance gap, estimable by ordinary least squares (OLS). The dynamic field evidence shows why that estimator is fragile. Three assumptions must hold for the OLS coefficients to carry their structural interpretation, and the Oliver and Burke (1999) findings attack each in turn.

The first assumption is exogenous expectations: \(E\) must be uncorrelated with the error \(\varepsilon\). Backward assimilation violates this directly. When \(E\) is elicited after consumption, the measured expectation is contaminated by the realized performance, so \(\mathrm{Cov}(E, \varepsilon) \neq 0\) and \(\hat \beta_E\) is biased. Eliciting \(E\) strictly before consumption—as the manipulation in the field study did—is the design that restores exogeneity; a post-hoc expectation measure cannot.

The second assumption is separability of \(E\) and \(P\). Forward assimilation violates it: because expectations shape perceived performance, \(P = \gamma E + u\) with \(\gamma > 0\), the regressors \(E\) and \(D = P - E\) are mechanically correlated, inflating standard errors and making the separate identification of the assimilative path (\(\beta_E\)) from the contrast path (\(\beta_D\)) delicate. Experimental manipulation of \(E\) independent of \(P\) is again the clean route to separation; in purely observational data the two effects are only weakly distinguishable.

The third assumption is stable coefficients. The decay of the expectation effect over the consumption episode means \(\beta_E\) is time-varying, so a single cross-sectional regression estimates a horizon-specific average rather than a structural constant. The practical implication is that satisfaction must be modeled dimension by dimension and stage by stage; a pooled global regression averages over heterogeneity the theory says is real. Figure 4.1 collects these paths: the solid arrows are the structural relationships of Equation 4.1, and the dashed arrows are the assimilation effects that defeat the OLS estimator.

flowchart LR
  E["Pre-consumption<br/>expectation E"] -->|beta_E| S["Satisfaction S"]
  E --> D["Disconfirmation<br/>D = P - E"]
  P["Perceived<br/>performance P"] --> D
  D -->|beta_D| S
  E -.->|forward<br/>assimilation| P
  P -.->|backward<br/>assimilation| E
Figure 4.1: The expectancy-disconfirmation process. Solid arrows are the structural paths of equation Equation 4.1; dashed arrows are the assimilation effects that break the identifying assumptions of an OLS estimator.

4.5 Consequences: The Satisfaction–Value Chain

Why does a post-purchase judgment matter to a firm’s owners? The answer is that satisfaction is a leading indicator of cash flow. Satisfied customers buy again, defect less, cost less to serve, and complain less, so a movement in satisfaction today forecasts the level, growth, and volatility of future cash flows—and the present value of those flows is what equity is worth. The chain runs satisfaction \(\to\) cash flow characteristics \(\to\) shareholder value (Figure 4.2), and each link has been estimated.

Gruca and Rego (2005) make the cash-flow link explicit. Combining longitudinal ACSI data with COMPUSTAT financials and estimating the relationship in a hierarchical Bayesian framework—appropriate because firms are sampled repeatedly and the satisfaction effect is allowed to vary across them—they find that customer satisfaction both raises the growth of future cash flows and lowers their variability. Because shareholder value rises with expected cash flow and falls with its risk, satisfaction contributes to value through both channels at once. The result is robust across firm and industry characteristics and survives multi-measure, multi-method re-estimation, which matters because a single ACSI series invites concern about measurement idiosyncrasy.

Not every satisfaction metric carries this predictive content, however. Firms collect many customer-feedback measures, and Morgan and Rego (2006) adjudicate among six of them—top-2-box satisfaction, average satisfaction, repurchase-intention loyalty, net promoters, repurchase likelihood, and the proportion of customers complaining—against COMPUSTAT and CRSP performance over 1994–2000. The ranking is instructive and somewhat deflating for popular practice. Average satisfaction scores are the most predictive of future business performance, with top-2-box scores close behind; repurchase likelihood and complaint proportion carry moderate, performance-dimension-specific value; and recommendation-based metrics—net promoters and the average number of recommendations—show little to no predictive value. The metric a firm chooses to optimize is therefore not innocuous: an obsession with recommendation scores can mistake a poorly predictive proxy for the asset itself. Keiningham et al. (2007a) reach the same verdict from a different direction, tracking net promoter against firm revenue growth over time and finding the relationship far weaker and less stable than the metric’s promoters claim.

Controversy: the one number, and what happened to it

The single most widely adopted customer metric in industry entered the academic literature as a claim that the literature then declined to confirm. The sequence is short and unusually clean.

