17 Influencer Marketing

Influencer marketing is the practice of leveraging an individual’s accumulated audience, credibility, and content to carry a firm’s marketing message. Formally, Leung et al. (2022) describe online influencer marketing as the deployment of an influencer’s resources—follower networks, personal positioning, communication content, and the trust followers place in them—to enhance the effectiveness of a firm’s marketing communication. The defining feature is that persuasion is rented from a third party whose authority was earned outside any single brand relationship, and whose continued authority depends on not appearing captured by the brands that pay them. That tension—between the sponsor’s desire for reach and control and the influencer’s need to preserve credibility—organizes nearly everything in this chapter.

Influencer marketing matters commercially because attention has migrated to social platforms and because a small set of accounts concentrates a disproportionate share of persuasive power. Industry audits routinely find that a handful of recommenders account for a large fraction of category recommendations—on the order of five percent of recommenders generating close to half of socially influenced purchases in apparel and footwear—so the selection of whom to hire dominates the economics of a campaign. It matters scientifically because influencers are a clean laboratory for classical questions in persuasion: source credibility, the endorsement of meaning, two-sided signaling, and the measurement of engagement and its translation (or failure to translate) into sales.

This chapter develops the topic from construct to model to measurement. We first locate the influencer within a communication-theoretic frame—sender, message, receiver—and define the constructs (the influencer as a person brand, credibility, engagement) formally. We then build the econometric workhorse of the empirical literature: a selection-corrected count model of engagement, with its estimator, assumptions, and identification requirements stated in full. We turn next to the economics of influencer type—why micro-influencers and celebrities play different equilibrium roles—and to the frontier problem that engagement is not sales, including deep-learning measures of product-focused attention. We close with the practical problem managers actually face: valuing an account or a single post, and a reproducible toolkit for doing so.

17.1 The Influencer as a Communication System

The cleanest way to organize the determinants of influencer effectiveness is the mathematical theory of communication (Weaver, Shannon, et al. 1963; Shannon 1948), which decomposes any act of communication into a sender, a message, and a receiver. Leung et al. (2022) map the three components onto the influencer setting and attach a measurable construct to each: the sender (the influencer) contributes source trustworthiness (Wilson and Sherrell 1993); the message (the post) contributes content value (Ducoffe 1996); and the receiver (the follower) contributes involvement with the message. Effectiveness is multiplicative rather than additive in these components—a trustworthy source delivering low-value content to an uninvolved audience persuades no one—which is why single-metric rankings of influencers (by follower count alone, say) predict outcomes so poorly. Figure 17.1 shows how the firm intervenes on this channel through selection and the brief while observing only downstream engagement.

flowchart LR
  F["Firm<br/>(selection + brief)"] --> S
  subgraph channel["Communication channel"]
    S["Sender<br/>influencer<br/><i>trustworthiness</i>"] --> M["Message<br/>post<br/><i>content value</i>"]
    M --> R["Receiver<br/>follower<br/><i>involvement</i>"]
  end
  R --> E["Engagement<br/>(likes, comments,<br/>shares)"]
  E -.->|noisy proxy| Sales["Sales lift"]
  R -.->|reputation feedback| S

Figure 17.1: The influencer as a communication system. Effectiveness is jointly determined by source trustworthiness, message content value, and receiver involvement; the firm intervenes through selection (whom to hire) and the brief (what to post), and observes only downstream engagement.

17.1.1 The Influencer as a Person Brand

An influencer is best understood as a person brand: a human being who functions as an economic asset because their name evokes a stable, ownable set of associations (Thomson 2006; Fournier and Eckhardt 2019; Kerrigan et al. 2011). This framing inherits the full apparatus of Chapter 11—brand associations, personality, and relationship quality—but with one decisive difference from a traditional celebrity. Classical celebrity endorsement works through meaning transfer: the celebrity, credentialed by an institution (a studio, a league, an awards body), carries culturally constituted meanings that flow through the advertisement into the brand and onward to the consumer (McCracken 1989). Social-media influencers are not institutionally credentialed in this way (McQuarrie, Miller, and Phillips 2013); their authority is conferred bottom-up by an audience and is therefore both cheaper to acquire and easier to lose. The influencer’s persuasive capital is endogenous to their own posting behavior in a way the movie star’s is not, which is the root of the reputation dynamics developed in Section 17.5.

Influencer (person brand). An individual whose accumulated audience, perceived expertise, and earned trust constitute a transferable marketing asset, and whose endorsement derives credibility from an audience relationship rather than from institutional credentialing.

