865475

research-article20192019
SMSXXX10.1177/2056305119865475Social Media <span class="symbol" cstyle="Mathematical">+</span> SocietyGräve

SI: Networked Influence

Social Media + Society

What KPIs Are Key? Evaluating

July-September 2019: 1­–9 
© The Author(s) 2019
Article reuse guidelines:
Performance Metrics for Social sagepub.com/journals-permissions
DOI: 10.1177/2056305119865475
https://doi.org/10.1177/2056305119865475

Media Influencers journals.sagepub.com/home/sms

Jan-Frederik Gräve

Abstract
Measuring the impact of social media communication is a prominent and pertinent challenge; the commercialization of social
media influencers (SMIs) in the form of so-called influencer marketing makes this effort even more complicated. Companies
that embrace influencer marketing have limited control over content and context, so they must evaluate both the SMIs
and the content they post, prior to and during their collaborations. Although quantitative success metrics (e.g., number
of followers, number of likes) are readily available, it remains unclear whether such metrics offer appropriate proxies
for evaluating an SMIs or the outcomes of an influencer marketing campaign. By combining secondary data on influencer
marketing campaigns from Instagram with an online survey among marketers, this study finds that professionals generally rely
on an SMI’s reach and number of interactions as success metrics. When they must trade off across multiple metrics, these
professionals predominantly rely on comment sentiment, indicating their implicit awareness that the commonly used metrics
are inadequate. A regression analysis affirms that only the sentiment measure correlates positively with professional content
evaluations, so this study both challenges the use of common quantitative metrics to evaluate SMI content and emphasizes
the relevance of content-based metrics.

Keywords
social media influencer, influencer marketing, metrics, KPI

Introduction marketers and agencies therefore use various metrics as key

performance indicators (KPIs), both for the selection of
Social media influencers (SMIs) are increasingly prevalent SMIs and for measuring the outcomes (Fay & Larkin, 2017;
in social media domains. These content providers have Peters, Chen, Kaplan, Ognibeni, & Pauwels, 2013). For
acquired large audiences of thousands or even millions of example, to measure the reach and impact of SMIs, they
followers on social media platforms such as Facebook, might use the number of followers or number of interactions
YouTube, or Instagram, such that they can communicate to with each post (Peters et al., 2013). Such metrics are publicly
and influence a vast number of people with their messages available, easy to interpret, and comparable across social
(Uzunoğlu & Misci Kip, 2014). These large audiences also media platforms. Yet they might not be optimal or even suit-
attract the attention of brand marketers, which enter into paid able for selecting SMIs or predicting and then evaluating a
collaborations with SMI for advertising or product place- campaign’s advertising effectiveness.
ment purposes, in a tactic called influencer marketing (de Consider the number of followers metric as an example.
Veirman, Cauberghe, & Hudders, 2017). Many followers might indicate the substantial experience of
Selecting SMIs to collaborate with and measuring the out- SMIs, who likely adopt a professional communication
comes of the resulting campaigns represent two main chal- approach that enables them to design appealing, appropriate
lenges to influencer marketing. The social media environment
is fragmented across different platforms, so reviewing and University of Hamburg, Germany
evaluating SMIs manually would be extremely laborious.
Corresponding Author:
Then each influencer marketing campaign could include Jan-Frederik Gräve, Institute of Marketing, University of Hamburg,
dozens or hundreds of sponsored posts by the SMIs, making Moorweidenstraße 18, 20148 Hamburg, Germany.
it impossible for managers to assess them individually. Brand Email: [email protected]

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-
NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction
and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages
(https://us.sagepub.com/en-us/nam/open-access-at-sage).
2 Social Media + Society