2003 — the claim. A Harvard Business Review article proposes that a single question—how likely a customer is to recommend the firm—yields a score that predicts growth better than any other customer measure, and that a firm needs essentially no other. The idea is adopted with remarkable speed; within a few years the score is a board-level metric at a large share of major firms and, in many, an input to executive compensation.

2006–2007 — the academic response. Morgan and Rego (2006) adjudicate six customer-feedback metrics against COMPUSTAT and CRSP performance and rank recommendation-based measures last. Keiningham et al. (2007b) test the growth claim directly and longitudinally, using the metric’s own published methodology, and do not find the claimed superiority. Keiningham et al. (2008) extend the assessment.

What followed. No retraction, because there was nothing to retract—the original claim was made in a practitioner outlet, not a peer-reviewed one, and practitioner outlets do not run replications. The academic finding and the industry practice simply proceeded in parallel, and largely still do. Two decades later the metric remains in wide managerial use, and the peer-reviewed evidence that it predicts growth better than alternatives remains thin.

That divergence is the lesson, and it is not really about this metric. A measure can win on properties that have nothing to do with predictive validity: it is one question, it produces a single number, it is comparable across units, and it is easy to put in a bonus formula. Those are real virtues for a management-control system. They are not evidence, and Section 31.2.6 develops what goes wrong when a metric is optimized on that basis. The reviewer’s version of the question is in Section 76.8.5: when a claim is popular and the disconfirming evidence is published, what mechanism is supposed to close the gap?

flowchart LR
  SAT["Customer<br/>satisfaction"] --> RET["Retention &<br/>repurchase"]
  SAT --> COST["Lower<br/>cost-to-serve"]
  RET --> CF["Cash-flow<br/>growth"]
  RET --> VOL["Lower cash-flow<br/>volatility"]
  COST --> CF
  CF --> SV["Shareholder<br/>value"]
  VOL --> SV
Figure 4.2: The satisfaction-to-shareholder-value chain. Satisfaction is priced because it forecasts the level, growth, and volatility of future cash flows.

Figure 4.2 draws every arrow with a positive sign, and that is the version of the chain most often reproduced in practice. It is incomplete in a way that matters. Satisfaction is bought, not found, and the price rises as the level does. Guenther and Guenther (2021) follow roughly one hundred U.S. firms over fifteen years and separate the benefits of a satisfaction improvement from its costs along the satisfaction continuum. Revenues and savings in marketing and acquisition costs can indeed accelerate at high satisfaction levels; but for many firms operating costs and capital investment accelerate too, and shareholder returns to satisfaction improvements can become negative. Worse, the two sides decay at different rates: operating-cost increases are sticky while top-line benefits fade. The payoff is contingent on the firm’s strategic situation—offering customization, word-of-mouth potential, and competitive innovation pressure all condition it. The managerial reading is not that satisfaction fails to create value but that the chain in Figure 4.2 is a gross rather than a net relationship, and that a firm can climb it past the point where the marginal dollar of satisfaction returns less than it costs.

4.6 How Markets Price Satisfaction

If satisfaction forecasts cash flows, an efficient equity market should impound it into prices. Whether it fully does—and where—has been the subject of a long empirical argument that turns on a subtle point: a satisfaction level that is already public should be priced, so the testable object is whether unexpected changes in satisfaction earn abnormal returns.

4.6.1 The Satisfaction Anomaly and Its Replication Record

That argument is worth following in full, because it is the most instructive replication episode the marketing literature has produced, and because it ended somewhere other than where most textbook treatments leave it. It runs for three decades, and it is best read in order.

Controversy: three decades of the satisfaction anomaly

Every statement below is drawn from a published article or a public regulatory filing, cited at the point it is made. The account describes what was written and filed, and the dates on which that happened. It attributes no motive to anyone, and asserts no view about any person’s conduct.

1994 — the instrument is built. The American Customer Satisfaction Index is established by researchers at the University of Michigan together with the American Society for Quality and CFI Group, adapting the earlier Swedish barometer.1 Its structure and first cross-industry estimates are set out in Fornell et al. (1996). At this stage the ACSI is a measurement project, and its scientific standing rests on the quality of the index.

2006 — the claim. Fornell et al. (2006), in the Journal of Marketing, report that portfolios formed on ACSI strength earn positive abnormal returns at the overall market level. The paper’s own title states the finding as High Returns, Low Risk. Two years later Aksoy et al. (2008) extend it to long-horizon valuation. The implication is large: a marketing construct is systematically mispriced by equity markets, which would make satisfaction not merely a strategic asset but a source of alpha.