17.1.2 Theoretical Foundations

The drivers and models that follow are anchored in five theories of persuasion and social structure, and naming them clarifies why influencer effectiveness behaves as it does. The foundational account is source credibility and attractiveness: a message persuades partly through who delivers it, with the source’s perceived expertise and trustworthiness driving credibility (Hovland and colleagues’ communication-and-persuasion program, a monograph cited here by name) and attractiveness and likability supplying a peripheral route to influence. This is why source trustworthiness (Wilson and Sherrell 1993) is the deep construct beneath the sender-side drivers, and why signals of independence raise perceived influence (Sokolova, Seenivasan, and Thomas 2020). The match-up or congruence hypothesis refines this: an endorser is most effective when their image fits the endorsed product, so that a fitness influencer credibly carries athletic apparel but not, say, financial services (the match-up literature, an edited-volume and journal tradition cited by name). Congruence is the theoretical reason follower-brand fit appears among the drivers, and the reason its effect is inverted-U, because perfect fit can read as paid placement.

Two further theories explain effects unique to the social-media setting. Parasocial interaction holds that audiences form one-sided, intimacy-like relationships with media personae, experiencing a creator as a friend despite the absence of reciprocity (Horton and Wohl’s parasocial-interaction concept, cited by name); the strength of this illusory relationship is what converts a follower’s attention into trust and what sponsorship can erode (Section 17.5). Homophily, the principle that ties form between similar people and that similarity raises influence (the homophily literature, cited by name), explains why audience-creator similarity predicts engagement and why micro-influencers, who resemble their followers, can out-persuade larger but more distant celebrities. Finally, signaling theory (Spence 1973) frames the whole exchange as one of asymmetric information: under uncertainty about product quality, a costly, hard-to-fake endorsement signals quality, which is precisely the screening role that makes a micro-influencer’s refusal to endorse low-quality products informative (Nistor and Selove (2024)). Meaning transfer (McCracken 1989) is the classical-celebrity counterpart of this signaling logic, distinguished by its reliance on institutional rather than audience-conferred authority.

17.1.3 Engagement, Formally

The dependent variable throughout the empirical literature is engagement. Following Hollebeek (2011), engagement is “a customer’s cognitive, emotional, and behavioral activities” with respect to brand-related interactions—a deliberately multidimensional construct. In practice it is operationalized through observable behavioral traces: comments, likes, shares, saves, and watch time. These traces are counts, bounded below at zero and right-skewed, a fact that dictates the estimator (Section 17.3).

A useful summary of effectiveness is the engagement elasticity, the percentage change in consumer engagement produced by a one-percent change in influencer marketing spend. Leung et al. (2022) estimate this elasticity at roughly $0.46$: a one-percent increase in spend raises engagement by about $0.457$ percent. Letting $g$ denote engagement and $s$ denote spend, the elasticity is

\[ \varepsilon_{g,s} \;=\; \frac{\partial \log g}{\partial \log s} \;=\; \frac{\partial g}{\partial s}\,\frac{s}{g}, \tag{17.1}\]

a unit-free quantity comparable across influencers, platforms, and campaigns in a way that raw engagement counts are not. Because Equation 17.1 is below one, engagement returns to spend are inelastic—doubling the budget less than doubles engagement—an empirical regularity consistent with diminishing returns to reach.

17.2 Drivers of Influencer Effectiveness

What makes an influencer campaign work? The literature converges on a set of drivers attached to the three communication components, several of which act non-monotonically or interact with campaign intent and platform. We organize them by component and then treat the most important interactions explicitly.

Sender (influencer) characteristics. Three sender attributes raise effectiveness: originality (distinctive content that does not read as a template), follower size (reach), and the moderating role of sponsor salience (Leung et al. 2022). Source trustworthiness is the deeper construct underneath these (Wilson and Sherrell 1993): influencers persuade to the extent that their endorsement is read as a sincere recommendation rather than a paid placement. A counterintuitive corollary is that signals of independence can raise perceived influence—an influencer who follows relatively few accounts while commanding many followers is read as autonomous and therefore more credible (Sokolova, Seenivasan, and Thomas 2020). Several drivers, however, are inverted-U shaped: influencer activity (posting frequency), follower–brand fit, and content positivity each help up to a point and then hurt, as over-posting reads as spam, perfect fit reads as paid, and relentless positivity reads as inauthentic (Leung et al. 2022).

Message (content) characteristics. Content value is the message-level analogue of source credibility (Ducoffe 1996). A practically important content distinction is between product-launch posts and ordinary endorsement posts: product-launch content decreases effectiveness (Leung et al. 2022), plausibly because launch posts foreground the commercial transaction and suppress the audience-relationship cues that make influencer content persuasive. Identifying which posts are launch- or promotion-related at scale is itself a measurement task; the literature uses dictionary-based and supervised text classification on post text to code launch versus non-launch and product versus service (Lovett, Peres, and Shachar 2013).