posts that evoke positive responses by the audience. Yet the makeup and cosmetics with more than 27 million followers
number of followers marks the maximum potential reach, on Instagram (Socialblade, 2018c).
and an algorithmic selection of posts (as used by Facebook On these platforms, SMIs cover various topics and target
and Instagram) could lead to a significantly smaller audience groups, including fashion, gaming, and entertainment, by
(Isaac, 2016). Similarly, a large number of interactions may (allegedly) sharing their personal lives. As a result, SMIs are
indicate engaging, interesting content that is likely to pro- perceived as accessible, friendly, people next door, who
voke the audience’s positive attitude toward the influencer offer unbiased opinions and therefore possess high credibil-
marketing campaign. However, a high number of interac- ity and trustworthiness (Abidin, 2015; de Veirman et al.,
tions (e.g., comments) do not reveal the content or sentiment 2017). In turn, they represent efficient sources of electronic
of the interactions. Because SMIs also are well aware of the word of mouth and can be effective marketing tools (King,
use and importance of such metrics, they might try to opti- Racherla, & Bush, 2014). Brand marketers thus increasingly
mize their results, using acceptable methods (e.g., raffles to engage in influencer marketing, entering into paid collabo-
increase engagement) or disputable practices (e.g., buying rations with SMIs to promote brands and products (Carter,
followers) (Smith, 2017). To help clarify the value of metrics 2016; Lee & Watkins, 2016; Schwemmer & Ziewiecki,
for influencer marketing, this article establishes three main 2018). Industry reports suggest that most brands and compa-
research questions: nies plan on increasing their investments in influencer mar-
keting too, so this advertising channel appears likely to
Research Question 1: What metrics do brand marketers remain relevant for marketing communication (Linqia,
and agencies use primarily to support their influencer 2017). Collaborations with SMIs might range from single
marketing efforts? posts endorsing a product to long-term ambassadorships to
Research Question 2: Which metrics are most important launches of comprehensive product ranges dedicated to and
for evaluating influencer marketing campaigns? branded by a single SMIs (Hobbs, 2015; Lee & Watkins,
Research Question 3: Can metrics proxy for the actual 2016; Strugatz, 2016). The compensation for SMIs varies
perceived quality of the content posted by SMI in an widely too, from a free product sample to hundreds of thou-
influencer marketing campaign? sands of dollars (Carter, 2016).
As noted in the “Introduction” section, though, for brand
To address these questions, the next section outlines a marketers, influencer marketing strategies are complicated
conceptual background of social media metrics and influ- by the identification and selection of relevant, suitable col-
encer marketing. The “Research design and data collec- laboration partners (Carter, 2016), as well as the difficulties
tion” section describes the methodology, followed by the of measuring the outcomes of campaigns. The social media
“Data analysis and results” section. Finally, the “Discussion” environment is opaque in this sense, in that it is dynamic and
section provides a discussion of the findings. fast-paced, as well as fragmented across different platforms.
Beyond a few notoriously popular SMIs, marketers would
have a hard time finding effective SMIs that are relevant for
Conceptual Background their brand communication, in terms of reach and target mar-
ket. Some influencer marketing campaigns include dozens of
SMIs and Influencer Marketing
different SMIs, which makes it even more effortful to make
With the rise of social media, adopted by consumers world- a selection and then assess their impacts. Evaluations of
wide, SMIs have become increasingly broad phenomena. influencer marketing campaigns might seek to quantify the
These influencers are “opinion leader[s] in digital social outcome of a campaign in general or determine whether the
media who communicate[s] to an unknown mass audi- collaboration should continue. For these various tasks, mar-
ence” (Gräve, 2017, p. 1; Uzunoğlu & Misci Kip, 2014). keters often turn to metrics to quantify the (potential) impact
Digital social media encompass various platforms, includ- of SMI communications.
ing major networks such as Facebook, YouTube, Instagram,
and Twitter, and also special interest platforms like Twitch.
Social Media Metrics
In terms of their audience, SMIs range from individuals
with a few thousand followers to celebrities with several Technological advances and the introduction of new plat-
million fans. The former are sometimes referred to as forms and participatory features have profoundly changed
micro-influencers (Wissman, 2018) or micro-celebrities the Internet, transforming individual users from mere con-
(Abidin, 2016; Marwick, 2015); the latter often attract vast tent consumers into active participants and content creators
audiences and can be considered international celebrities. (Gerlitz & Helmond, 2013). Social media networks offer
For example, Felix Arvid Ulf Kjellberg (also known as many ways to interact with content, including subscribing,
“PewDiePie”), who publishes entertaining videos, has (dis)liking, sharing, and commenting. Such activities pro-
more than 66 million subscribers on YouTube (Socialblade, duce data and metrics that together constitute meaningful
2018b). Huda Kattan (“hudabeauty”) shares posts about information, and much of it is publicly available on social
Gräve 3