2009 — the field pushes back, in one issue. Marketing Science devotes a section of volume 28, issue 5 to the question, and it is a genuine confrontation rather than a symposium. Jacobson and Mizik (2009b) re-examine the evidence and locate the abnormal returns in a handful of Internet and computer firms rather than across the market. Fornell, Mithas, and Morgeson (2009) reply in a commentary defending the economic and statistical significance of the effect. Ittner, Larcker, and Taylor (2009) — an accounting team with no prior stake in the marketing debate — run conventional multifactor asset-pricing tests and report no evidence that ACSI data earn abnormal returns. Jacobson and Mizik (2009a) then close the section with a rejoinder. In the same year and independently, O’Sullivan, Hutchinson, and O’Connell (2009) apply standard risk adjustments in the International Journal of Research in Marketing and also find nothing significant. Four pieces in one issue, plus a fifth elsewhere, in a single year, on one paper.

2014 — the portfolio construction is reopened. Bell, Ledoit, and Wolf (2014) return to the question with a new portfolio-formation method, applied specifically to customer satisfaction; the working-paper version carried the blunter title Reexamining Possible Mispricing of Customer Satisfaction. Their contribution locates the dispute where it belongs. Whether an anomaly appears depends on how portfolios are formed and weighted, so a claim that survives only under one construction is a claim about that construction.

August–October 2016 — the claim is productized. On 5 August a registration statement is filed with the SEC for an exchange-traded fund tracking an ACSI-based index. The filing states that the fund’s investment adviser is also its index provider, that the index organization is “an affiliate of the Adviser,” and that “Mr. Claes Fornell has a controlling interest in the Adviser and ACSI”; it discloses the resulting conflict and describes a mitigation, an unaffiliated third party retained to calculate the index.2 In September, Fornell, Morgeson, and Hult (2016b) appears in the Journal of Marketing reporting cumulative returns of roughly \(518\%\) against \(31\%\) for the S&P 500 over 2000–2014. On 31 October the fund begins trading.

September 2016 — the second confrontation. The Journal of Marketing publishes the critiques in the same issue as the claim, and the framing is itself remarkable: V. Kumar (2016) opens the section with the question Is Customer Satisfaction (Ir)relevant as a Metric?—a startling thing for the field’s flagship journal to ask about its most famous construct. Sorescu and Sorescu (2016) and Bharadwaj and Mitra (2016) then reply, and Fornell, Morgeson, and Hult (2016a) closes with a rejoinder titled An Abnormally Abnormal Intangible.

The substance of those two replies matters more than the arithmetic, because it is a stronger objection than “we got a different number”: both report that they could not evaluate the result at all, because the trading strategy on which it rested had not been disclosed. A finding whose construction is proprietary cannot be reproduced and cannot be falsified. That this was stated plainly in the pages of the field’s flagship journal, and that the exchange nonetheless concluded without resolution, is the part of the episode a doctoral reader should sit with longest.

2019–2026 — the meta-analyses arrive, and reframe everything. Otto, Szymanski, and Varadarajan (2020) synthesize over a quarter century of empirical work on satisfaction and firm performance. Mittal et al. (2023) go wider, pooling 535 correlations from 245 articles with a combined sample above \(1.1\) million, and report positive associations across customer- and firm-level outcomes whose magnitude varies substantially with context and measurement. Mittal, Singh, and Malshe (2026) replicate and extend that evidence on more than 285 million ratings across 1,126 brands. These do not vindicate either camp outright, and that is the point. They establish an association that is real, positive, and heavily conditional—which is a different object from a stable, exploitable pricing anomaly (Section 4.6.3).

The prior nobody brought. One thing is conspicuously missing from twenty years of this exchange: any reference to what happens to published return predictors in general. Finance has audited its own anomaly literature systematically, and the verdict is that such findings replicate poorly, decay after publication, and are highly sensitive to weighting and universe choices (Section 25.8.1). A referee armed with that reference class would have asked, in 2006, whether the result was value weighted, whether it survived outside small illiquid stocks, and what happened after the strategy became public. Those three questions do not require any view about customer satisfaction at all.

What the episode is, and is not. No data were alleged to be fabricated. No paper was retracted. No institution found misconduct. The literature simply produced a striking claim, five independent teams could not reproduce it, the disagreement was published twice without resolution, and the strategy has since accumulated a decade of public out-of-sample record that any reader can check rather than take on authority.3 That is a failure mode with no villain and no enforcement mechanism, which is precisely why it repays study: it is far more common than fraud, and much harder to see. Section 76.8.4 takes up what a reviewer should do about it.

