Whose and What Social Media Complaints Have Happier Resolutions? Evidence from Twitter

Whose and What Social Media Complaints

Have Happier Resolutions? Evidence from

Twitter

PRIYANGA GUNARATHNE, HUAXIA RUI, AND

ABRAHAM SEIDMANN

PRIYANGA GUNARATHNE (priyanga.gunarathne@simon.rochester.edu) is a Ph.D. can-

didate in computer information systems at the Simon Business School of the

University of Rochester. Before joining Simon Business School, she worked as a

software engineer in financial markets. Her primary research interests are consumer

and organizational behavior on social media, and the use of deep learning and

natural language processing in business.

UAXIA RUI ([email protected]ter.edu) is the Xerox Assistant Professor at the

Simon Business School of the University of Rochester. He received his Ph.D. from

the University of Texas at Austin in 2012. His work has been published in

Management Science, Information Systems Research, Journal of Management

Information Systems, MIS Quarterly, Production and Operations Management,

and others.

BRAHAM SEIDMANN ([email protected]chester.edu; corresponding author) is

the Xerox Professor of Computers and Information Systems and Operations

Management at the Simon Business School of the University of Rochester. His

research and consulting work focuses on the development of advanced analytical

tools for solving real business problems in information-intensive industries, most

recently in health care, web services, software development, digi tal supply chains,

and automated manufacturing. His research also involves the economic aspects of

information technology and its strategic interactions with organizations, markets,

and value chain operations. He is a Distinguished Fellow of the INFORMS

Information Systems Society.

BSTRACT: Many brands try to manage customer complaints on social media, help-

ing their customers on a real-time basis. Inspired by this popular practice, in this

study, we aim to understand whose and what complaints on social media are likely

to have happier resolutions. We analyzed the complaint resolution experience of

customers of a major U.S. airline, by exploiting a unique data set combining both

customer–brand interactions on Twitter and how customers felt at the end of these

interactions. We find that complaining customers who are more influential in online

social networks are more likely to be satisfied. Customers who have previously

complained to the brand on social media, and customers who complain about

process-related rather than outcome-related issues are less likely to feel better in

the end. To the best of our knowledge, this study is the first to identify the key

factors that shape customer feelings toward their brand–custo mer interactions on

Journal of Management Information Systems / 2017, Vol. 34, No. 2, pp. 314–340.

ISSN 0742–1222 (pr int) / ISSN 1557–928X (online)

DOI: https://doi.org/10.1080/07421222.2017.1334465

social media. Our results provide practical guidance for successfully resolving

customers’ complaints through the use of social media—an area that expects expo-

nential growth in the coming decade.

EY WORDS AND PHRASES: airline industry, CLASS methodology, complaint manage-

ment, customer service, Klout score, online complaints, social influence, social

media, Twitter.

On March 13, 2014, Lauren Munhoven, a customer in Ketchikan, Alaska, turned to

Twitter after wasting an hour on the phone with General Motors (GM), trying to get

help with her 2006 Saturn Ion regarding a GM vehicle recall [16]. After she wrote a

public tweet—“@GM your agents keep telling me to take my car to a GM dealer for

the recall, after I’ve explained I live on an isl and in Alaska! Help!!!!”—a member of

GM’s Twitter team helped. The company agreed to pay the $600 cost of a round-trip

ferry to ship Ms. Munhoven’s car to the nearest dealer, about 300 miles away in

Juneau, and to pay for a rental car for the time she would be without the Saturn. Ms.

Munhoven credited the public nature of complaining on social media with getting

GM’s attention, and she was so pleased that she posted a public thank-you to GM on

Twitter.

Empowered by social media and mobile technologies, more and more customers

are turning to social media platforms such as Twitter and Facebook to post their

complaints to brands in real-time. In response, brands are striving to monitor and

quickly respond to those complaints to prevent them from festering and damaging

their reputation. Many consumer brands equip their social media teams with sig-

nificant organizational customer relationship management (CRM) experience, as

well as access to the associated CRM system, so that complaints can be effectively

and efficiently managed online. For instance, airline social media teams today can do

flight reservations and rebooking for customers in real-time. In the pre-social-media

era, customers directly contacted the brand’s custo mer service call center to begin

the organizational complaint management process, and the communication between

the customer and the brand was always kept private and confidential. In contrast,

social media has enabled customers to publicly express their dissatisfaction about a

brand, and the brand’s dedicated social media team starts a conversation with the

customer openly. Such conversations are typically open to third-party audiences such

as the followers of the customer, or to anybody if the posts do not assume any

privacy. The simplicity and timeliness of delivering customer service through social

media might quickly turn an angry or unhappy customer into a calm, relieved, or

even happy customer at the end of the interaction, thereby making it more likely that

the brand will retain the customer.

Inspired by this growing phenomenon, in this study, we aim to understand whose

and what complaints on social media are likely to have happier resolutions. On the

customer side (i.e., whose complaint), we first examine whether a custo mer’s online

social influence is related to whether the customer would feel better after

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 315

complaining to and interacting with a brand’s social media team. Second, we

examine whether a customer’s past complaints are related to the c ustomer’s com-

plaint resolution experience on social media. A customer’s online influence and

complaint history on social media are probably the two most salient and relevant

customer characteristics in our context. Both are likely to a ffect a customer’s

satisfaction with the complaint resolution, and both can be directly measured and

easily used by a brand to customize its social media custo mer service. On the

complaint side (i.e., what complaint), we investigate whether customers complaining

about outcome-related issues (i.e., operations) and customers complaining about

process-related issues (i.e., employees) have different degrees of satisfaction about

their complaint resolution experience on social media. Prior literature [4, 34] has

suggested different customer perceptions of outcome-related service failure and

process-related service failure. Therefore, it is both intellectually interesting and

practically important to understand whether the distinction between these two types

of complaints has any implications for customer satisfaction regarding complaint

resolution on social media.

To address our research questions, we first developed our Closed Loop Analytical

Social Survey (CLASS) methodology, which is an innovative way of learning how

happy a customer felt about an interaction in complaining to a brand’s social media

customer service team. We directly queried 1,500 randomly selected customers with

recent complaints to a major U.S. airline on Twitter.

We then analyzed the profiles of the complaining customers, their complaints, and

their responses to our survey questions to estimate an ordered logit model. The

estimation results suggest that a customer’s satisfaction with the complaint resolution

experience on social media is positively associated with the customer’s online social

influence and is negati vely associated with whether the customer has complained

previously to the airline on social media. We also find that a customer’s satisfaction

is positively associated with whether the complaint is outcome-related. To alleviate

the concern about potential nonresponse bias, as complaining customers who are

unhappier about their social media interaction may also be more likely to respond to

surveys such as ours, we augmented our data with customers who did not respond to

our survey and employed a Heckman-type procedure to estimate our model. The

results are quali tatively the same as our benchmark model and indicate that non-

response bias is not likely a major concern in this study.

Our study makes important contributions to the field of information systems (IS)

and service management in the social media era. Previous studies mostly looked at

the causes and the sources of consum er complaint behavior [11, 15, 32, 37], and the

procedural determinants of the organizational compl aint management process, with

specific focus on repurchase intentions, potential word-of-mouth, and customer

satisfaction with the outcomes [4, 7, 9, 34]. To the best of our knowledge, our

study is the first to investigate factors that may potentially affect how complaining

customers would actually feel after their interaction with customer service on social

media. Given the growing importance of using social media to deliver customer

service and the sparse literature studying this new phenomenon, our arti cle offers

316 GUNARATHNE, RUI, AND SEIDMANN

timely and much-needed empirical evidence to guide managers in practice. For

example, our findings suggest that it is worthwhile to customize a brand’s response

to a customer complaining on social media based on the customer’s complaint

history, rather than treating each complaint as completely new. Although a brand

may track complaints received via the traditional complaint management process,

whether the same is true for the complaints received via social media is not evident.