media networks. Depending on the platform, the number of as previously noted, metrics often serve as KPIs to evaluate
followers (e.g., Instagram), likes and comments (e.g., SMIs, single posts, and campaigns overall. The empirical
Facebook), or views of a post (e.g., YouTube) are observ- study in the next section seeks to determine which metrics
able to everyone. Some of these metrics are meaningful per practitioners should use.
se (e.g., number of subscribers), but others can have more
abstract meanings (e.g., clicking a Like button; Sumner,
Research Design and Data Collection
Ruge-Jones, & Alcorn, 2018). Platforms such as influ-
encerdb.net, socialblade.com, or socialbakers.com system- The research design relies on real influencer marketing cam-
atically collect such metrics in databases, compute paigns focused on a German audience, which then were
additional metrics (e.g., influencerdb.net, 2018, offers included as examples in a survey of German marketing pro-
“Estimated Post Values” in U.S. dollars and “Like- fessionals. Germany is generally comparable to other coun-
Follower-Ratios”), and offer to help marketing profession- tries, though the share of companies that engage in social
als find SMIs and measure their impact. These metrics media activities is slightly below the European average
actually serve as proxies; for example, the actual number of (Eurostat, 2019). Whereas SMIs communicating in English
attentive views of a post would be more meaningful than have the potential to reach an international audience, SMIs
the number of followers of an account. However, private communicating in German likely engage in national cam-
companies own the most popular social media networks paigns. This study leverages Instagram, which is especially
and decide what information to make publicly available, relevant for influencer marketing, with its more than 800 mil-
which information is restricted to the account owner (i.e., lion users (de Veirman et al., 2017; Linqia, 2017; Systrom,
business accounts on Facebook and Instagram receive 2017). The well-established practice of using campaign-­
details about the number of views of their posts), and what specific hashtags on this platform also makes it possible to
information is not available at all. identify posts related to a campaign more easily.
In research on social media metrics, most studies adopt a First, I identified influencer marketing campaigns on
brand communication perspective, such as a stream of Instagram directed toward a German audience with a consis-
research that investigates how companies and brands can tent campaign hashtag, by researching online press coverage
manage their social media presence efficiently and commu- and Instagram profiles of popular German SMIs. Only cam-
nicate effectively, e.g. by eliciting many interactions (de paigns with a distinctive hashtag were included (i.e., a
Vries, Gensler, & Leeflang, 2012; Peters et al., 2013). Other hashtag used only in the context of the campaign). I identi-
authors examine the outcomes of communication on social fied five suitable campaigns in 2017, featuring the brands
media on consumers’ offline behavior, sales, or returns on L’Oréal Paris (cosmetics), Rama (margarine), Rittersport
investment (Fay & Larkin, 2017; Fisher, 2009; Kumar, (chocolate), Coral (detergent), and Fitvia (detox products).
Bezawada, Rishika, Janakiraman, & Kannan, 2016). Such The data collection relied on Netlytic, a cloud-based tool
insights might help establish efficient influencer marketing that performs automated collections of publicly available
campaigns, but influencer marketing also creates a special social media posts (Gruzd, 2017), which gathered all
challenge in this regard. That is, for other types of paid and Instagram posts that included the focal campaign hashtag.
owned content on social media, brand marketers typically All posts were screened to eliminate those that were not part
produce the content themselves and retain control of the con- of the original campaign by the brand (e.g., posts mimicking
text in which it appears. In contrast, influencer marketing the campaign, posts by individuals with small audiences of
leaves brand managers with very limited control of the con- <1,000).
tent that SMIs post and the context in which it appears, Second, I expanded this data set with information about
because SMIs create the content themselves; brand market- the SMI, such as the number of followers they had at the time
ers can only define rough guidelines. Brand marketers accept the post was created, collected from Socialblade (2018a). To
this limited control, because it allows SMIs to communicate compute a sentiment-related metric, “net sentiment” (i.e.,
genuinely and retain their authenticity and trustworthiness number of positive comments minus number of negative
(de Veirman et al., 2017). Therefore, the posted content does comments; Fay & Larkin, 2017), the Linguistic Inquiry and
not necessarily adhere to a carefully crafted brand identity Word Count (LIWC) dictionary tool classified all SMI posts
but rather is composed according to the intuition of the influ- as positive, neutral, or negative (Kahn, Tobin, Massey, &
encer. A similar limitation exists with regard to the context: Anderson, 2007; Wolf et al., 2008). Sentiment analysis pro-
A firm can review a channel for general brand fit, but subse- vides a method to make affect quantifiable, which is espe-
quent posts could feature conflicting content, leading to neg- cially helpful for very large data sets in social media contexts
ative perceptions of the posts in this context. Thus, the firm (Puschmann & Powell, 2018). The analysis was conducted
needs some way to evaluate the content quality and to mea- using a German LIWC dictionary (Wolf et al., 2008), so no
sure the advertising effectiveness of the sponsored content in translations were necessary. Table 1 summarizes the descrip-
the context in which it appears. Because doing so is difficult, tive statistics of the data set.
4 Social Media + Society

Table 1.  Descriptive Statistics of Collected Campaign Posts.

L’Oréal Rama Coral Rittersport Fitvia

No. of SMIs 33 35 31 10 47
No. of posts 95 48 36 17 70
Mean followers (k) 197 148 173 340 83
Minimum followers (k) 1.6 7.2 10.6 82.3 3.3
Maximum followers (k) 2,900 916 1,050 1,010 526
Ø Interaction rate per post 3.9% 3.7% 5.1% 3.6% 3.1%
Ø No. of likes per post (k) 7.7 5.4 8.7 12.3 2.6
Ø No. of comments per post 50 42 83 75 45
Net sentiment 1,673 587 593 312 202

Note. SMIs = social media influencers.