Table 4.2: The replication record of the customer-satisfaction stock-return anomaly. The affirmative results come from one research group; the null results come from five independent teams.
Study Venue Verdict on tradable abnormal returns
Fornell et al. (2006) Journal of Marketing Positive, market-wide
Aksoy et al. (2008) Journal of Marketing Positive, long-horizon
Jacobson and Mizik (2009b); Jacobson and Mizik (2009a) Marketing Science Confined to a few Internet/computer firms
Ittner, Larcker, and Taylor (2009) Marketing Science None under standard multifactor tests
O’Sullivan, Hutchinson, and O’Connell (2009) IJRM None after standard risk adjustment
Bell, Ledoit, and Wolf (2014) Customer Needs and Solutions Reopens portfolio formation; counted among the non-replications
Fornell, Morgeson, and Hult (2016b) Journal of Marketing Positive; \(518\%\) vs. \(31\%\), 2000–2014
Sorescu and Sorescu (2016); Bharadwaj and Mitra (2016) Journal of Marketing Not evaluable; strategy undisclosed

Why the structure matters, stated carefully. The configuration the filing documents—the same individual having created the measure, co-authored the studies establishing that the measure predicted abnormal returns, and held a controlling interest in the firm that sold the resulting product—does not by itself invalidate any finding, and this book asserts no view about anyone’s conduct. Nor was any of it concealed: it is filed with the SEC and public, and the prospectus discloses the conflict and describes a mitigation. The narrow, structural point for a reader of the academic literature is that this is exactly the kind of information bearing on how much independent replication a claim should require before a field treats it as established. In this instance the field required, and got, rather a lot.

What survives. It would be wrong to conclude that satisfaction is unrelated to firm value—that is a different claim, and it holds. The distinction is between an association and a tradable anomaly, and the meta-analytic evidence in Section 4.6.3 separates them cleanly. The tradable anomaly is the claim that failed.

The methodological lesson generalizes past this case. Asset-pricing tests of marketing constructs are acutely sensitive to portfolio formation, risk adjustment, and the sample window; a result that requires a particular undisclosed construction to appear is, for that reason alone, not yet a finding. Section 76.8.4 returns to this episode as the field’s own contribution to the credibility revolution, and Section 57.5.2 works through the estimating equations in detail.

The pricing of satisfaction is also asymmetric, and the asymmetry has a behavioral signature consistent with prospect theory. Malshe, Colicev, and Mittal (2020) estimate that a one-unit increase in customer satisfaction raises abnormal returns by about \(0.56\) percentage points, while a one-unit decrease lowers them by about \(1.34\) percentage points—dissatisfaction is priced roughly \(2.4\) times as heavily as an equal-sized satisfaction gain, the loss-aversion pattern in which losses loom larger than gains. The mechanism is informative: short interest, a measure of short-seller activity, mediates the effect of customer (dis)satisfaction on abnormal returns, so sophisticated traders act on satisfaction news and transmit it into prices. The policy implication for managers is stark—defending against dissatisfaction protects value more than an equivalent investment in raising already-satisfied customers does.

4.6.2 Satisfaction and the Risk of Returns

Most of the discussion above concerns the level of returns. Satisfaction also affects their risk, which matters independently because a risk reduction lowers the discount rate applied to future cash flows and thus raises value even if expected returns are unchanged. Tuli and Bharadwaj (2009) decompose this effect: positive changes in customer satisfaction produce negative changes in systematic risk (both overall market-movement exposure and downside exposure specifically) and in idiosyncratic risk (the firm-specific volatility of returns). A more satisfied customer base therefore makes a firm’s equity less volatile and less exposed to market downturns—precisely the cash-flow-stability channel of Gruca and Rego (2005), now visible in the second moment of returns rather than the first. The marketing–finance machinery used to estimate these abnormal-return and risk effects—market models, event windows, and the screening of confounds—is developed in Chapter 25.

4.6.3 How Large Is the Effect? The Meta-Analytic Anchor

Individual studies of the satisfaction–performance link disagree, so the disciplined way to read the literature is through its meta-analyses, which also supply the magnitude against which any headline claim should be judged.

Szymanski and Henard (2001) provided the first synthesis. Two decades later Otto, Szymanski, and Varadarajan (2020) revisited the question across more than a quarter century of empirical research.4 Mittal et al. (2023) went wider still, synthesizing 535 correlations from 245 articles with a combined sample exceeding \(1.1\) million, and found positive associations of satisfaction with customer-level outcomes (retention, word of mouth, spending, price) and with firm-level outcomes across product-market, accounting, and financial-market performance—while emphasizing that the magnitude varies substantially with context and measurement. The most recent statement, Mittal, Singh, and Malshe (2026), replicates and extends this evidence on more than 285 million ratings across 1,126 brands and again reports robust association with mindset metrics, accounting outcomes, and stock-market performance. The parallel synthesis on the marketing side of the interface is Edeling and Fischer (2016), with the broader review in Edeling, Srinivasan, and Hanssens (2021).