Thus, our findings on the importance of accounting for customers’ past complaints

on social media in providing happier complaint resolutions contribute to the current

literature on customer complaint management in the age of social media.

Another important contribution of our research is the CLASS methodology that we

used to directly survey customers who received customer service through social

media. Although many studies have used social media data to study various inter-

esting questions, we leverage the power of social surveillance to establish the

missing link between researche rs and actual customers by using social media data

in an active rather than passive manner. Such a methodology can be easily auto-

mated and leveraged to generate large amounts of customer satisfaction data, which

could be extremely valuable to researchers, companies, and policymakers.

Literature Review

The study of complaint management is challenging because complaints and their

handling are only triggered by a service failure, making systematic empirical

research almost impossible to conduct in either a laboratory or a field environ-

ment [34]. As a result, it is difficult to find an ideal setup to do causal inference

in a strict sense. Nevertheless, studies in traditional complaint management have

used hypothetical service failure scenarios or incident recall techniques to capture

customer perceptions about organizations’ complaint-handling processes.Thus,

several complaint management frameworks have been developed over the years

to model satisfaction, and other postcomplaint customer attitudes and behaviors

[5 , 9, 34, 41].

Tax et al. [41] examined the influence of a customer’s justice evaluations on

complaint-handling satisfaction, trust, and commitment, differentiating between the

distributive, procedural, and interactional justice aspects of service. Distributive

justice in complaint handling refers to whether the complaint outcome was perceived

to be deserved, met the customer’s needs, or was fair. Procedural justice refers to the

perceived fairness of the complaint-handling process, including quality attributes

such as accessibility, timing/speed, flexibility, process control (i.e., freedom to

communicate views on a decision process), and decision control (i.e., the extent to

which the customer is free to accept or reject an outcome). Interactional justice

refers to the treatment the complainant receives in the direct interaction with the

employees of the organization, which includes quality attributes such as explanation/

causal account, empathy, politeness, honesty, and effort. This justice-evaluation

approach provides a comprehensive list of complaint management quality attributes

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 317

and calls for a dimensional structure, if the different types of perceived justice are

interpreted as quality dimensions of complaint management [36].

Another empirically tested approach to conceptualize customer satisfaction with

service encounters involving failure and recovery was presented by Smith et al. [34].

Their model provided a framework for considering how the context of service failure

(failure type, magnitude) and the attributes of service recovery (compensat ion,

response speed, apology, initiation) influence customer evaluations through discon-

firmation and perceived justice, thereby influencing satisfaction with the service

encounter. Their findings suggest that satisfaction is related positively to perceptions

of distributive, procedural and interactional justice. Also, a customer’s satisfaction

level after a service failure is shown to depend on both the type and the magnitude of

the failure. Moreover, all the service recover y attributes show positive effects on

perceptions of justice, thereby positively influencing complaint satisfaction.

Davidow [9] presented a comprehensive framework of organizational complaint

management, empirically differentiating between organizational response dimen-

sions, customer satisfaction, and postcomplaint customer behaviors such as word-

of-mouth acti vity and intention to repurchase. His model subsumed six organiza-

tional response dimensions (timeliness, facilitation, redress, apology, credibility, and

attentiveness) that incorporated almost all the dimensions mentioned in prior studies

of complaint management. His empirical study found only timeli ness, credibility,

redress, and attentiveness to have significant positive impacts on satisfaction.

Several other studies on traditional complaint management also showed mixed

results regarding which aspects of organizational responses to complaints are most

effective in shaping postcomplaint customer behavior [10]. Levesque and

McDougall [25] investigated the connection between the type of problem and

customer dissatisfaction with issues associated with service outcomes, service pro-

cess, pricing, and location. Their findings suggest that customers are more likely to

voice a complaint than to exit when they encounter problems and the importance of

the problem is link ed to the rate of taking action. Estelami [13] examined the impact

of various procedural determinants of complaint handling such as compensation,

employee behavior, and promptness, on the creation of outstanding complaint

resolutions. The author found that consumer delight and disappointment with com-

plaint outcomes are primarily influenced by the compensatory aspects of complaint

resolutions. Strauss and Hill [39] explored company responses to genuine com-

plaints via e-mails, and consumer reactions to those responses. They found 47

percent of the firms responded to the complaint e-mails, which in turn resulted in

higher customer satisfaction and purchase likelihood. Additionally, response e-mails

that were sent quickly, addressed the specific problem, and were signed with an

employee’s name resulted in higher customer satisfaction.

Although delivering customer service on social media has become very popular

today, there is little empirical research about the effectiveness of this practice. Our

research thus fills this gap and contributes to the stream of research literature on

customer complaint management in the digital age by investigating the key factors in

postcomplaint customer satisfaction in the context of social media and offering

318 GUNARATHNE, RUI, AND SEIDMANN

guidance on how customer service can be successfully delivered through social

media platforms.

Development of Hypotheses

Building on the current research literature on complaint management, theories from

social psychology, and anecdotal evidence on social customer service, in this sec-

tion, we discuss the theoretical background of our hypotheses.

Social Influence

Online social influence is an important concept arising from the popularity of social

media and companies’ grow ing desire to identify potential opinion leaders and

influential word-of-mouth (WOM) on social media [31, 40]. One popular measure

of a person’s online social influence is the person’s Klout score . According to

Klout’s official website, klout.com, the Klout score algorithm uses more than 400

signals from several leading online social networks (e.g., Twitter, Facebook,

Instagram, Foursquare, Blogger, Tumblr, Google+, and LinkedIn) and data from

places such as Bing and Wikipedia to construct Klout scores. Since its launch in

2008, the Klout score has become a popular marketing tool, as several leading

brands offer free and exclusive products and experiences (i.e., perks) to high Klout

scorers, who are often happy to spread the brand message on social medi a, even

though they are not required to do so. For example, for limited periods in recent

years, American Airlines and Cathay Pacific Airways granted high Klout scorers

access to their exclusive airport lounges, which would otherwise have been available

only to their first-class or business-class passengers. Therefore, it is natural to

examine the role of online social influence in shaping the satisfaction judgments

of customers complaining to brands on social media.

There can be at least three potential mechanisms through which customers’ online

social influence might be associated with their judgment about satisfaction regarding

the resolution of a complaint received via social media: (1) the preferential customer

service that may be offered to customers of high social influence, (2) certain

personality characteristics of highly influential individuals, and (3) highly influential

customers’ likely higher level of expectations of social media customer servi ce.

Next, we discuss each of these three mechanisms in detail.

Preferential Customer Service

First, a customer with greater online social influence might have a better relationship

with a brand as a result of preferential treatment (e.g., access to airport lounges). A

more positive relationship between a customer and a brand moderates the customer’s

emotional reaction to, and results in higher satisfaction with a complaint resolution

experience [ 44]. Preferential treatment is the practice of giving some customers

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 319

elevated recognition, and additional or enhanced products and services above and

beyond standard firm value propositions and customer service practices [23]. For

example, a company may offer preferential treatment to a customer by placing the

customer higher on a priority list if there is a queue, and giving the customer more

attention or faster service than other customers [35]. Practitioners and academic

researchers have long been interested in identifying high-value customers [8 ]. The

traditional view of preferential treatment assumes that the customers earned the

special treatment through loyalty (i.e., their economic value) or effort. For example,

frequent flyer programs offer priority boarding and first-class/busin ess-class

upgrades to airlines’ frequent travelers. However, in the age of social media,

customers’ online social influence has become an important factor driving a brand’s

prioritization decisions. As Allon and Zhang [1] argue, it is not only the value that

customers bring in that matters to a company, but also the ability of those customers

to influence others in the social network. Pr ioritization based on a customer’s online

social influence has also become technically convenient in recent years. For exam-

ple, Genesys, a global omnichannel customer experience and contact center solution

provider for business clients, including major airlines, banks, and telecommunica-

tions companies, integrated the Klout score into its solutions. This enabled compa-

nies that used the Genesys platform to recognize their customers with high Klout

scores and route them to specialized customer service agents, if they wished to do

so. Althoug h it is unclear how widely and to what extent preferential customer

service that is based on online social influence has been adopted in practice, recent

research [20] finds evidence that in the airline industry, customers with larger online

social influence are more likely to receive a response, and also are more likely to

receive a response faster when they complain through social media.