Third, I conducted an online survey among social media Table 2.  Values of KPIs in the Conjoint Analysis.
professionals, including brand managers, public relations
Interactions Interaction Net
managers, and employees of advertising agencies. To collect (No.) rate (%) sentiment (k)
a convenience sample, I used posts on social media platforms
(e.g., Facebook, LinkedIn) and personal contacts in various Low 181.700 3.1 3.3
companies in the consumer goods industry and advertising Medium 459.100 5.2 6
agencies, whom I encouraged to share the survey within their High 736.500 7.3 8.7
organizations. Prior to the actual survey, the respondents Note. KPIs = key performance indicators.
answered questions regarding their professional experience
to verify their fit with the target group. Specifically, respon-
dents indicated whether they worked in the field of market- nine fictitious influencer marketing campaigns. The respon-
ing and for how long, and whether they had prior work dents then had to evaluate the campaigns from a professional
experience with social media marketing. Only respondents perspective and indicate which campaigns they considered
with professional experience completed the actual survey most successful by ranking them, 1 to 9, with a drag and drop
questions, offering their professional opinions. In total, 76 tool.
respondents (M age = 27.7 years) answered the questionnaire, Finally, in the third section of the survey, each respondent
of whom 57.9% were women and 47.4% had more than evaluated three randomly chosen SMI posts out of a sample
2 years of professional experience. In terms of professional of 30 posts that came from the initial data set, equal to 6 posts
background, 58% reported experience in influencer market- per campaign. The posts for each campaign reflect overall
ing within a company, 30% within an advertising agency, 7% variance in the campaign (i.e., scoring low, medium, and
in personal collaborations with companies, and 5% in another high values for each KPI). For every post, the number of
function. likes and comments, as well as the comments themselves,
The first section of the survey featured an open question, was deleted. Only the SMI’s user name, picture, and caption
to gain insight about the most relevant KPIs used in practice. were visible, as the examples in Figure 1 depict. Before each
Respondents had to identify the metrics they considered evaluation, respondents indicated if they were aware of the
most relevant for influencer marketing. Each respondent SMI and the brand. If they were, they also had to complete
could state up to five KPIs. Next, they had to trade off differ- items indicating their attitudes toward the SMI (three-item
ent KPIs to rank nine fictitious influencer marketing cam- scale; Martin, Kwai-Choi Lee, & Yang, 2004) and the brand
paigns presented to these respondents, according to their (three-item scale; Sengupta, Goodstein, & Boninger, 1997).
level of success. These unnamed campaigns were character- After respondents viewed the post, they offered general
ized by three KPIs: number of interactions (i.e., sum of all attitudes toward the advertisement and their professional
likes and comments), interaction rate (i.e., number of follow- evaluations. The professional evaluation consisted of three
ers divided by the sum of all likes and comments; Fisher, items on 7-point Likert-type scales: “The post shown is an
2009), and a sentiment index (i.e., positive minus the nega- example of well-implemented influencer marketing,” “The
tive comments; Fay & Larkin, 2017). To ensure realistic val- post shown is of high quality,” and “The influencer is profi-
ues, I derived three different values for each KPI from the cient in performing professional influencer marketing”
data related to the influencer marketing campaigns: mini- (Cronbach’s α  = .857; developed according to Singh,
mum, maximum, and mean. Table 2 lists the KPI values Balasubramanian, & Chakraborty, 2000).
included in the stimuli. These nine values were combined in All respondents were informed that their participation
a Latin square design (Green, 1974) to obtain values for the was anonymous and that their data would be used only for
Gräve 5

Figure 1.  Exemplary campaign posts for L’Oréal and Rama.

research purposes. All Instagram posts presented in the sur- evaluation of the SMIs’ posts and the respective KPIs pro-
vey and used in the data analysis were publicly available on vided input for a linear regression analysis. Estimating the
the Internet. relation between abstract KPIs and professional evaluation
of a specific SMI post should help clarify the extent to which
professional evaluations of a post are reflected by the corre-
Data Analysis and Results sponding KPIs used for the post and the SMI.
Following the structure of the survey, the data analysis con-
sists of three parts. First, to classify the freely stated KPIs, I
Stated KPI Classification
grouped identical and similar terms. The grouping reveals
the most relevant types of KPI. Second, I used traditional The classification of stated KPIs features two dimensions:
conjoint analysis to assess the rank orders of the fictitious quantitative versus qualitative and post/influencer related
influencer marketing campaigns. The respondents’ task is to versus sales/brand related. The results show a clear prefer-
sort a given number of choice options, with a given number ence for quantitative KPIs (Table 3); qualitative measures are
of characteristics, according to their personal preference. hardly mentioned. Furthermore, KPIs directly related to
Thus, the respondents must make trade-offs across the differ- posts are mentioned more often, especially the number of
ent choice characteristics, which reveals the relative impor- interactions/interaction rate and total reach (which ultimately
tance of each characteristic (Johnson, 1974). Third, the relates to the SMI, given its correlation with the number of
6 Social Media + Society