Receiver (audience) characteristics. Effectiveness ultimately depends on follower involvement—how cognitively and affectively engaged the audience is with the message. Involvement is not directly controllable by the firm, but it is shaped by the match between content and audience and by platform affordances, which is why the same post performs differently across platforms.

17.2.1 Sponsored Blogging: Expertise, Intent, and Platform

A foundational empirical study of these drivers is Hughes, Swaminathan, and Brooks (2019), who study sponsored blogging—content that is simultaneously paid (the firm compensates the blogger) and earned (it appears as the blogger’s own editorial voice), placing it at the intersection of paid and earned media (Colicev et al. 2018; Colicev, Kumar, and O’Connor 2019). Grounding their hypotheses in the elaboration likelihood model, which distinguishes central (argument-driven) from peripheral (cue-driven) routes to persuasion, the authors show that the same blogger characteristic can help or hurt depending on the campaign’s position in the purchase funnel and the platform on which the content appears.

Two moderators are central. Campaign intent is whether the goal is to raise awareness (top of funnel) or to encourage trial (lower funnel). Platform involvement contrasts a high-involvement environment (a blog, where readers elaborate) with a low-involvement environment (a Facebook feed, where they skim). Table 17.1 summarizes the key contingencies.

Table 17.1: How blogger and content characteristics interact with campaign intent and platform. Source: Hughes, Swaminathan, and Brooks (2019).

Driver	Awareness campaign	Trial campaign	Platform contingency
Blogger expertise	More effective	Less effective	On Facebook (low involvement) source expertise has no effect on engagement
Hedonic content	Less effective	More effective	On Facebook, hedonic content lifts engagement, especially for trial
Campaign freebies	Mixed	Mixed	Giveaways can backfire, cannibalizing blog engagement in favor of the Facebook channel

The headline interaction is between expertise and intent: when the goal is to build awareness, high blogger expertise is more effective, because an expert source lends credibility to an unfamiliar product; when the goal is to drive trial, expertise matters less and experiential, hedonic content matters more, because the audience already knows the product and needs a reason to act. Critically, this expertise effect is a high-involvement phenomenon: on Facebook, where readers do not elaborate, source expertise washes out entirely. The managerial implication is that influencer selection and the creative brief must be chosen jointly with the funnel objective and the platform, not optimized in isolation.

17.3 Modeling Engagement with Selection

We now formalize the workhorse model behind studies such as Hughes, Swaminathan, and Brooks (2019) and Leung et al. (2022). Two econometric facts dominate the specification. First, engagement is a count with overdispersion—the conditional variance exceeds the conditional mean—so ordinary least squares and even the Poisson model are inappropriate. Second, the influencers and posts in any dataset are not randomly assigned: brands choose which influencers to hire and influencers choose which campaigns to accept, so the observed sample is selected on unobservables that may also drive engagement. Ignoring this yields a selection bias that contaminates every coefficient.

17.3.1 The Negative Binomial Outcome Model

Let $y_i$ be the engagement count for post $i$ (e.g., number of comments or likes), and let $\mathbf{x}_i$ collect influencer, content, campaign, and platform covariates. The negative binomial model specifies a log-linear conditional mean,

\[ \mathbb{E}[y_i \mid \mathbf{x}_i] = \mu_i = \exp(\mathbf{x}_i^\top \boldsymbol{\beta}), \tag{17.2}\]

with the variance inflated relative to the mean by an overdispersion parameter $\alpha > 0$,

\[ \operatorname{Var}(y_i \mid \mathbf{x}_i) = \mu_i + \alpha\,\mu_i^2 . \tag{17.3}\]

The model arises as a Poisson–gamma mixture: conditional on an unobserved multiplicative heterogeneity term $\nu_i \sim \text{Gamma}(1/\alpha, 1/\alpha)$, $y_i \mid \nu_i \sim \text{Poisson}(\mu_i \nu_i)$, and integrating out $\nu_i$ yields the negative binomial likelihood (Cameron and Trivedi 2005). The parameter $\alpha$ measures the residual variance left by influencer- and post-level factors the covariates do not capture; as $\alpha \to 0$ the model collapses to Poisson. The coefficients $\boldsymbol{\beta}$ are estimated by maximum likelihood and interpreted as semi-elasticities: a unit change in covariate $k$ multiplies expected engagement by $\exp(\beta_k)$, holding the rest fixed.

17.3.2 Correcting for Endogenous Selection

The threat to identification is that the set of (influencer, campaign) pairs we observe is the outcome of a choice. Brands pursue influencers they expect to perform well; influencers accept campaigns that fit their persona. If the unobservables driving selection are correlated with the unobservables driving engagement, the estimated $\boldsymbol{\beta}$ in Equation 17.2 confounds the causal effect of a covariate with its correlation with the selection process.