While preferential treatment might make a customer happier with the brand and

thus happier with the complaint resolution experience, less-influential customers

may feel unhappy about their overall complaining experience on social media not

only because of the poorer treatment or a less-positive relationship due to poorer

treatment, but also because they perceive influence-based preferential treatment as

unfair [20]. Previous research indicates that perceived service unfairness induces

negative emotional reactions, such as feelings of betrayal and anger, as well as

behavioral responses, such as venting and revenge.

Customer Personality Traits

Certain personality characteristics might drive both a customer’s online social

influence and satisfaction with complaint resolution. For instance, customers with

greater online social influence may be happier people in general, so even in the

absence of preferential treatment, the very personalities of these influential custo-

mers could lead to more-positive feelings about the complaint resolution experi ence.

In fact, empirical studies in psychology show that happy people tend to be more

likable and thus more popular than unhappy people [3, 14]. Although many

320 GUNARATHNE, RUI, AND SEIDMANN

definitions of happiness have been used in the literature, in general, happy indivi-

duals are characterized as those who experience frequent positive emotions, such as

joy, interest, and pride, and infrequent (though not absent) negative emotions such as

sadness, anxiety, and anger [27]. Social influence is often associated with social

dominance [42]. Lucas et al. [26] found that across the world, positive feelings were

associated with tendencies for affiliation, dominance, excitement seeking, and social

interaction. Furthermore, influence has been shown to be associated with the Big

Five personality trait, emotional stability, which is a person’s ability to remain calm

when facing press ure or stress [26 , 29].

Does this imply that the unobserved

personality features related to a person’s general level of happiness and emotional

stability cause people to be more influential in online social networks? Although the

literature in psychology seems to support this claim, we may never know for sure.

However, if this is the ca se, the positive emotions and emotional stability that make

a customer influential in online social networks might make the customer perceive

the experience with a brand as more pleasant.

Differential Customer Expectations

A customer who is highly influential online might have higher expectations for customer

service and be less likely to be satisfied with the complaint resolution experience. This

conjecture may be supported by theories from behavioral psychology, where influence

has long been studied as a prominent style of human behavior. For instance, the

celebrated psychologist, William Marston, who proposed the dominance, inducement,

submission, and compliance model of human behavior (DISC) [28], recognized influ-

ence as “inducement” behavior, characteristic of people who can persuade, attract,

convince, convert, and lead other people. Recently, the DISC model of human behavior

has been adapted and customized to customer service [22], where the inducement

behavior is characteristic of the “influential customers,” who are optimistic, persuasive,

inspiring, and trusting in their approach toward customer service organizations and

expect the same in return. Such influential customers have a need for social recognition,

and always want to be accepted. Therefore, we could expect influential customers to

have higher than normal expectations for social media customer service.

Hence, from a purely theoretical perspective, it is not clear which mechanism

would dominate in shaping a customer’s satisfaction regarding the resolution of a

complaint. However, considering the nature of individuals of higher online social

influence, it is more likely that the first two mechanisms would dominate the third

mechanism. Thus, if we use a customer’s Klout score to measure his or her online

social influence, we might observe a positive or negative correlation between the

Klout score and the customer’s satisfaction with the complaint resolution. To

investigate this, we propose the following hypothesis for empirical testing:

Hypothesis 1 (The Social Influence Hypothesis): The higher a complaining

customer’s Klout score is, the more likely the customer will feel better at the

end of a conversation with a brand on social media.

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 321

Prior Complaint Experience

The effect of past behavior on individuals’ attitudes, intentions, and behavior has

long been recognized in studies of personality and social psychology. The idea that

the extent of past complaining experiences becomes assimilated into an indivi-

dual’s attitude tow ard complaining, is consistent with the behavioris t and situa-

tionist theories of psychology, which explain how past behaviors and exposure to

situations shape and reinforce an individual’s behavioral dispositions in future

situations [33]. From the perspective of customer–brand relationship, customers

with prior complaint experience are likely to have a weaker relationship with the

brand than customers without prior complaint experi ence do. Prior literatures

suggest that there is a buffering effect of relationship strength. For example, Xia

[44] finds that consumers with a stronger existing relationship with a brand will

perceive a defens ive reaction toward criticism by the brand as less inappropriate

than those with a weaker existing relationship. Hence, the threshold for successful

complaint resolution tends to be higher for customers who have had a negative

experience with the brand in the past than for customers who have had no such

experience. The se differences in expectations may lead to differences in postcom-

plaint satisfaction. Based on these arguments, we propose the following hypothesis

for empirical testing:

Hypothesis 2 (The Prior Complaint Experience Hypothesis): A complaining

customer is more likely to feel worse at the end of a conversation with a brand

on social media if the customer has prior complaint experience with the brand.

Complaint Type

When customers deal with service firms, the two main reasons for complaints are

the failure to deliver the service and how the service was delivered [4]. The

marketing literature recognizes these two types of servi ce failures as outcome

and process failures [4, 19, 25]. The outcome dimension involves what customers

actually receive from the service, or the performance aspects of the service, and the

ability of the organization to keep its promises and to solve problems when they

arise [19]. Processes involve the functional or people aspects of the service and are

a consequence of the behavior and customer-oriented service-mindedness of the

employees [19]. Therefore, in an outcome failure, the organization does not fulfill

the core service need, whereas in a process failure, the delivery of the core service

is flawed or deficient in some way [34]. For example, in the airline industry,

outcome-related complaints might include flight delays, flight cancellations, mis-

handled baggage, in-flight service-related issues (e.g., seats, wi-fi, meals), long

queues at check-in counters, and boarding issu es. Process-related complaints may

include issues such as unprofessional employees, and issues related to the airline’s

dedicated customer service (e.g., long on-hold times, pending refunds, mishandled

complaints).

322 GUNARATHNE, RUI, AND SEIDMANN

As per social exchange and equity theories [21, 43], a complaint encounter can

be viewed as an exchange in which the customer experiences a loss due to the

failure and the organization attempts to provide a gain, in the form of effective

handling of the complaint, to make up for the customer’s loss. Service failures can

result in the loss of economic resources (money, time) or psychological /social

resources (status, empathy, esteem) for customers, and organizations often offer

customers economic resources in the form of compe nsation, or psychological/social

resources such as an apology [34 ]. An outcome failure involves a loss of economic

resources, whereas a process failure involves a loss of psychological/social

resources. Thus, we expect customers’ complaint satisfaction judgments to differ

by the type of complaint, as outcome and process failures represent different

categories of loss to the customer. The marketing literature provides very limited

evidence on which type of failure has more influence on customers’ postcomplaint

satisfaction. Smith et al. [34] found that customers who experienced process

failures were more dissatisfied than those who experienced outcome failures.

Bitner et al. [ 4 ] found that a large percentage of unsatisfactory service encounters

were related to employees’ inability or unwillingness to respond effectively to

service failure. Furthermore, prior studies indicate that operational failures them-

selves do not necessarily lead to customer dissatisfaction, since most customers

accept that things may sometimes go wrong [12]. However, if it is the organiza-

tion’s employees who failed to live up to customer expectations of service, it is

less likely that the customers will be satisfied with their experience. Based on these

arguments, we propose the following hypothesis for empirical test:

Hypothesis 3 (The Complaint Type Hypothesis): A complaining customer is

more likely to feel better at the end of a conversation with a brand on social

media if the complaint is outcome-related rather than process-related.