followers). The ultimate goal of influencer marketing—that all three KPIs; the higher the KPI, the higher its assigned
is, performance measures related to the firm—appears less utility value (see Table 4). Goodness-of-fit measures also can
often in these free responses. help evaluate the estimation: Pearson’s r coefficient mea-
Thus, the findings support the notion that managers sures the correlation between the estimated total utilities and
mostly rely on KPIs that are easily quantifiable and readily the empirical input ranks, and Kendall’s Tau indicates the
available. This choice is reasonable in a way; it minimizes discrepancy between the actual and predicted ranks (Green
the effort needed to evaluate influencer marketing cam- & Srinivasan, 1978; Rao, 2014, p. 105). Both measures
paigns. However, it is questionable whether this shortcut also (Pearson’s r = .991; Kendall’s Tau = .873) indicate the high
is a proxy for actually desired outcomes, namely, authentic, internal validity of the estimation.
high-quality content corresponding to the aspired brand Besides the estimated partworth utilities, I seek to deter-
image, which increases brand awareness, enhances the brand mine the relevance assigned to a KPI if multiple KPIs are
image, and ultimately leads to increased sales. available, as reflected in relative importance values (Johnson,
1974). If forced to trade off the three KPIs, respondents
assign the highest weight (43.3%) to the net sentiment in
Conjoint Analysis their evaluations of an influencer marketing campaign. The
The KPIs in the second section of the survey include the number of interactions (30.3%) ranks second, and the inter-
post-related dimensions mentioned most often in the free action rate is least important (26.4%). These results are
response: interactions, reach, and sentiment. These KPIs rep- inconsistent with the findings from the first section of the
resent measures that practitioners use to evaluate campaigns. survey, in which participants mentioned the number of inter-
The conjoint analysis yields plausible partworth utilities for actions most often. This discrepancy indicates that the impor-
tance of the number of interactions and reach may be due to
their mere presence and availability. When considered in
Table 3.  Classification of Stated KPIs. relation to other measures like net sentiment, managers
attach greater importance to this sentiment measure rather
Quantitative (225) Qualitative (30)
than relying solely on the quantity of reactions.
Post/ No. of interactions/ Sentiment of
influencer interaction rate (82) comments (8)
related Reach (41) Influencer quality (7) Regression Analysis
(182) No. of views/clicks (23) Target group fit (5) A linear regression can reveal the extent to which the three
No. of followers (12) Post quality (4)
abstract metrics used in the conjoint analysis relate to actual
Sales/brand Sales/turnover (15) Brand image (6)
related (73) Click-through rate (13)
professional attitudes toward an SMI post, as expressed in a
Conversion rate (12) direct evaluation. The regression model includes Professional
Social media impact (8) Evaluation of a Post as a dependent variable; the indepen-
Brand awareness (6) dent variables are three metrics: Number of Followers of the
Cost (6) SMI, Interaction Rate of the post, and Net Sentiment of the
Customer acquisition (4) post. Several covariates help control for individual differ-
ROI (3) ences across the posts and the respondents. The post-related
Note. Numbers in brackets indicate how often this type of KPI was variables include Promotional Code (dummy variable, indi-
mentioned. KPIs = key performance indicators; ROI = return on investment. cating whether a coupon code for an online shop appeared in

Table 4.  Results of the Conjoint Analysis.

KPI Value of KPI Partworth utility Relative importance SD

Number of 181,700 1.083 30.3 .139
interactions 459,100 2.167 .278
736,500 3.250 .416
Interaction rate 3.1 1.029 26.4 .139
5.2 2.058 .278
7.3 3.087 .416
Net sentiment 3.3 1.699 43.3 .139
6 3.399 .278
8.7 5.098 .416
Constant –2.623 .494

Note. KPI = key performance indicators; SD = standard deviation.

Gräve 7

Table 5.  Results of the Regression Analysis.

DV: Professional Evaluation of a Post

2
Adjusted R  = .413, F statistic = 8.113
Independent variables ß SE T p
Constant .953 .820 1.162 .248
Direct effects
  No. of followers (million) .260 .179 1.452 .150
  Interaction rate –.018 .030 –0.597 .552
  Net sentiment .018 .007 2.496 .014
Covariates
  Promotional codea 2.194 .758 2.894 .005
  Post length –.002 .001 –3.010 .003
  Brand visibility .395 .267 0.1481 .142
  General ad attitude .549 .081 6.747 .000
  Pre-brand attitude .366 .100 3.663 .000
 Age –.012 .017 –0.677 .500
 Gender –.056 .235 –0.240 .811
  n = 102