The standard remedy is a Heckman-type selection model (Heckman and Vytlacil 2007; Greene 2003). Two equations are specified. A selection equation models whether an influencer is chosen for a campaign,

\[ d_i = \mathbf{1}\!\left\{ \mathbf{z}_i^\top \boldsymbol{\gamma} + u_i > 0 \right\}, \tag{17.4}\]

and an outcome equation models engagement conditional on selection (Equation 17.2). Identification hinges on an exclusion restriction: the selection covariates $\mathbf{z}_i$ must contain at least one variable that affects selection but does not directly affect engagement once selection is accounted for. Concretely, Hughes, Swaminathan, and Brooks (2019) construct a blogger psychographic index—travel/foodie, persona, lifestyle, and values factors extracted by varimax-rotated factor analysis of profile descriptions—as instruments: these traits plausibly drive which bloggers a brand recruits without directly moving the engagement a post generates. Leung et al. (2022) use an analogous strategy, instrumenting selection with the characteristics of the second-most-similar influencer the focal brand could have chosen, following the matching logic of Hughes, Swaminathan, and Brooks (2019).

The exclusion restriction is precisely where identification can break. If the purportedly excluded variable does influence engagement directly—if, say, a blogger’s “lifestyle” loading also makes their posts more engaging—the instrument is invalid, the correction is contaminated, and the corrected estimates can be more biased than the naive ones. There is no purely statistical test of the restriction; it must be defended on substantive grounds. A credible influencer study therefore states its exclusion restriction explicitly and argues why the excluded variables touch selection but not the outcome.

17.3.3 A Reproducible Selection-Corrected Count Model

The following simulation builds a dataset with genuine selection on unobservables, shows the bias from naively fitting the outcome model on the selected sample, and recovers the truth with a control-function correction. It depends only on base R.

Code

set.seed(15)

n <- 4000

# --- Latent traits -------------------------------------------------------
# Exclusion-restriction variable: psychographic "lifestyle" index z.
# It drives SELECTION but NOT engagement directly.
z          <- rnorm(n)
expertise  <- rnorm(n)               # influencer expertise (in both equations)
hedonic    <- rnorm(n)               # content hedonic value (outcome only)

# Correlated unobservables linking selection and engagement.
rho   <- 0.6
e_sel <- rnorm(n)
e_out <- rho * e_sel + sqrt(1 - rho^2) * rnorm(n)

# --- Selection equation (who gets hired) ---------------------------------
sel_index <- -0.2 + 0.9 * z + 0.5 * expertise + e_sel
selected  <- as.integer(sel_index > 0)

# --- Outcome: engagement counts (negative binomial) ----------------------
beta0 <- 1.0; b_exp <- 0.40; b_hed <- 0.30
log_mu <- beta0 + b_exp * expertise + b_hed * hedonic + 0.8 * e_out
mu     <- exp(log_mu)
theta  <- 2                                  # NB size (overdispersion)
y      <- rnbinom(n, size = theta, mu = mu)

dat <- data.frame(y, z, expertise, hedonic, selected, e_out)
obs <- dat[dat$selected == 1, ]              # we only observe hired influencers

# --- Naive NB on the selected sample (biased) ----------------------------
naive <- MASS::glm.nb(y ~ expertise + hedonic, data = obs)

# --- Control-function correction -----------------------------------------
# Stage 1: probit selection on z + expertise; form the inverse Mills ratio.
sel_fit <- glm(selected ~ z + expertise, family = binomial("probit"), data = dat)
lp      <- predict(sel_fit, newdata = obs)
imr     <- dnorm(lp) / pnorm(lp)             # inverse Mills ratio for selected

# Stage 2: NB outcome including the IMR as a control function.
corr <- MASS::glm.nb(y ~ expertise + hedonic + imr, data = obs)

round(rbind(
  truth     = c(expertise = b_exp, hedonic = b_hed),
  naive     = coef(naive)[c("expertise", "hedonic")],
  corrected = coef(corr)[c("expertise", "hedonic")]
), 3)
#>           expertise hedonic
#> truth         0.400   0.300
#> naive         0.321   0.317
#> corrected     0.438   0.309

The naive estimates inflate the expertise coefficient because hired influencers are positively selected on the very unobservables that also drive engagement; adding the inverse Mills ratio as a control function absorbs that correlation and returns the expertise coefficient close to its true value of $0.40$. The hedonic coefficient, which enters only the outcome and is uncorrelated with selection, is recovered well by both methods—an illustration that selection bias contaminates selected-on covariates in particular.