Data, Measures, and Methodology

Data Collection

We used the Twitter API (application program interface) to collect all the user tweets

mentioning the official Twitter account of a major U.S. airline, which we have

purposely kept anonymous, as well as all the tweets posted by that airline, from July

2014 to January 2015. We processed the tweets daily, when constructing our main

data set of complaint-based conversations between the users (i.e., customers) and the

airline on Twitter. We define a conversation as a dialogue between a customer and

the airline on Twitter, containing all the tweets the customer sent to the airline

regarding a particular complaint and the associated reply tweets from the airline. We

obtained a variety of conversations users had with the airline, particularly on

complaints, compliments, and on information sharing in general. Then we processed

our data to identify all the complaint-based conversations, and randomly selected 40

percent of the conversations for further analysis. Taking the concise nature of

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 323

communication on Twitter into account, we picked only conversations with at least

two replies from the airline, for our main data set. This process was repeated until we

obtained 1,500 single-complaint-based conversations of different customers. On

average, a conversation contained 6.4 total tweets (i.e., both user tweets and airline

tweets), and 2.75 airline tweets. Customers interacted with the airline to complain

mostly about flight delays, flight cancellations, mishandled baggage, in-flight ser-

vice-related issues (e.g., seats, wi-fi, meals), long queues at check-in counters,

boarding issues, rude flight attendants, and issues related to the airline’s dedica ted

customer service. To learn how these customers felt at the end of the conversation

they had with the airline on Twitter, we developed the Closed Loop Anal ytical

Social Survey (CLASS) approach to survey these customers using Twitter.

CLASS Methodology

We began by creating a dedicated Twitter account and started following each

customer, as the instantaneous Twitter notification this creates is likely to capture

the customer’s immediate attention. Next, we sent out a tweet to the customer

asking her to follow us back, so we could communicate via direct messages (DM),

keeping the conversation private and confidential. This tweet took the following

form: “Hi Amy, we are studying how airlines treat customers on Twitter. Could

you follow us so we can DM you 2 short questions? Thanks!” If the customer

followed us back indicating a preference to interact, we sent a couple of direct

messages asking two short questions: “Thx Amy. We are collecting voices on

@airline to monitor their service. We want to learn your Twitter experience with

them on December 7

” and then “(Q1) Did @airline solve your problem? (Q2)

Did your conversation with @airline make you feel better, worse, or the same?”

Upon receiving respon ses from the customer, we ended the conversation with a

thank-you note. In Figure 1, we present a more general framework for this

proposed new survey method. It can easily be adapted to any other social media

user survey such as ours.

As expect ed, not all the customers followed us back. Some customers followed

us, but did not respond to our DMs. Some customers who responded to our DMs

did not stop at providing the answers, but explained their actual experience with

the airline in detail. We offered the survey to 1,500 different customers and heard

back from 503 of them, which is a response rate of 33.54 percent. Although our

data collection methodology can be completely automated to increase the sample

size, we are currently prevented from doing so due to Twitter API rules and rate

limits.

As a result, we did the survey manually and could only obtain a few data

points each day.

Surprisingly, 53.2 percent of t he customers reported that they felt worse at the

end of the conversation with the airline on Twitter, while only 19.8 percent of the

customers felt better and 27 percent felt the same. Among the various types of

complaints present in the conversations, flight delays, cancellations, mishandled

324 GUNARATHNE, RUI, AND SEIDMANN

baggage , i n-f li gh t se r vi ce, and other operati on s-r e late d issues co ntr i but ed to

about 65 percent of the total complaints. The rest of the complaints were

process-related, including complaints related to unprofessional employees or the

airline’s dedicated customer service. Furthermore, only 10.6 percent of the

customers believed that the airline’s social media team resolved their problem.

This was more evident among the customers who felt worse at the end, as 94.36

percent of them did not perceive their problem as resolved. Moreover, 39 percent

of the customers reported handoffs, instead of having their complaint rectified by

the social media team.

Variables

Dependent variable: Our dependent variable is Emotional Outcome, which equals 1

if the customer felt better, –1 if the customer felt worse, and 0 if the customer felt the

same at the end of the conversation with the airline.

Did the

customer

follow back?

Send a tweet to the customer

mentioning the research purpose

and requesting to follow back

Follow the customer on Twitter

[Yes]

[No]

[Yes]

[Did the customer

respond to the

survey?]

Send the survey

questions to the customer

via direct message (DM)

Send a thank-you message.

Design corrective measures

Track a customer

complaint on Twitter

END

[No]

Observe the conversation between the

customer and the Customer Service Team

of the target brand on Twitter

Figure 1. The CLASS Methodology

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 325

Independent variables: The primary independent variables of interest are

the custo mer’s Klout score, whether the customer has prior experience complaining

to the airline, and the complaint type (i.e., whether the complaint is outcome-related

or process-related). Users’ Klout scores were obtained using the Klout API. To

capture users’ prior experience for complaining to the airline, we used the Twitter

API to collect the users’ historical tweets up to a maximum of 3,200, and examined

whether they contained complaints sent to the airline at least 24 hours before the

start of the conversation under consideration.

Control variables: We include a set of control variables to account for unobserved

heterogeneity at the conversation level and the customer level. The first set of

control variables includes the characteristics specific to the conversation between

the customer and the airline such as whether:

●

the airline apologized, provided an explanation, expressed gratitude, made a

handoff during the conversation, or posted consecutive tweets during the

conversation;

●

the customer had their problem solved, ended the conversation, warned the

airline about potential brand switching in future, or posted consecutive tweets

during the conversation; and

●

the airline or customer mentioned using direct mess aging.

In addition, we controlled for the airline’s average response time, and the total

number of tweets exchanged during the conversation.

The second set of control variables includes characteristics specific to the custo -

mer, such as gender, race, whether the customer had a verified Twitter account, the

age of the Twitter account, whether the customer ’s location/website/profile descrip-

tion was publicly available on Twitter, and the customer’s personality.

Controlling for the customer’s personality is important here, as personality traits

are likely to influence a customer’s evaluation of the outcome. Therefore, for each

customer, we derived the Big Five personality traits (openness, consci entiousness,

extraversion, agreeableness, neuroticism) via a lexicon-based approach, using the

customer’s past tweets as input to the linguistic inquiry and word count (LIWC)

dictionary [ 17, 30]. Past tweets could be collected for only 453 customers, as some

user timelines were private and some other profiles were no longer on Twitter. We

augment our empirical model with the derived Big Five personality traits, account-

ing for the likely omitted variable bias due to differences in customer personality.

The details on how we derived the personality traits are reported in Section A1 of the

Online Appendix.

Table 1 explains the key variables in our empi rical analysis. The summary

statistics are presented in Table 2. The correlation matrix is presented in Section

A2 of the Online Appendix.

Some of the service quality variables (e.g., handoff, apology) required manual

coding. They are presented in Table 3, along with some sample tweets used to

identify the respective constructs.

326 GUNARATHNE, RUI, AND SEIDMANN

Table 1. Definitions of Variables

Variable Description

Emotional Outcome How the customer felt at the end of the conversation (obtained from

Q1 of the survey)

(–1 = worse, 0 = the same, 1 = better)

Klout Score Klout score of the customer as obtained via the Klout API (numeric

value between 1 and 100)

Complaint Type Binary variable indicating the complaint type (1 = outcome/

operations,

e.g., flight delay/cancellation, mishandled baggage, in-flight

service, non-employee-related issues at airports, etc.)

0 = process/employees/dedicated customer service-related

e.g., rude flight attendants, longer than usual holding times in

contacting customer service, delays in responses from customer

service, etc.)