Note. DV = dependent variable; ß = unstandardized regression coefficient; SE = standard error.

a
Dummy variable indicating whether a promotional code is included in the post.

the post), Post Length (number of words in the caption of the Discussion
post), and Brand Visibility (dummy variable, indicating
whether the brand name or logo is visible in the post). At the This article addresses two of the main challenges to influ-
respondent level, the model controls for Age and Gender, encer marketing: the identification and selection of SMIs and
general Attitude Toward the Advertisement in the post, and the evaluation of influencer marketing campaigns using met-
Pre-Brand Attitudes. Respondents who were not aware of the rics as KPIs. To do so, it starts from a foundation of three
brand previously were excluded from this analysis. In total, research questions, related to the metrics brand marketers
the data set comprises 102 evaluations of SMI posts. and agencies use, the ones they consider most important for
The results reveal an adjusted R2 value of .413, resulting evaluating an influencer marketing campaign, and the met-
in a highly significant F statistic (p < .001; Table 5). The rics that can proxy for actual perceived quality of the content
variance inflation factors for the independent variables (all of an influencer marketing campaign.
<2) indicate no risk of multicollinearity. The number of fol- Regarding the first two questions, the empirical study
lowers (p = .150) and interaction rate (p = .552) have no sig- shows that marketers mostly rely on quantitative, readily
nificant effect on professional attitudes toward a post. In available metrics, like the number of interactions or reach, to
contrast, the analysis uncovers a significant positive effect evaluate influencer marketing activities. Almost 53% of all
of the net sentiment value (p = .014), underlining the impor- mentions fall into this category. The overreliance on these
tance of this kind of measure to evaluate the quality and KPIs seems driven largely by their mere availability. In con-
effect of an influencer marketing campaign. The higher the trast, the conjoint analysis revealed that in the presence of
net sentiment of the campaign, the more positive is the pro- additional KPIs that resemble the sentiment of the audience,
fessional evaluation of campaign posts. Therefore, the net managers attach greater importance to such measures than to
sentiment is an abstract KPI that can determine whether purely count-based metrics. Managers therefore seem (at
content is of high quality and can be considered good influ- least implicitly) aware that the informative value of the met-
encer marketing, without evaluating every single post rics they predominantly use is subpar. The preference for the
individually. net sentiment metric could arise because positive net senti-
The model estimation for the covariates shows several ment is the desired outcome, such that marketers strive for
significant effect sizes, indicating the need to control for more positive comments and few negative comments and are
individual differences. However, the effect of the dummy willing to pay for it; it also could indicate that net sentiment
variable Influencer Awareness (dummy variable, whether the provides a seemingly effective proxy of general advertising
respondent knows the influencer whose post is shown) is effectiveness, in that this metric contains more information
insignificant (p = .796). The variable thus was excluded from than a simple difference in the number of positive and nega-
the final model. tive comments. The former explanation might be embraced
8 Social Media + Society