17.4 Influencer Type and the Economics of Trust

Not all influencers play the same economic role, and the difference is an equilibrium phenomenon rather than a matter of scale alone. Nistor and Selove (2024) model a market with celebrity influencers (very large reach) and micro-influencers (smaller, more engaged audiences) who differ in how endorsement affects their future standing. The model yields a sharp separation. Celebrity influencers, whose value to followers is reach and awareness rather than curation, endorse all products to maximize the number of brand deals; their endorsements drive visibility but convey little reliable information about quality. Micro-influencers instead endorse only high-quality products, because their value proposition to followers is curation—acting as a quality screener so followers need not sift through comments to assess a product. In equilibrium, micro-influencers raise the information quality available to their audience while celebrities raise awareness; the two are complements, not substitutes, and a brand’s optimal influencer mix depends on whether it needs reach or credibility.

This screening role rationalizes a pricing puzzle: per follower, micro-influencers often command higher rates than their reach alone would justify, because part of what the brand buys is the influencer’s refusal to endorse low-quality products, which is exactly what makes the endorsement informative.

17.4.1 Selecting Influencers Under a Reach–Cost Trade-off

The practical selection problem is to trade reach against cost. Tian, Dew, and Iyengar (2024) formalize this with the follower elasticity of impressions (FEI)—the percentage change in impressions delivered per percentage change in follower count—and use it to quantify the marginal value of a larger audience against its marginal price. Because impressions scale sub-proportionally with followers (mega-influencers reach a smaller share of their audience), FEI typically falls with size, and the cost-effective hire is often a mid-tier influencer rather than the largest one available. The general lesson echoes Equation 17.1: the right object is an elasticity, normalizing the outcome by the input, not the raw level of reach.

17.5 Reputation Burning: The Cost of Sponsorship to the Influencer

Because an influencer’s authority is conferred by their audience rather than by an institution (Section 17.2), sponsorship is not free to the influencer: visibly accepting payment can erode the trust that made them valuable. Cheng and Zhang (2024) provide direct empirical evidence of this reputation-burning effect. Posting a sponsored video reduces an influencer’s reputation—measured by subscriber count—by about $0.19\%$ relative to comparable organic videos. The effect is larger for influencers with bigger audiences, and the gap in audience response (likes and comments) between sponsored and organic content widens with following size, consistent with larger audiences holding their influencers to a higher standard of editorial independence.

The loss is not a fixed tax; it is contingent, and the contingencies are managerially actionable. Reputation damage is mitigated when the sponsored content aligns with the influencer’s typical style—so the endorsement does not read as a break in character—and when the promoted brand is less well known, because endorsing an obscure brand looks more like discovery than like cashing in. This is the micro-influencer’s screening logic (Nistor and Selove (2024)) viewed from the supply side: the influencer’s optimal acceptance policy weighs the per-deal fee against the content-fit-dependent reputation cost.

Identifying this effect requires comparing sponsored and organic videos that are otherwise alike, which they are not by default—influencers choose which videos to monetize. Cheng and Zhang (2024) construct matched treatment (sponsored) and control (organic) groups at the influencer–video level using DiNardo–Fortin–Lemieux reweighting, which reweights the control distribution of observed covariates to match the treated distribution, so that any residual difference is attributable to sponsorship rather than to composition. The design is a useful template for any setting where treatment is a non-random managerial choice.

17.6 Engagement Is Not Sales: Product-Focused Attention

The single most consequential critique of influencer marketing is that engagement need not translate into sales. A video can be wildly engaging while directing none of that attention to the product—the viewer remembers the creator, not the brand. Two strands of recent work confront this gap by opening up the unstructured content of the video itself.

Rajaram and Manchanda (2020) build an interpretable deep-learning framework that decomposes a YouTube influencer video into its textual (“what is said”), audio, and visual streams and relates each to viewer engagement. Their central finding is that textual content matters more than audio or images for engagement, and they separate drivers of shallow from deep engagement using a dual-process view of viewer cognition. The contribution is methodological as much as substantive: the framework predicts engagement and attributes it to interpretable content features, which is what makes it useful to a manager writing a brief.

Yang, Zhang, and Zhang (2021) attack the engagement-versus-sales problem head-on by constructing a Product Engagement score (PE-score) that measures how much of a video’s engaging moments are actually about the product. The construction proceeds in three stages. First, a deep three-dimensional convolutional neural network is trained to produce an engagement heatmap assigning each pixel of each frame a weight reflecting its contribution to viewer engagement; critically, the network is trained on engagement residuals after partialling out influencer fixed effects, product fixed effects, acoustic features, and transcript embeddings, so it learns visual drivers of engagement rather than confounds. Second, an object-detection pass (SIFT-based) yields a product heatmap locating the product in each frame. Third, the PE-score combines the two: it is high when the engaging regions of the video coincide with the product’s location.