Prior Complaint

Experience

Binary variable indicating whether the customer has prior

complaining experience with the airline on Twitter (1 = Yes, 0 = No)

Handoff Binary variable indicating whether the social media team handed the

customer off to some other department (1 = Yes, 0 = No)

Problem Solved Binary variable indicating whether the airline resolved the complaint

on social media (obtained from Q2 of the survey) (1 = Yes, 0 = No)

Apology Binary variable indicating whether the airline apologized (1 = Yes, 0 = No)

Explanation Binary variable indicating whether the airline provided an explanation

(1 = Yes, 0 = No)

Gratitude Binary variable indicating whether the airline expressed its gratitude

to the customer (1 = Yes, 0 = No)

Total Tweets

Exchanged

Total number of tweets exchanged during the conversation

Average Airline

Response Time

Average of response times between airline tweets and their

respective parent user tweets, in seconds

Direct Messaging

(DM)

Binary variable indicating whether the customer or the airline

mentioned direct messaging (1 = Yes, 0 = No)

Ended by Customer Binary variable indicating whether it was the customer who ended the

conversation (1 = Yes, 0 = No)

Brand Switch

Warning

Binary variable indicating whether the customer warned the airline

about possible brand switching in the future (1 = Yes, 0 = No)

Consecutive User

Tweets

Binary variable indicating whether consecutive user tweets exist in

the conversation (1 = Yes, 0 = No)

Consecutive Airline

Tweets

Binary variable indicating whether consecutive airline tweets exist in

the conversation (1 = Yes, 0 = No)

Gender Categorical variable indicating the customer’s gender, as obtained

via Kairos (kairos.com) face detection API (1 = Female, 2 = Male,

3 = Unidentifiable)

Race Categorical variable indicating the customer’s race, as obtained via

Kairos face detection API (1 = White, 2 = Black, 3 = Other,

4 = Unidentifiable)

Verified Account Binary variable indicating whether the customer’s account is verified

on Twitter (1 = Yes, 0 = No)

(continues)

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 327

Econometric Analysis

Benchmark Model

The latent perceived satisfaction from complaining to an airline on social media for

customer i in conversation j is Y



; where



¼ β

þ D

þ C

þ ε

Here, D

refers to the vector of observable characteristics of conversation j, and

refers to the vector of observable characteristics of customer i in conversation j.

ε is an error term with cumulative distribution function G such that

GxðÞ¼1  G xðÞ.

Let Y

be an ordered outcome of whether the customer felt worse, the same, or

better at the end of the conversation with the brand, taking on the values { – 1, 0, +1}

respectively. Let τ

< τ

be unknown thresholds such that:

¼1 if Y



 τ

¼ 0 if τ



 τ

¼þ1 if Y



> τ

For simplicity, we denote by X

all independent variable s including

the conversation and custom er-related variab les as w ell as the unit for th e

Table 1. Continued

Variable Description

Customer Account

Age

Number of days since the creation of the customer’s Twitter account

Public Web Site/

Location/Profile

Bio

Binary variable indicating whether the user’s location, website, or

profile description is publicly available (1 = Yes, 0 = No)

Agreeableness Numeric value representing a person’s tendency to be

compassionate and cooperative toward others (altruism,

cooperation, trustworthiness, empathy)

Conscientiousness Numeric value representing a person’s tendency to be organized and

dependable (organization, persistence, self-assurance)

Extraversion Numeric value representing a person’s tendency to seek stimulation

in the company of others (outgoingness, sociability, energy,

positive emotions, assertiveness, sociability, talkativeness)

Neuroticism Numeric value representing a person’s tendency to experience

unpleasant emotions easily, such as anger, anxiety, and

depression

Openness Numeric value representing the extent to which a person is open to

experiencing a variety of activities (creativity, intellect, preference

for novelty)

328 GUNARATHNE, RUI, AND SEIDMANN

constant term, a nd we denote by β the vector of all coefficients including the

constant term β

. The conditional dist ribution of Y

given X

can be defined as

follows:

Pr Y

¼1jX



¼ PrðY



 τ

Þ¼G τ

 X



;

Pr Y

¼ 0jX



¼ Pr τ



 τ



¼ G τ

 X



 G τ

 X



;

Pr Y

¼ 1jX



¼ PrðY



> τ

Þ¼1  G τ

 X



The log likelihood function is given by:

τ; βðÞ¼1 Y

¼1



log G τ

 X



þ 1 Y

¼ 0



log G τ

 X



 G τ

 X



þ 1 Y

¼ 1



log 1  G τ

 X



Table 2. Summary Statistics (N = 453)

Variable Mean Std. dev.

Emotional Outcome –0.34 0.79

Klout Score 30.70 14.29

Complaint Type 0.70 0.46

Prior Complaint Experience 0.20 0.40

Handoff 0.39 0.49

Problem Solved 0.10 0.31

Apology 0.72 0.45

Explanation 0.34 0.48

Gratitude 0.25 0.44

Total Tweets Exchanged 6.57 2.74

Log of Average Airline Response Time (seconds) 1,056.60 1,141.71

Direct Messaging (DM) 0.16 0.37

Ended by Customer 0.24 0.43

Brand Switch Warning 0.23 0.42

Consecutive User Tweets 0.34 0.48

Consecutive Airline Tweets 0.01 0.09

Gender 1.81 0.73

Race 3.19 1.20

Verified Account 0.01 0.105

Customer Account Age (days) 1,514.76 632.93

Public Web Site/Location/Profile Bio 0.86 0.35

Agreeableness 5.69 1.33

Conscientiousness –1.83 0.70

Extraversion 2.42 0.98

Neuroticism 1.38 0.63

Openness –7.87 2.53

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 329

We further assume that the error term ε follows a logistic distrib ution, and we

estimate an ordered-logit model to test our hypotheses. Here we adopt the propor-

tional odds assumpt ion [18] or the parallel regression assumption such that the

relationship between each pair of outcome categories of the dependent variable is the

same. To test this assumption, we performed a Brant test [6]; it generated nonsigni-

ficant test statistics, providing evidence that the parallel regression assumption had

not been violated. The regression results are reported in Columns (1) and (2) of

Table 4.

Table 3. Sample Tweets for Manually Coded Service Quality Variables

Variable Sample Tweets

Handoff

●

@user Sonya; have you reached out to our Customer Relations

team? They can help you with past date travel issues.

●

@user We’re sorry to hear this. Contact our Central Baggage

Services for assistance with this.

●

@user Our agents will help you with available options. Please see

them as soon as you can

Apology

●

@user No one likes delays; especially on their birthday☺. We

apologize for the inconvenience.

●

@user We hear your frustration. Please accept our apology.

●

@user We expect our team to always be cordial at all times; Ashley.

Our apologies that you experienced otherwise.

Explanation

●

@user We show the equipment is out of service; Justin. Safety of

our customers and crew is always our top priority.

●

@user We’re sorry your flights delayed. There are major Air Traffic

Control delays beyond our control.

●

@user Weather can back up Air Traffic Control flows once things

get moving again; Bianca. Our apologies for the inconvenience.

Gratitude

●

@user We’re glad we have you to Sacramento. Have a good rest of

your Sunday and thanks for flying with us.

●

@user We appreciate you feedback and thank you for the kind

words for our staff.

●

@user Our goal is to provide exceptional customer service. We’re

sorry we missed the mark today. We appreciate your loyalty.

Direct Messaging

(DM)

●

@user Please send us a DM with your bag tag number.We’ll take

a look.

●

@user We’re very sorry to hear this; Sharina. Did you file a report at

<URL>? If so; DM your CR file number.

●

@user Rachel; please DM your record locator or Baggage Report

number. We’d like to check on that for you.

Brand Switch

Warning

●

@airline Never Again! 4 flight changes lost luggage. And my

UNACCOMPANIED minor still not where she needs to be. 1st and

last <airline> flight.