by marketing professionals, but the regression analysis sug- ORCID iD

gests the latter is the case. For the last research question, only Jan-Frederik Gräve https://orcid.org/0000-0001-5408-3348
the net sentiment metric relates positively to professional
attitudes toward a post, therefore emphasizing the impor- References
tance of such KPIs for evaluating the success of influencer
Abidin, C. (2015). Communicative intimacies: Influencers and
marketing activities on an abstract level, without requiring perceived interconnectedness. Ada: A Journal of Gender, New
assessments of every single campaign post. Managers should Media, and Technology 8, 1–16. doi:10.7264/N3MW2FFG
assign greater emphasis to sentiment measures drawn from Abidin, C. (2016). “Aren’t these just young, rich women doing
comments to evaluate influencer marketing activities rather vain things online?”: Influencer selfies as subversive frivolity.
than looking at simple heuristics like reach or interaction Social Media + Society, 2(2), 1–17. doi:10.1177/2056305116
rate. However, this finding should not be taken to mean that 641342
the latter measures are not important: Managers should con- Carter, D. (2016). Hustle and Brand: The sociotechnical shap-
sider (potential) reach and interaction rate when choosing ing of influence. Social Media + Society, 2(3), 1–12.
SMIs for collaborations. Still, it is important to understand doi:10.1177/2056305116666305
that these conditions are necessary, not sufficient, to guaran- de Veirman, M., Cauberghe, V., & Hudders, L. (2017). Marketing
tee high-quality SMI content. through Instagram influencers: The impact of number of fol-
This research also is subject to several limitations that lowers and product divergence on brand attitude. International
offer potential for further research advances. First, I only Journal of Advertising, 36, 798–828. doi:10.1080/02650487.2
used 30 posts from 5 example influencer marketing cam- 017.1348035
de Vries, L., Gensler, S., & Leeflang, P. S. H. (2012). Popularity of
paigns on Instagram to gather professional attitudes, and the
brand posts on brand fan pages: An investigation of the effects
survey covered a limited number of KPIs. In addition, this of social media marketing. Journal of Interactive Marketing,
study focused on German influencer marketing campaigns 26, 83–91. doi:10.1016/j.intmar.2012.01.003
and accordingly surveyed German professionals. Continued Eurostat. (2019). Social media—Statistics on the use by enterprises.
research could broaden the scope by reviewing a greater Retrieved from https://ec.europa.eu/eurostat/statistics-explained/
variety of influencer marketing activities on different social index.php?title=Social_media_-_statistics_on_the_use_by_
media platforms in different countries, as well as more and enterprises
varied KPIs, to ensure that the results generalize to all cases Fay, B., & Larkin, R. (2017). Why online word-of-mouth measures
of influencer marketing. Second, the computation of certain cannot predict brand outcomes offline. Journal of Advertising
KPIs could be subject to further investigation. Recent Research, 57, 132–143. doi:10.2501/JAR-2017-021
research indicates that, among automated text classification Fisher, T. (2009). ROI in social media: A look at the argu-
methods, machine learning algorithms such as random for- ments. Journal of Database Marketing & Customer Strategy
Management, 16, 189–195. doi:10.1057/dbm.2009.16
ests outperform lexicon-based approaches like LIWC
Gerlitz, C., & Helmond, A. (2013). The like economy: Social but-
(Hartmann, Huppertz, Schamp, & Heitmann, 2018). Thus, it
tons and the data-intensive web. New Media & Society, 15,
would be worthwhile to investigate whether the findings of 1348–1365. doi:10.1177/1461444812472322
the current analysis hold when advanced classification meth- Gräve, J.-F. (2017, July). Exploring the perception of influencers
ods with higher accuracy compute the sentiment metric. vs. traditional celebrities. Proceedings of the 8th International
In the near future, the fields of influencer marketing and Conference on Social Media & Society—#SMSociety17,
SMIs promise to remain dynamic and challenging, due to ACM Press, New York, NY. doi:10.1145/3097286.3097322
the development of new features (e.g., Instagram stories), Green, P. E. (1974). On the design of choice experiments involv-
emerging social media platforms, regulatory changes, and ing multifactor alternatives. Journal of Consumer Research, 1,
increasing professionalization of both SMIs and specialized 61–68. doi:10.1086/208592
marketers. Researchers thus must continue to extend knowl- Green, P. E., & Srinivasan, V. (1978). Conjoint analysis in con-
edge and best practices in this field. sumer research: Issues and outlook. Journal of Consumer
Research, 5, 103–123. doi:10.1086/208721
Acknowledgement Gruzd, A. (2017). Netlytic: Software for automated text and social
network analysis. Available from http://Netlytic.org
Special thanks to Annika Greff for her help in the data collection. Hartmann, J., Huppertz, J., Schamp, C., & Heitmann, M. (2018).
Comparing automated text classification methods. International
Declaration of Conflicting Interests Journal of Research in Marketing, 36, 20–38.
The author(s) declared no potential conflicts of interest with respect Hobbs, T. (2015). Superdrug’s marketing director on champion-
to the research, authorship, and/or publication of this article. ing Zoella and why vloggers are “not a fad.” Retrieved from
https://www.marketingweek.com/2015/07/27/superdrugs-mar-
keting-director-on-championing-zoella-and-why-vloggers-are-
Funding not-a-fad/
The author(s) received no financial support for the research, author- Influencerdb.net. (2018). Analyze influencers. Retrieved from
ship, and/or publication of this article. https://www.influencerdb.net/insights/
Gräve 9