Formally, let $E_{ft}(p)$ be the engagement weight and $P_{ft}(p)$ the product indicator for pixel $p$ in frame $f$ of video $t$. The frame-level product engagement is the overlap

\[ \text{PE}_{ft} \;=\; \sum_{p} E_{ft}(p)\, P_{ft}(p), \tag{17.5}\]

aggregated over frames to a video-level PE-score. The measure requires only the ad and a product image, making it usable for pre-release testing.

Validated against TikTok influencer video ads matched to Taobao product sales (May–November 2019, 2,685 ads with linked sales), the findings are striking. Traditional engagement metrics (likes, comments, shares) show no significant correlation with sales, confirming the central critique; the PE-score, by contrast, significantly predicts sales lift, identified through a difference-in-differences design around video-posting time that uses a product search index to address reverse causality. Neither product-placement intensity (mere presence) nor influencer popularity nor price predicts sales—what matters is whether engagement is aimed at the product. An additional dataset shows influencers’ PE-scores are higher when promoting their own products, evidence that incentive alignment shapes how product-focused the content is. Reassuringly for the method’s validity, the neural saliency maps correlate highly with actual gaze maps from eye-tracking.

These results sit alongside a broader move to mine engagement and its drivers from unstructured social data (Yang, Zhang, and Zhang 2025; Lovett, Peres, and Shachar 2013), and they carry a blunt practical lesson: a brand paying for engagement should contract on product-focused engagement, because general engagement is, empirically, uncorrelated with the outcome the brand actually wants.

17.7 Valuing Influencer Accounts and Posts

The recurring managerial decision—what to pay an influencer, or what a single sponsored post is worth—is a valuation problem, and it is productive to import the discipline of financial valuation while respecting where social-media assets differ. The parallels are close enough to borrow frameworks and loose enough to demand caution: an influencer’s “cash flows” are sponsorship and affiliate revenues that are sporadic, algorithm-dependent, and contingent on an audience that can defect overnight.

Table 17.2 maps the three canonical financial approaches onto their social-media analogues.

Table 17.2: Financial valuation frameworks and their influencer-marketing analogues.

Finance method	Traditional use	Social-media analogue	Principal challenge
Discounted cash flow (DCF)	Discount projected firm cash flows to present value	Project and discount an account’s future sponsorship/affiliate income	Revenue is sporadic and algorithm-dependent; the discount rate must reflect platform volatility
Comparable multiples	Value via peers’ EBITDA/earnings multiples	Benchmark price-per-follower or price-per-engagement against peers in the niche	Metrics are not standardized across niches and platforms
Precedent transactions	Price from recent comparable acquisitions	Infer buyout value from comparable account/page sales	Private-account transaction data are scarce

17.7.1 Post-Level Multiples

At the post level, valuation reduces to a chosen unit price multiplied by an expected quantity. The common units are cost-per-thousand-impressions (CPM), cost-per-engagement (CPE), cost-per-view (CPV), and cost-per-click (CPC):

\[ \text{Value}_{\text{CPM}} = \frac{\text{Impressions}}{1000}\times \text{CPM}, \qquad \text{Value}_{\text{CPE}} = \text{Engagements}\times \text{CPE}. \tag{17.6}\]

A closely related construct, earned media value (EMV), monetizes organic reach by pricing it as if it had been bought, scaled by an adjustment factor for sentiment and fit, $\text{EMV} = \text{Impressions}\times \text{CPM}\times \phi$, where $\phi$ captures qualitative quality. EMV’s well-known weakness is its sensitivity to the analyst’s choice of $\phi$, which is why it should be reported alongside the raw inputs rather than as a single headline figure. Because no single multiple captures both reach (CPM) and interaction (CPE), practitioners often blend them and apply a qualitative adjustment for brand fit and sentiment, exactly as Yang, Zhang, and Zhang (2021) would counsel by emphasizing product-focused rather than gross engagement.

Code

followers       <- 200000
engagement_rate <- 0.03                       # 3% of followers engage
impressions     <- 0.75 * followers           # reach < followers
engagements     <- engagement_rate * followers

cpe <- 0.12; cpm <- 10                         # market unit prices
val_cpe <- engagements * cpe
val_cpm <- (impressions / 1000) * cpm

# Blend (weight CPE 40%, CPM 60%) then apply qualitative adjustment.
blend  <- 0.40 * val_cpe + 0.60 * val_cpm
final  <- blend * 1.20                          # +10% brand fit, +10% sentiment

data.frame(
  metric = c("CPE value", "CPM value", "Blended", "Final (adj.)"),
  value  = round(c(val_cpe, val_cpm, blend, final))
)
#>         metric value
#> 1    CPE value   720
#> 2    CPM value  1500
#> 3      Blended  1188
#> 4 Final (adj.)  1426