Read together, these give the honest position. The association between satisfaction and firm performance is real, positive, replicated across enormous samples, and—crucially—heavily conditional: it moves with the setting, the satisfaction measure, and the performance measure chosen. That is a defensible foundation for treating satisfaction as a strategic asset. It is a poor foundation for a trading strategy, and the distinction is where Section 4.6.1 went wrong. An association whose sign is dependable but whose magnitude depends on measurement choices is not the profile of a stable, exploitable mispricing; it is the profile of a construct that matters and resists being reduced to one number.

Two cautions for the reader. First, these syntheses are not fully independent of the primary literature they assess—several share authors with prominent studies in the underlying corpus—which is ordinary in a specialized field but worth knowing. Second, a meta-analytic mean is an average over heterogeneous designs; the moderators matter more than any pooled estimate, which is precisely what Mittal et al. (2023) emphasize.

4.7 Two Disciplining Findings

A reader could leave the preceding sections believing that satisfaction is an unambiguous good to be maximized without limit. Two findings discipline that view, and Table 4.3 contrasts each with the naive expectation it overturns.

4.7.1 The Market-Share–Satisfaction Tension

Practitioners often assume that market share and customer satisfaction move together—that a firm winning the market must be pleasing its customers. The data say otherwise. Re-examining the relationship over an extended horizon in a representative sample of U.S. consumer markets, Rego, Morgan, and Fornell (2013) find a consistently significant negative relationship: satisfaction is a weak predictor of future market share, while market share is a strong negative predictor of future satisfaction. The mechanism is preference heterogeneity. A brand that serves a large share of the market necessarily serves a more diverse set of customers, whose heterogeneous preferences cannot all be satisfied by a single offering, so average satisfaction falls as share rises. The negative relationship is mediated by this heterogeneity, and it attenuates under two conditions—when satisfaction is benchmarked against the firm’s nearest competitor rather than an industry average, and when brand switching costs are low. The strategic resolution is a diversified brand portfolio: marketing a broader range of brands lets a firm cover heterogeneous segments without forcing one offering to satisfy them all, easing the inherent trade-off between share and satisfaction.

4.7.2 Financial Leverage as a Moderator

The satisfaction–value link is also conditioned by the firm’s capital structure. Malshe and Agarwal (2015) show that financial leverage shapes marketing outcomes and firm value through two distinct effects. First, leverage lowers customer satisfaction, an effect mediated by advertising intensity—debt service crowds out the marketing investment that sustains satisfaction. Second, leverage moderates the satisfaction–firm-value relationship itself. The damage is not uniform: service firms and firms in competitive industries suffer more from financial leverage, because their value depends more heavily on the customer relationships that leverage erodes. The implication is that satisfaction cannot be managed in isolation from the balance sheet; a leveraged firm extracts less value from a given level of satisfaction and finds it harder to sustain that level in the first place.

Table 4.3: Findings that discipline a naive “maximize satisfaction” view
Finding Naive expectation Empirical reality Mechanism
Market share vs. satisfaction Positively related Negative; share predicts lower future satisfaction Preference heterogeneity (Rego, Morgan, and Fornell 2013)
Pricing of satisfaction news Symmetric Dissatisfaction priced ~2.4x as heavily Loss aversion; short-interest channel (Malshe, Colicev, and Mittal 2020)
Financial leverage Neutral to value link Lowers satisfaction and weakens its value effect Advertising crowd-out; relationship erosion (Malshe and Agarwal 2015)

4.8 Satisfaction in Monopoly: The Cost Channel

The marketing–finance evidence above is drawn almost entirely from competitive markets, where satisfaction plausibly works through retention and repurchase—a customer who can leave but chooses to stay. This raises a sharp test: does satisfaction matter where customers cannot leave? Regulated utilities supply the natural experiment, both because they are monopolies and because their stringent reporting yields unusually precise operating and accounting data (Bhattacharya, Morgan, and Rego 2020).

Studying U.S. public utility firms, Bhattacharya, Morgan, and Rego (2021) find that customer satisfaction has no detectable effect on future revenues—unsurprising, since captive customers cannot redirect their demand—yet a positive effect on future profitability. The reconciliation is a cost channel. Higher satisfaction substantially reduces operating costs, specifically the costs of distribution, customer service, and general administration. Two post-hoc mechanisms support this. First, satisfaction lowers the cost of direct customer interactions and improves employee engagement by relieving the burden of managing dissatisfied customers. Second, a satisfied customer base is associated with greater trust and cooperation, which lubricates the firm’s regulated operations. The monopoly case is therefore not a null result but a clarifying one: it strips away the revenue channel and reveals a cost channel that operates even when demand is fixed, completing the account of why satisfaction is valuable.