Isaac, M. (2016). Instagram may change your feed, personalizing low-involvement conditions. Journal of Consumer Research,
it with an algorithm. Retrieved from https://www.nytimes. 23, 351–361. doi:10.1086/209488
com/2016/03/16/technology/instagram-feed.html Singh, M., Balasubramanian, S. K., & Chakraborty, G. (2000).
Johnson, R. M. (1974). Trade-off analysis of consumer val- A comparative analysis of three communication formats:
ues. Journal of Marketing Research, 11, 121–127. doi: Advertising, infomercial, and direct experience. Journal of
10.2307/3150548 Advertising, 29, 59–75. doi:10.1080/00913367.2000.10673624
Kahn, J. H., Tobin, R. M., Massey, A. E., & Anderson, J. A. (2007). Smith, M. (2017). Why you shouldn’t buy Instagram followers from
Measuring emotional expression with the Linguistic Inquiry vending machines—Opinion: Is automating your Instagram
and Word Count. The American Journal of Psychology, 120, too good to be true? Retrieved from http://www.adweek.com/
263–286. doi:10.2307/20445398 digital/matt-smith-later-guest-post-why-you-shouldnt-buy-
King, R. A., Racherla, P., & Bush, V. D. (2014). What we know and instagram-followers-from-vending-machines/
don’t know about online word-of-mouth: A review and syn- Snukieful. (2017, September). Anzeige Was ich im Kiteurlaub gel-
thesis of the literature. Journal of Interactive Marketing, 28, ernt habe: . .. [Instagram Post]. Retrieved from https://www.
167–183. doi:10.1016/j.intmar.2014.02.001 instagram.com/p/BZGlCnAlQFp/
Kumar, A., Bezawada, R., Rishika, R., Janakiraman, R., & Kannan, Socialblade. (2018a). Social blade Instagram stats. Retrieved from
P. K. (2016). From social to sale: The effects of firm-gener- https://socialblade.com/instagram/
ated content in social media on customer behavior. Journal of Socialblade. (2018b). Top 50 subscribed YouTube channels.
Marketing, 80, 7–25. doi:10.1509/jm.14.0249 Retrieved from https://socialblade.com/youtube/top/50/most-
Lea_van_acken. (2017, September). Was ist Schönheit für dich? subscribed
[Instagram post]. Socialblade. (2018c). Top 100 Instagram users by followers.
Lee, J. E., & Watkins, B. (2016). YouTube vloggers’ influ- Retrieved from https://socialblade.com/instagram/top/100/fol-
ence on consumer luxury brand perceptions and intentions. lowers
Journal of Business Research, 69, 5753–5760. doi:10.1016/j. Strugatz, R. (2016). Bloggers and digital influencers are reshaping
jbusres.2016.04.171 the fashion and beauty landscape. Retrieved from http://www.
Linqia. (2017). The state of influencer marketing 2018. Retrieved from latimes.com/fashion/la-ig-bloggers-20160809-snap-story.html
http://www.linqia.com/wp-content/uploads/2017/12/Linqia-The- Sumner, E. M., Ruge-Jones, L., & Alcorn, D. (2018). A func-
State-of-Influencer-Marketing-2018.pdf tional approach to the Facebook like button: An exploration
Martin, B. A. S., Kwai-Choi Lee, C., & Yang, F. (2004). The influ- of meaning, interpersonal functionality, and potential alterna-
ence of ad model ethnicity and self-referencing on attitudes. tive response buttons. New Media & Society, 20, 1451–1469.
Journal of Advertising, 33, 27–37. doi:10.1080/00913367.20 doi:10.1177/1461444817697917
04.10639172 Systrom, K. (2017). Strengthening our commitment to safety and
Marwick, A. E. (2015). Instafame: Luxury selfies in the atten- kindness for 800 million. Retrieved from http://blog.instagram.
tion economy. Public Culture: Bulletin of the Center for com/post/165759350412/170926-news
Transnational Cultural Studies, 27, 137–160. doi:10.1215/ Uzunoğlu, E., & Misci Kip, S. (2014). Brand communication
08992363-2798379 through digital influencers leveraging blogger engagement.
Peters, K., Chen, Y., Kaplan, A. M., Ognibeni, B., & Pauwels, K. International Journal of Information Management, 34, 592–
(2013). Social media metrics—A framework and guidelines for 602. doi:10.1016/j.ijinfomgt.2014.04.007
managing social media. Journal of Interactive Marketing, 27, Wissman, B. (2018). Micro-influencers: The marketing force of
281–298. doi:10.1016/j.intmar.2013.09.007 the future? Retrieved from https://www.forbes.com/sites/bar-
Puschmann, C., & Powell, A. (2018). Turning words into consumer rettwissman/2018/03/02/micro-influencers-the-marketing-
preferences: How sentiment analysis is framed in research force-of-the-future/
and the news media. Social Media + Society, 4(3), 1–12. Wolf, M., Horn, A. B., Mehl, M. R., Haug, S., Pennebaker,
doi:10.1177/2056305118797724 J. W., & Kordy, H. (2008). Computergestützte quantitative
Rao, V. R. (2014). Applied conjoint analysis. Heidelberg, Germany: Textanalyse [Computer-assisted quantitative text analysis].
Springer. Diagnostica, 54, 85–98. doi:10.1026/0012-1924.54.2.85
Schwemmer, C., & Ziewiecki, S. (2018). Social media sellout:
The increasing role of product promotion on YouTube. Author Biography
Social Media + Society, 4(3), 1–20. doi:10.1177/2056305 Jan-Frederik Gräve is a doctoral student and research associate at
118786720 the Institute of Marketing at University of Hamburg, Germany. His
Sengupta, J., Goodstein, R. C., & Boninger, D. S. (1997). All cues research focuses on brand communication in social media, in par-
are not created equal: Obtaining attitude persistence under ticular influencer marketing and advertising endorsements.