●

@airline . . . Worst customer service. My mom has been through hell

with you. Now both of us won’t fly with you. @airline2 wins again

●

Stuck on a plane for over an hour; gets told to wait at least another

20mins before further info? Never EVER flying with @airline again

330 GUNARATHNE, RUI, AND SEIDMANN

Estimation Results

From Table 4, we see that Klout Score is positive and statistically significant

(0.032, p < 0.01). In terms of magnitude, for a one-unit increase in Klout Score,

the odds of feelin g bett er i ncrease by a factor of 1. 032 ( 3.2 p ercent ) mor e tha n

Table 4. Ordered Logistic Regression: Benchmark Model

Variable

(1)

Ordered Logit

Coefficient

(2)

Ordered Logit

Odds Ratio

Klout Score 0.032*** (0.008) 1.032*** (0.008)

Complaint Type 0.643*** (0.232) 1.903*** (0.441)

Prior Complaint Experience –0.528** (0.256) 0.590** (0.151)

Handoff –0.465** (0.213) 0.628** (0.134)

Problem Solved 1.381*** (0.330) 3.981*** (1.314)

Apology 0.295 (0.233) 1.343 (0.313)

Explanation –0.178 (0.219) 0.837 (0.184)

Gratitude 0.609*** (0.226) 1.838*** (0.415)

Total Tweets Exchanged –0.145*** (0.048) 0.865*** (0.042)

Log of Average Response Time 0.018 (0.105) 1.018 (0.107)

Direct Messaging (DM) 0.355 (0.277) 1.426 (0.395)

Ended by Customer –0.407* (0.245) 0.666* (0.163)

Brand Switch Warning –0.659** (0.258) 0.518** (0.133)

Consecutive User Tweets 0.283 (0.247) 1.327 (0.328)

Consecutive Airline Tweets 0.056 (1.016) 1.058 (1.075)

Gender—Male (base: Female) –0.117 (0.232) 0.890 (0.206)

Gender—Unidentifiable (base: Female) 0.203 (0.784) 1.226 (0.961)

Race—Black (base: White) –0.323 (0.740) 0.724 (0.536)

Race—Other (base: White) 0.399 (0.525) 1.490 (0.781)

Race—Unidentifiable (base: White) –0.053 (0.316) 0.949 (0.300)

Verified Account –1.478 (0.976) 0.228 (0.223)

Customer Account Age –0.000 (0.000) 1.000 (0.000)

Public Web Site/Location/Profile Bio –0.426 (0.332) 0.653 (0.217)

Agreeableness –0.104 (0.142) 0.901 (0.128)

Conscientiousness 0.514* (0.274) 1.673* (0.459)

Extraversion 0.033 (0.191) 1.033 (0.197)

Neuroticism –0.055 (0.318) 0.946 (0.301)

Openness –0.141 (0.097) 0.869 (0.084)

Cut 1 Constant –0.174 (0.998) 0.840 (0.839)

Cut 2 Constant 1.312 (1.000) 3.713 (3.714)

Observations 453

Log likelihood –404.155

AIC 868.340

BIC 991.786

Notes: Standard errors are in parentheses. Significance: ***p < 0.01; **p < 0.05; *

p <

0.1.

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 331

the odds of feeling the same or feeling worse.

This finding suggests that as

the Klout score incre ases, there is a corresponding increase in the probability

of the customer feeling better at the end of a conversation with the airline

on social media, thereby providing support for the Social Influence

Hypothesis (H1).

To better evaluate how the probabilities of each emotional outcome changes as

Complaint Type and Prior Complaint Experience vary, we generate their respective

predicted probabilities while keeping the rest of the variables at their means. The

results are reported in Table 5.

From Table 4, we also see that Prior Complaint Experience is negative and

statistically significant (–0.528, p < 0.05). In terms of magnitude, having prior

complaint experience decreases the odds of feeling better by a factor of 0.59 (41

percent), than the odds of feeling the same or feeling worse. Moreover, the predicted

probabilities (Table 5) indicate that there is 64.4 percent chance that the customer

feels worse if the customer has had prior complaint experience, as opposed to a 51.7

percent probability of feeling worse if the customer has had no such experience.

Thus, our findings suggest that a customer who previously complained is more

likely to feel worse than to feel the same or better, thereby providing support for the

Prior Complaint Experience Hypothesis (H2).

Finally, we see in Table 4 that Complaint Type is positive and statistically

significant, (0.643, p < 0.01). In terms of magnitude, customers with outcome-

related complaints are more likely to feel better than those with process-related

complaints by a factor of 1.903 (90.3 percent) more than the odds of feeling the

same or feeling worse. As per the predicted probabilities (Table 5), there is a 65

percent chance that the customer feels worse at the end when the complaint is

process-related, and just a 49.4 percent chance when the complaint is outcome-

related. Accordingly, our findings suggest that process-related complaints are

less likely to make a customer feel better at the end of a conversation with an

airline on social media, thereby providing support for the Complaint Type

Hypothesis (H3).

Table 5. Predicted Probabilities

Variable Emotional outcome Probability at 0 Probability at 1

Complaint Type Worse 0.650*** (0.045) 0.494*** (0.031)

Same 0.242*** (0.030) 0.318*** (0.026)

Better 0.109*** (0.021) 0.188*** (0.022)

Prior Complaint

Experience

Worse 0.517*** (0.029) 0.644*** (0.053)

Same 0.309*** (0.025) 0.245*** (0.034)

Better 0.175*** (0.020) 0.111*** (0.025)

Notes: Standard errors are in parentheses. Significance: ***p < 0.01

332 GUNARATHNE, RUI, AND SEIDMANN

We found that several other variables were statistically significant. For example,

Handoff is negative and statistically significant (–0.465, p < 0.05). In terms of

magnitude, handing the customer off to some other department, rather than the

social media team taking care of the customer, decreases the odds of feeling better

by a factor of 0.628 (37.2 percent), as compared with the odds of feeli ng the same or

feeling worse. Thus, our findings suggest that a complaining customer is more likely

to feel worse, if the airline’s social media team hands that custo mer off to some other

department.

Furthermore, Problem Solved is positive and statistically significant (1.381, p <

0.01). In terms of magnitude, a customer’s perception of the problem as being fixed

by the social media team increases the odds of feeling better by a factor of 3.981

(298.1 percent) as compared to the odds of feeling the same or feeling worse. Thus,

Problem Solved has the largest positive effect on a customer’s emotional outcome at

the end of the complaining encounter. This makes sense because one of the

responses that a customer expects when a problem arises is a “fair fix” that at

least returns the customer to the starting point before the service failure [9]. Our

analysis reveals that this could mean a variety of possible resolutions including

replacement, refund, repair, discounts, corrections, and appropriate remedial action.

In the airline industry, these may include paying damaged baggage claims, rebook-

ing, hotel and food vouchers in case of flight delays or cancellations, free miles,

refunds, customer status upgrades, reporting unprofessional employees to manage-

ment, and so on.

Robustness Test: Nonresponse Bias

Although we sent the survey to 1,500 Twitter users who had a complaint-based

conversation with the airl ine on Twitter, we heard back from only about a third

of them. Our main analysis was primarily based on the users who responded to

the survey. If the unobserva bles that (1) determine cust omers ’ satisfaction after

the i nteract ion with the b rand on social media, and (2) increase the likelihood

of not responding to the survey are correlated, then our model suffers from

nonresponse bias. This would have been the case if users who did not respond

to our survey were significantly different from those who res ponded to the

survey. To correct for this, we randomly selected 200 users who did not

respond to our su rvey and introduced these users into the ma in data se t. Then

we formulated an ordered probit model for customers’ emotional outcome, with

selection on whether they responded to the survey. As the outcome covariates,

we u sed the same set o f variables us ed in the benchmark model. We particu-

larly assume that how influential the customer is on socia l media (Klout score),

gender, race, complaint type, customer account age, whether the customer had

his or her location/website/profile description publicly available on Twitter, and

whether the custo mer warned the airli ne a bout s witching to another brand,

affect the selection. Then we employ the Heckman procedure to estimate our

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 333

model. The results, w hich are prese nted in Table 6, are qualitatively the same

as the benchmark model. Moreover, error correlati on was not statistically

significantatthep < .05 level, suggesting that nonresponse bias is not likely

a ma jor concern.