4.9 Communicating Uncertainty: A Lever on Satisfaction

The chapter has treated satisfaction as a consequence of performance relative to expectations. The expectancy–disconfirmation logic of Equation 4.1 implies a managerial corollary: because satisfaction depends on the expectation a firm sets, the way a firm communicates can move satisfaction without changing performance at all. Time predictions under uncertainty—a delivery window, a wait time, a service estimate—are a clean setting because the firm chooses how to frame an inherently uncertain quantity.

Hu, Gaertig, and Dietvorst (2024) show that customers prefer range estimates (for example, “40–50 minutes”) to point estimates (“45 minutes”) for time predictions in uncertain contexts, and that communicating uncertainty through ranges enhances satisfaction with the estimate. The preference is robust across domains, time durations, and outcome distributions, and it holds even when the point estimate is conservative—that is, when the single number is set to the cautious upper-bound value a firm might otherwise use to manage expectations. In the terms of Equation 4.1, a range sets an expectation \(E\) that the realized performance is more likely to confirm, reducing the chance of a negative disconfirmation \(D < 0\) and thereby protecting satisfaction.5 The result is a concrete instance of the chapter’s central theme: satisfaction is managed not only by improving \(P\) but by shaping \(E\), and the framing of uncertainty is one of the cheapest available levers on the gap between them.

4.10 Connections

Satisfaction does not stand alone among marketing’s intangible assets. Its relationship to brand equity is close: perceived quality is an input to satisfaction and a component of brand value, and the attribute-level dynamics of satisfaction—how warranty-resolvable versus irresolvable attributes decay at different rates and carry different weight over time (Slotegraaf and Inman 2004)—are developed alongside the quality-signaling material in Chapter 11. The asset-pricing and risk results invoked above rest on the event-study and risk-decomposition machinery formalized in Chapter 25.