17.7.2 Account-Level Discounted Cash Flow

At the account level, the appropriate object is the present value of the influencer’s future brand-deal income, net of production and agency costs, plus a terminal value capturing the residual worth if the influencer pivots platforms or winds down. With net annual brand income $C_t$ growing at rate $g$, discount rate $r$, horizon $T$, and a terminal multiple or Gordon-growth tail, account value is

\[ V_0 \;=\; \sum_{t=1}^{T} \frac{C_t}{(1+r)^t} \;+\; \frac{\text{TV}}{(1+r)^T}, \qquad C_t = C_1 (1+g)^{t-1}. \tag{17.7}\]

The discount rate $r$ should exceed a typical corporate weighted-average cost of capital, because an influencer’s income is more exposed to platform-algorithm changes, audience fickleness, and competitive entry than a diversified firm’s. The following example values an account generating $120{,}000 in year-one net income, growing $5\%$ annually, discounted at $15\%$, with a terminal value set to twice final-year income.

Code

C1 <- 120000; g <- 0.05; r <- 0.15; T <- 5
t   <- 1:T
C   <- C1 * (1 + g)^(t - 1)                    # projected net income
pv  <- C / (1 + r)^t                            # discounted income
TV  <- 2 * C[T]                                 # terminal value (2x final year)
pv_TV <- TV / (1 + r)^T

account_value <- sum(pv) + pv_TV
data.frame(
  year = t, income = round(C), pv_income = round(pv)
)
#>   year income pv_income
#> 1    1 120000    104348
#> 2    2 126000     95274
#> 3    3 132300     86989
#> 4    4 138915     79425
#> 5    5 145861     72519
cat("PV of terminal value: $", format(round(pv_TV), big.mark = ","), "\n", sep = "")
#> PV of terminal value: $145,037
cat("Account value:        $", format(round(account_value), big.mark = ","), "\n", sep = "")
#> Account value:        $583,592

These valuations are estimates resting on assumptions about growth, persistence, and discounting, and they should be screened for the pathologies specific to social data: fake followers and bot engagement that inflate the inputs, market saturation that compresses multiples as supply grows, and abrupt algorithm changes that invalidate the projection. A defensible valuation states its assumptions, blends multiple methods (Table 17.2), and is refreshed as realized campaign outcomes arrive—much as the marketing–finance event-study machinery of Chapter 23 updates the market’s valuation of a firm’s brand assets as news arrives.

17.8 Key Takeaways

Influencer marketing is persuasion rented from a third party; its central tension is that the sponsorship which generates the firm’s reach also erodes the influencer’s audience-conferred credibility (Section 17.5).
Effectiveness is jointly determined by source trustworthiness, content value, and receiver involvement; several drivers (activity, fit, positivity) are inverted-U shaped, and key effects (expertise, hedonic content) interact with campaign intent and platform involvement (Section 17.2.1).
The empirical workhorse is a selection-corrected overdispersed count model: negative binomial for the engagement outcome (Equation 17.2), Heckman-type correction for non-random influencer selection, identified by an exclusion restriction that must be defended on substance, not statistics (Section 17.3).
Influencer type is an equilibrium role, not a size class: micro-influencers act as quality screeners and raise information quality, while celebrities raise awareness (Nistor and Selove (2024)); selection should trade reach against cost via an elasticity such as FEI (Tian, Dew, and Iyengar (2024)).
Engagement is not sales. Traditional engagement metrics do not predict sales lift; product-focused engagement, measured by the PE-score (Equation 17.5), does (Yang, Zhang, and Zhang 2021).
Account and post valuation borrows DCF, multiples, and precedent-transaction logic from finance, adjusted for platform volatility and screened for inflated metrics (Section 17.7).

Cameron, A Colin, and Pravin K Trivedi. 2005. Microeconometrics: Methods and Applications. Cambridge university press.

Cheng, Mengjie, and Shunyuan Zhang. 2024. “Reputation Burning: Analyzing the Impact of Brand Sponsorship on Social Influencers.” Management Science.

Colicev, Anatoli, Ashish Kumar, and Peter O’Connor. 2019. “Modeling the Relationship Between Firm and User Generated Content and the Stages of the Marketing Funnel.” International Journal of Research in Marketing 36 (1): 100–116. https://doi.org/10.1016/j.ijresmar.2018.09.005.

Colicev, Anatoli, Ashwin Malshe, Koen Pauwels, and Peter O’Connor. 2018. “Improving Consumer Mindset Metrics and Shareholder Value Through Social Media: The Different Roles of Owned and Earned Media.” Journal of Marketing 82 (1): 37–56. https://doi.org/10.1509/jm.16.0055.

Ducoffe, Robert H. 1996. “Advertising Value and Advertising on the Web-Blog@ Management.” Journal of Advertising Research 36 (5): 21–32.