Table 6. Robustness Test: Heckman Selection

Variable

(1)

Ordered Probit Model

(Emotional Outcome)

(2)

Selection Model

(Responded to Survey)

Klout Score 0.016*** (0.005) -0.001 (0.004)

Complaint Type 0.351*** (0.130) 0.061 (0.119)

Prior Complaint Experience –0.287** (0.140)

Handoff –0.266** (0.119)

Problem Solved 0.698*** (0.227)

Apology 0.151 (0.125)

Explanation –0.094 (0.117)

Gratitude 0.317** (0.133)

Total Tweets Exchanged –0.072*** (0.027)

Log of Average Response Time 0.018 (0.061)

Direct Messaging (DM) 0.174 (0.152)

Ended by Customer –0.221* (0.132)

Brand Switch Warning –0.273 (0.169) 0.284** (0.137)

Consecutive User Tweets 0.147 (0.136)

Consecutive Airline Tweets 0.026 (0.576)

Gender—Male (base: Female) –0.143 (0.129) – 0.243* (0.134)

Gender—Unidentifiable (base: Female) –0.421 (0.494) – 1.555*** (0.565)

Race—Black (base: White) 0.016 (0.425) 0.625 (0.550)

Race—Other (base: White) 0.092 (0.315) –0.420 (0.280)

Race—Unidentifiable (base: White) –0.032 (0.172) –0.126 (0.170)

Verified Account –0.791 (0.558)

Customer Account Age 0.00002 (0.000) 0.000** (0.000)

Public Web Site/Location/Profile Bio –0.218 (0.191) 0.033 (0.156)

Agreeableness –0.063 (0.071)

Conscientiousness 0.283* (0.161)

Extraversion 0.024 (0.105)

Neuroticism –0.040 (0.175)

Openness –0.082 (0.051)

Cut 1 Constant 0.289 (0.615)

Cut 2 Constant 1.063* (0.585)

Constant 0.549*** (0.210)

atanh ρ (ρ = error correlation) 0.971 (0.891)

Observations

653.00 653.00

Log

likelihood –773.21 –773.21

AIC 1,630.43 1,630.43

BIC 1,818.65 1,818.65

Notes: Standard errors are in parentheses. Significance: ***p < 0.01; **p < 0.05; *p < 0.1.

334 GUNARATHNE, RUI, AND SEIDMANN

Extension

Given that 53.2 percent of the customers in our sample felt worse at the end of their

conversations with the airline on Twitter, one natur ally questions whether doing

customer service on social media is worthwhile for a brand. As an extension to our

study, we investigated this question by doing a short survey among customers who

did not receive any response from the airline after compl aining to the airline on

Twitter. Again, we followed the CLASS approach; the survey questionnaire took the

following form: “Our records show that @airline did not respond to your complaint

on Twitter on January 14

. We want to learn your experience. (Q1) Rate your

overall satisfaction regarding the way @airline handled your complaint on Twitter:

Very Dissatisfied/Dissatisfied/Neither Satisfied Nor Dissatisfied/Satisfied/Very

Satisfied (Q2) What is the likelihood that you would use Twitter again to complain

to @airline in the future? Very Unlikely/Unlikely/Not Sure/Likely/Very Likely.”

We sent out 222 survey requests and heard back from 38 people.

Of the

respondents, 92.1 percent reported that they were either very dissatisfied or dissa-

tisfied regarding the way the airline handled their complaint on Twitter (i.e., by

choosing not to respond). Surprisingly, 81.58 percent of these respondents claimed

that they are very likely or likely to use Twitter to complain to the airline in the

future. One possible explanation of these results is that customers whose complaints

are ignored by a brand are more motivated to punish the brand by publicly

complaining on social media.

Considering the high percentage of people unhappy with the airline for not

responding to their complaint, and their intention to keep complaining on social

media, it seems that brands would still be much better off investing in social media

customer service, even though it does not always effectively transform a disgruntled

customer into a happy one. Moreover, recent research [38] has suggested that

emotionally charted tweets tend to be retweeted more often and more quickly

compared to neutral ones. Simply put, brands today have no option but to listen to

and engage with their customers on a real-time basis in order to succeed. The

democratization of media by social media platforms like Twitter has effectively

raised the bar for customer service and wi ll ultimately lead to more transparency

and better service.

Managerial Implications

Our findings have important implications for the brands striving to harness the

power of social media to deliver custo mer service.

First, our empirical test of the Social Influence Hypothesis (H1) indicates that

complaining customers with a higher Klout score are more likely to feel better at the

end of a conversation with a brand on social media. As we argued earlier, this may

be because customers with greater social influence receive preferential customer

service from the airline, or simply because socially influential customers are happier

and emot ionally more stable individuals in general. If it is social-influence-based

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 335

preferential customer service that is mainly at work, the result would suggest that a

brand’s influence-based preferential treatment pays off during complaint resolution,

at least for those customers who are treated better. However, given the controversial

nature of this practice, and its implications on perceptions of fairness, brands should

carefully examine the drivers of this practice within their social media teams and act

accordingly.

On the other hand, if it is the happy and emotionally more stable nature of socially

influential individuals that makes them more satisfied about the complaint resolution

received, then this suggests that those customers with greater online social influence

are simply easier to please. Hence, the brand can customize its marketing strategies

accordingly, in targeting this particular customer segment. On the other hand, the

result also reveals a challenge in achieving high satisfaction with less-influential

customers. Nevertheless, the extension of our study shows that it is imperative for

brands to monitor and engage with complaining custo mers on social media even

though the customers may still be dissatisfied after their complaint resolution

experience on social media.

Second, our empirical test of the Prior Complaint Experience Hypothesis (H2)

indicates that custo mers with prior complaint experience with the airline are more

likely to feel dissatisfied about their social media interactions with that brand.

Hence, it may be worthwhile to customize the response to a customer complaining

on social media, based on the customer’s social media-complaint history, instead of

treating each complaint as completely new. Although a brand may keep track of all

historical complaints received from each customer via the traditional complaint

management process, whether the same is true for complaints received via social

media is not evident. Thus, a direct implication for practice would be to keep track

of all complaints from each custo mer and train the social media team to handle those

customers accordingly.

Third, our empirical test of the Com plaint Type Hypothesis (H3) shows that

customers complaining about process-related issues (e.g., unprofessional empl oyees

or dedicated customer service) are more likely to feel worse at the end than those

who complained about outcome or operations-related issues. Most customers under-

stand that things may sometimes go wrong in airline operations, but when it comes

to issues of employee attitude, they find it harder to forgive. Therefore, it is

important to devise a separate response strategy to manage process-related com-

plaints on social media. For example, in addition to reassuring the customer that

action will be taken against the reported unprofessionalism , it may also be worth-

while to cheer the customer with some sort of compensation.

Another important implication for practice would be the pressing need to empower

the social media team. We find that customers who were handed off to other

departments are more likely to feel worse at the end. It appears that customers

tend to perceive a service handoff as a way of “passing the buck,” rather than the

social media team’s lack of ability to resolve the complaints. The reasons for the low

problem resolution rate and the high handoff rate may be a lack of technology

infrastructure, training opportunities, and budget available to social media teams.

336 GUNARATHNE, RUI, AND SEIDMANN

Therefore, a careful social media investment strategy should be defined at the

corporate level, enabling seamless integration between the social media team and

the dedicated customer service of the brand. For instance, rather than letting the

social media team ask the customer to contact the baggage claims department

regarding lost baggage, the social media team should be able to access the relevant

corporate databases to provide more complete and worthwhile social media customer

service. Furthermore, social media teams should be given continuous and mandatory

opportunities to learn to provide high-quality complaint resolutions faster.

Conclusion

As consumers become more empowered by social media, companies are under

increasing pressures to improve not only the core value they deliver to their

customers but also everything their customers experience [24]. To succeed in such

a competitive environment, companies need to place customer experience into one

holistic view for their present customers as well as for past and future customers. A

key component of building a successful customer experience is establishing a

systematic approach to ongoing listening to customers and to their perceptions of

the way in which the compa ny addresses their concerns. There are several common

approaches to doing that, including focus groups, mystery shoppers, advisory panels,

periodic surveys, and transactional surveys. Transactional surveys have the advan-

tage of getting feedback while the service experience is still fresh, and they allow the

company to act quickly if it detects a major service gap. This approach motivated the

development of our Closed Loop Analytical Social Survey (CLASS) methodology,

which leverages the popularity of social media as a novel and rapid way to conduct

such transactional surveys. At a broad level, this research can be viewed as an

example from service science, management, and engineering (SSME) which is an

important multidisciplinary area [2].