4.11 Key Takeaways

Aksoy, Lerzan, Bruce Cooil, Christopher Groening, Timothy L. Keiningham, and Atakan Yalçın. 2008. “The Long-Term Stock Market Valuation of Customer Satisfaction.” Journal of Marketing 72 (4): 105–22. https://doi.org/10.1509/jmkg.72.4.105.
Bell, David R., Olivier Ledoit, and Michael Wolf. 2014. “A New Portfolio Formation Approach to Mispricing of Marketing Performance Indicators: An Application to Customer Satisfaction.” Customer Needs and Solutions 1 (4): 263–76. https://doi.org/10.1007/s40547-014-0028-6.
Bharadwaj, Sundar G., and Debanjan Mitra. 2016. “Satisfaction (Mis)pricing Revisited: Real? Really Big?” Journal of Marketing 80 (5): 116–21. https://doi.org/10.1509/jm.16.0236.
Bhattacharya, Abhi, Neil A. Morgan, and Lopo L. Rego. 2020. “Customer Satisfaction and Firm Profits in Monopolies: A Study of Utilities.” Journal of Marketing Research 58 (1): 202–22. https://doi.org/10.1177/0022243720962405.
Bhattacharya, Abhi, Neil A Morgan, and Lopo L Rego. 2021. “Customer Satisfaction and Firm Profits in Monopolies: A Study of Utilities.” Journal of Marketing Research 58 (1): 202–22.
Botti, Simona, and Ann L. McGill. 2011. “The Locus of Choice: Personal Causality and Satisfaction with Hedonic and Utilitarian Decisions.” Journal of Consumer Research 37 (6): 1065–78. https://doi.org/10.1086/656570.
Edeling, Alexander, and Marc Fischer. 2016. “Marketing’s Impact on Firm Value: Generalizations from a Meta-Analysis.” Journal of Marketing Research 53 (4): 515–34. https://doi.org/10.1509/jmr.14.0046.
Edeling, Alexander, Shuba Srinivasan, and Dominique M. Hanssens. 2021. “The Marketingfinance Interface: A New Integrative Review of Metrics, Methods, and Findings and an Agenda for Future Research.” International Journal of Research in Marketing 38 (4): 857–76. https://doi.org/10.1016/j.ijresmar.2020.09.005.
Fornell, Claes, Michael D. Johnson, Eugene W. Anderson, Jaesung Cha, and Barbara Everitt Bryant. 1996. “The American Customer Satisfaction Index: Nature, Purpose, and Findings.” Journal of Marketing 60 (4): 7–18. https://doi.org/10.1177/002224299606000403.
Fornell, Claes, Sunil Mithas, and Forrest V. Morgeson. 2009. “Commentary—The Economic and Statistical Significance of Stock Returns on Customer Satisfaction.” Marketing Science 28 (5): 820–25. https://doi.org/10.1287/mksc.1090.0505.
Fornell, Claes, Sunil Mithas, Forrest V. Morgeson, and M. S. Krishnan. 2006. “Customer Satisfaction and Stock Prices: High Returns, Low Risk.” Journal of Marketing 70 (1): 3–14. https://doi.org/10.1509/jmkg.2006.70.1.3.
Fornell, Claes, Forrest V. Morgeson, and G. Tomas M. Hult. 2016a. “An Abnormally Abnormal Intangible: Stock Returns on Customer Satisfaction.” Journal of Marketing 80 (5): 122–25. https://doi.org/10.1509/jm.16.0248.
———. 2016b. “Stock Returns on Customer Satisfaction Do Beat the Market: Gauging the Effect of a Marketing Intangible.” Journal of Marketing 80 (5): 92–107. https://doi.org/10.1509/jm.15.0229.
Gruca, Thomas S, and Lopo L Rego. 2005. “Customer Satisfaction, Cash Flow, and Shareholder Value.” Journal of Marketing 69 (3): 115–30.
Guenther, Miriam, and Peter Guenther. 2021. “The Complex Firm Financial Effects of Customer Satisfaction Improvements.” International Journal of Research in Marketing 38 (3): 639–62. https://doi.org/10.1016/j.ijresmar.2020.10.003.
Higgins, E. Tory. 1997. “Beyond Pleasure and Pain.” American Psychologist 52 (12): 1280–300. https://doi.org/10.1037/0003-066x.52.12.1280.
Hu, Beidi, Celia Gaertig, and Berkeley J Dietvorst. 2024. “How Should Time Estimates Be Structured to Increase Customer Satisfaction?” Management Science.
Ittner, Christopher, David Larcker, and Daniel Taylor. 2009. “Commentary—The Stock Market’s Pricing of Customer Satisfaction.” Marketing Science 28 (5): 826–35. https://doi.org/10.1287/mksc.1090.0526.
Jacobson, Robert, and Natalie Mizik. 2009a. “RejoinderCustomer Satisfaction-Based Mispricing: Issues and Misconceptions.” Marketing Science 28 (5): 836–45. https://doi.org/10.1287/mksc.1090.0531.
———. 2009b. “The Financial Markets and Customer Satisfaction: Reexamining Possible Financial Market Mispricing of Customer Satisfaction.” Marketing Science 28 (5): 810–19. https://doi.org/10.1287/mksc.1090.0495.
Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory: An Analysis of Decision Under Risk.” Econometrica 47 (2): 263–91. https://doi.org/10.2307/1914185.
Keiningham, Timothy L., Lerzan Aksoy, Bruce Cooil, Tor Wallin Andreassen, and Luke Williams. 2008. “A Holistic Examination of Net Promoter.” Journal of Database Marketing & Customer Strategy Management 15 (2): 79–90. https://doi.org/10.1057/dbm.2008.4.
Keiningham, Timothy L., Bruce Cooil, Tor Wallin Andreassen, and Lerzan Aksoy. 2007a. “A Longitudinal Examination of Net Promoter and Firm Revenue Growth.” Journal of Marketing 71 (3): 39–51. https://doi.org/10.1509/jmkg.71.3.039.
———. 2007b. “A Longitudinal Examination of Net Promoter and Firm Revenue Growth.” Journal of Marketing 71 (3): 39–51. https://doi.org/10.1509/jmkg.71.3.039.
Kumar, Sachin, and Mikhail Zymbler. 2019. “A Machine Learning Approach to Analyze Customer Satisfaction from Airline Tweets.” Journal of Big Data 6 (1). https://doi.org/10.1186/s40537-019-0224-1.
Kumar, V. 2016. “Introduction: Is Customer Satisfaction (Ir)relevant as a Metric?” Journal of Marketing 80 (5): 108–9. https://doi.org/10.1509/jm.80.5.1.
Malshe, Ashwin, and Manoj Agarwal. 2015. “From Finance to Marketing: Impact of Financial Leverage on Customer Satisfaction.” Journal of Marketing, June, 150626124337002. https://doi.org/10.1509/jim.15.0312.
Malshe, Ashwin, Anatoli Colicev, and Vikas Mittal. 2020. “How Main Street Drives Wall Street: Customer (Dis)satisfaction, Short Sellers, and Abnormal Returns.” Journal of Marketing Research 57 (6): 1055–75. https://doi.org/10.1177/0022243720954373.
Mittal, Vikas, Kyuhong Han, Carly Frennea, Markus Blut, Muzeeb Shaik, Narendra Bosukonda, and Shrihari Sridhar. 2023. “Customer Satisfaction, Loyalty Behaviors, and Firm Financial Performance: What 40 Years of Research Tells Us.” Marketing Letters 34 (2): 171–87. https://doi.org/10.1007/s11002-023-09671-w.
Mittal, Vikas, William T. Ross, and Patrick M. Baldasare. 1998. “The Asymmetric Impact of Negative and Positive Attribute-Level Performance on Overall Satisfaction and Repurchase Intentions.” Journal of Marketing 62 (1): 33–47. https://doi.org/10.1177/002224299806200104.
Mittal, Vikas, Sonam Singh, and Ashwin Malshe. 2026. “Customer Satisfaction: A Multi-Level Framework.” Journal of the Academy of Marketing Science, July. https://doi.org/10.1007/s11747-026-01175-z.
Morgan, Neil A, and Lopo Leotte Rego. 2006. “The Value of Different Customer Satisfaction and Loyalty Metrics in Predicting Business Performance.” Marketing Science 25 (5): 426–39.
O’Sullivan, Don, Mark C. Hutchinson, and Vincent O’Connell. 2009. “Empirical Evidence of the Stock Market’s (Mis)pricing of Customer Satisfaction.” International Journal of Research in Marketing 26 (2): 154–61. https://doi.org/10.1016/j.ijresmar.2008.08.004.
Oliver, Richard L, and Raymond R Burke. 1999. “Expectation Processes in Satisfaction Formation: A Field Study.” Journal of Service Research 1 (3): 196–214.
Otto, Ashley S., David M. Szymanski, and Rajan Varadarajan. 2020. “Customer Satisfaction and Firm Performance: Insights from over a Quarter Century of Empirical Research.” Journal of the Academy of Marketing Science 48 (3): 543–64. https://doi.org/10.1007/s11747-019-00657-7.
Rego, Lopo L, Neil A Morgan, and Claes Fornell. 2013. “Reexamining the Market Share–Customer Satisfaction Relationship.” Journal of Marketing 77 (5): 1–20.
Slotegraaf, Rebecca J, and J Jeffrey Inman. 2004. “Longitudinal Shifts in the Drivers of Satisfaction with Product Quality: The Role of Attribute Resolvability.” Journal of Marketing Research 41 (3): 269–80.
Sorescu, Alina, and Sorin M. Sorescu. 2016. “Customer Satisfaction and Long-Term Stock Returns.” Journal of Marketing 80 (5): 110–15. https://doi.org/10.1509/jm.16.0214.
Szymanski, David M., and David H. Henard. 2001. “Customer Satisfaction: A Meta-Analysis of the Empirical Evidence.” Journal of the Academy of Marketing Science 29 (1): 16–35. https://doi.org/10.1177/0092070301291002.
Tuli, Kapil R., and Sundar G. Bharadwaj. 2009. “Customer Satisfaction and Stock Returns Risk.” Journal of Marketing 73 (6): 184–97. https://doi.org/10.1509/jmkg.73.6.184.
Tversky, A, and D Kahneman. 1981. “The Framing of Decisions and the Psychology of Choice.” Science 211 (4481): 453–58. https://doi.org/10.1126/science.7455683.
Wei, Qiang, Xiaowei Shi, Quan Li, and Guoqing Chen. 2020. “Hawaii International Conference on System Sciences.” In. Hawaii International Conference on System Sciences. https://doi.org/10.24251/hicss.2020.178.