Fournier, Susan, and Giana M. Eckhardt. 2019. “Putting the Person Back in Person-Brands: Understanding and Managing the Two-Bodied Brand.” Journal of Marketing Research 56 (4): 602–19. https://doi.org/10.1177/0022243719830654.

Greene, William H. 2003. Econometric Analysis. Pearson Education India.

Heckman, James J, and Edward J Vytlacil. 2007. “Econometric Evaluation of Social Programs, Part II: Using the Marginal Treatment Effect to Organize Alternative Econometric Estimators to Evaluate Social Programs, and to Forecast Their Effects in New Environments.” Handbook of Econometrics 6: 4875–5143.

Hollebeek, Linda. 2011. “Exploring Customer Brand Engagement: Definition and Themes.” Journal of Strategic Marketing 19 (7): 555–73. https://doi.org/10.1080/0965254x.2011.599493.

Hughes, Christian, Vanitha Swaminathan, and Gillian Brooks. 2019. “Driving Brand Engagement Through Online Social Influencers: An Empirical Investigation of Sponsored Blogging Campaigns.” Journal of Marketing 83 (5): 78–96. https://doi.org/10.1177/0022242919854374.

Kerrigan, Finola, Douglas Brownlie, Paul Hewer, and Claudia Daza-LeTouze. 2011. “‘Spinning’ Warhol: Celebrity Brand Theoretics and the Logic of the Celebrity Brand.” Journal of Marketing Management 27 (13-14): 1504–24. https://doi.org/10.1080/0267257x.2011.624536.

Leung, Fine F., Flora F. Gu, Yiwei Li, Jonathan Z. Zhang, and Robert W. Palmatier. 2022. “EXPRESS: Influencer Marketing Effectiveness.” Journal of Marketing, May, 002224292211028. https://doi.org/10.1177/00222429221102889.

Lovett, Mitchell J., Renana Peres, and Ron Shachar. 2013. “On Brands and Word of Mouth.” Journal of Marketing Research 50 (4): 427–44. https://doi.org/10.1509/jmr.11.0458.

McCracken, Grant. 1989. “Who Is the Celebrity Endorser? Cultural Foundations of the Endorsement Process.” Journal of Consumer Research 16 (3): 310. https://doi.org/10.1086/209217.

McQuarrie, Edward F., Jessica Miller, and Barbara J. Phillips. 2013. “The Megaphone Effect: Taste and Audience in Fashion Blogging.” Journal of Consumer Research 40 (1): 136–58. https://doi.org/10.1086/669042.

Nistor, Cristina, and Matthew Selove. 2024. “Influencers: The Power of Comments.” Marketing Science.

Rajaram, Prashant, and Puneet Manchanda. 2020. “Video Influencers: Unboxing the Mystique.” arXiv Preprint arXiv:2012.12311.

Shannon, C. E. 1948. “A Mathematical Theory of Communication.” Bell System Technical Journal 27 (3): 379–423. https://doi.org/10.1002/j.1538-7305.1948.tb01338.x.

Sokolova, Tatiana, Satheesh Seenivasan, and Manoj Thomas. 2020. “The Left-Digit Bias: When and Why Are Consumers Penny Wise and Pound Foolish?” Journal of Marketing Research 57 (4): 771–88. https://doi.org/10.1177/0022243720932532.

Spence, Michael. 1973. “Job Market Signaling.” The Quarterly Journal of Economics 87 (3): 355. https://doi.org/10.2307/1882010.

Thomson, Matthew. 2006. “Human Brands: Investigating Antecedents to Consumers’ Strong Attachments to Celebrities.” Journal of Marketing 70 (3): 104–19. https://doi.org/10.1509/jmkg.70.3.104.

Tian, Zijun, Ryan Dew, and Raghuram Iyengar. 2024. “Mega or Micro? Influencer Selection Using Follower Elasticity.” Journal of Marketing Research 61 (3): 472–95.

Weaver, Warren, Claude E Shannon, et al. 1963. The Mathematical Theory of Communication. Vol. 517. University of Illinois Press Champaign, IL.

Wilson, Elizabeth J., and Daniel L. Sherrell. 1993. “Source Effects in Communication and Persuasion Research: A Meta-Analysis of Effect Size.” Journal of the Academy of Marketing Science 21 (2): 101–12. https://doi.org/10.1007/bf02894421.

Yang, Jeremy, Juanjuan Zhang, and Yuhan Zhang. 2021. “First Law of Motion: Influencer Video Advertising on Tiktok.” Available at SSRN 3815124.

———. 2025. “Engagement That Sells: Influencer Video Advertising on TikTok.” Marketing Science 44 (2): 247–67. https://doi.org/10.1287/mksc.2021.0107.