  1. ETF Series Solutions, Form 485APOS (post-effective amendment to registration statement), filed 5 August 2016, SEC EDGAR accession 0000894189-16-010869. Quotations are from that filing. The fund has since changed advisers; current filings are under Tidal Trust I.↩︎

  2. ETF Series Solutions, Form 485APOS (post-effective amendment to registration statement), filed 5 August 2016, SEC EDGAR accession 0000894189-16-010869. Quotations are from that filing. The fund has since changed advisers; current filings are under Tidal Trust I.↩︎

  3. Read current figures from the fund’s own prospectus and annual reports rather than from any secondary source, including this one; performance statistics move, and the pedagogical point does not depend on the exact number.↩︎

  4. Specific pooled correlation values from this article circulate widely in secondary summaries. They are deliberately not reproduced here, because they could not be confirmed against the article itself during preparation of this edition—the publisher’s full text was not accessible. A reader who needs the magnitude should take it from the article, not from this book or from any other secondary source. The argument in this section does not depend on the precise value.↩︎

  5. A single conservative point estimate also lowers the chance of negative disconfirmation, which is why firms pad their quotes. The Hu, Gaertig, and Dietvorst (2024) result is that a range achieves this and is preferred to the padded point estimate, so it dominates the conventional conservatism heuristic rather than merely replicating it.↩︎