Our study has some limitations. It assumes that a customer’s satisfaction regarding

his or her interaction with the brand on social media genuinely reflects the custo-

mer’s true emot ional status at the end of the conversation with the airline. This

approach may not be perfect for at least two reasons. First, although we maintained

the minimum possible interval between the end of the conversation on Twitter and

the survey offer, this gap may psychologically cause customers to overestimate or

underestimate how they actually felt at the end of the interaction. Second, some

individuals, such as people with greater online social influence, may be particularly

cautious in their social media interactions with brands, such that their conversations

on social media may not reveal their actual preferences, while average customers

may choose to express their concerns freely. Although we are unable to determine

the extent to which these factors may affect our findings, their existence could

undermine the importance of studies such as this one in determining the drivers of

happier complaint resolutions on social media.

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 337

Acknowledgments: An early version of this study was previously circulated as “What Drives

Successful Complaint Resolutions on Social Media? Evidence from the Airline Industry.” The

authors thank the JMIS coeditors Rob Kauffman, Rajiv Dewan, Thomas Weber, Eric Clemons,

the HICSS minitrack chairs, Jie Zhang, Yabin Jiang, and all the participants of the HICSS

minitrack on “Integrating Business Operations, Information Technologies, and Consumer

Behavior” in the Organizational Systems and Technology Track, for useful discussion and

comments.

Supplemental File

Supplemental data for this article can be found on the publisher’s website at

10.1080/07421222.2017.1334465

NOTES

1. The Big Five personality traits refer to openness, conscientiousness, extraversion,

agreeableness, and neuroticism/degree of emotional stability; see [17] for details.

2. To automate the process, one needs to apply for special permission and be approved by

Twitter.

3. See Greene and Hensher [18], for details on the interpretation of ordered logit

coefficients.

4. The small sample size is due to both the low response rate and limiting rules imposed by

Twitter.

REFERENCES

1. Allon, G., and Zhang, D.J. Managing service systems in the presence of social net-

works. Available at: http://ssrn.com/abstract=267313 (accessed on October 12, 2015).

2. Bardhan, I.R.; Demirkan, H.; Kannan, P.K.; Kauffman, R.J.; and Sougstad, R. An

interdisciplinary perspective on IT services management and service science. Journal of

Management Information Systems, 26, 4 (April 2010), 13–64.

3. Bell, P.A. Affective state, attraction, and affiliation: Misery loves happy company too.

Personality and Social Psychology Bulletin, 4 (October 1978), 616–619.

4. Bitner, M.J.; Booms, B.H.; and Tetreault, M.S. The service encounter: Diagnosing

favorable and unfavorable incidents. Journal of Marketing, 54, 1 (January 1990), 71–84.

5. Boshoff, C. RECOVSAT: An instrument to measure satisfaction with transaction-spe-

cific service recovery. Journal of Service Research, 1, 3 (February 1999), 236–249.

6. Brant, R. Assessing proportionality in the proportional odds model for ordinal logistic

regression. Biometrics (December 1990), 1171–1178.

7. Conlon, D.E., and Murray, N.M. Customer perceptions of corporate responses to

product complaints: The role of explanation. Academy of Management Journal, 39,4

(August 1996), 1040–1056.

8. Cui, G.; Wong, M.L; and Wan, X. Cost-sensitive learning via priority sampling to

improve the return on marketing and CRM investment. Journal of Management Information

Systems, 29, 1 (July 2012), 341–374.

9. Davidow, M. The bottom line impact of organizational responses to customer com-

plaints. Journal of Hospitality and Tourism Research, 24, 4 (November 2000), 473–490.

10. Davidow, M. Organizational responses to customer complaints: What works and what

doesn’t. Journal of Service Research, 5, 3 (February 2003), 225–250.

11. Day, R.L. Modeling choices among alternative responses to dissatisfaction. Advances in

Consumer Research, 11, 1 (1984), 496–499.

338 GUNARATHNE, RUI, AND SEIDMANN

12. Del Río-Lanza, A. B.; Vázquez-Casielles, R.; and Díaz-Martín, A. M. Satisfaction with

service recovery: Perceived justice and emotional responses. Journal of Business Research,

62, 8 (August 2009), 775–781.

13. Estelami, H. Competitive and procedural determinants of delight and disappointment in

consumer complaint outcomes. Journal of Service Research, 2, 3 (February 2000), 285–300.

14. Feingold, A. Happiness, unselfishness, and popularity. Journal of Psychology, 115

(September 1983), 3–5.

15. Folkes, V.S. Consumer reactions to product failure: An attributional approach. Journal

of Consumer Research, 10, 4 (March 1984), 398–409.

16. Goel, V. G.M. uses social media to manage customers and its reputation. New York

Times. March 23, 2014. Available at: http://www.nytimes.com.

17. Goldberg, L.R. The structure of phenotypic personality traits. American Psychologist,

48, 1 (January 1993), 26–34.

18. Greene, W.H., and Hensher, D.A. Modeling Ordered Choices: A Primer. Cambridge:

Cambridge University Press, 2010.

19. Gronroos, C. Service quality: The six criteria of good perceived service quality. Review

of Business, 9, 3 (January 1988), 10–13.

20. Gunarathne, P.; Rui, H.; and Seidmann, A. When social media delivers customer

service: Differential customer treatment in the airline industry. Working Paper, Simon

Business School, University of Rochester, 2017.

21. Homans, G.C. Social Behavior: Its Elementary Forms. New York, NY: Harcourt, Brace

and World, 1961.

22. Karr, R., and Blohowiak, D.W. The Complete Idiot’s Guide to Great Customer Service.

New York, NY: Alpha Books, 1997.

23. Lacey, R.; Suh, J.; and Morgan, R.M. Differential effects of preferential treatment levels

on relational outcomes. Journal of Service Research, 9, 3 (February 2007), 241–256.

24. Lemon, K.N. The art of creating attractive consumer experiences at the right time: Skills

marketers will need to survive and thrive. GfK Marketing Intelligence Review, 8, 2 (November

2016), 44–49.

25. Levesque, T.J., and McDougall, G.H. Customer dissatisfaction: The relationship

between types of problems and customer response. Canadian Journal of Administrative

Sciences, 13, 3 (September 1996), 264–276.

26. Lucas, R.E.; Diener, E.; Grob, A.; Suh, E. M.; and Shao, L. Cross-cultural evidence for

the fundamental features of extraversion. Journal of Personality and Social Psychology, 79,3

(September 2000), 452–468.

27.

Lyubomirsky, S.; King, L.; and Diener, E. The benefits of frequent positive affect: Does

happiness lead to success? Psychological Bulletin, 131, 6 (November 2005), 803–855.

28. Marston, W.M. Emotions of Normal People. New York, NY: Harcourt Brace, 1928.

29. Mervielde, I., and de Fruyt, F. The Big Five personality factors as a model of the

structure of children’s peer nominations. European Journal of Personality, 14, 2 (March

2000), 91–106.

30. Pennebaker, J.W.; Boyd, R.L.; Jordan, K.; and Blackburn, K. The development and

psychometric properties of LIWC2015. LIWC development manual, University of Texas at

Austin, 2015.

31. Schlereth, C.; Barrot, C.; Skiera, B.; and Takac, C. Optimal product-sampling strategies

in social networks: How many and whom to target? International Journal of Electronic

Commerce, 18, 1 (October 2013), 45–72.

32. Singh, J. Voice, exit, and negative word-of-mouth behaviors: An investigation across

three service categories. Journal of the Academy of Marketing Science, 18, 1 (January 1990),

1–15.

33. Singh, J., and Wilkes, R.E. When consumers complain: A path analysis of the key

antecedents of consumer complaint response estimates. Journal of the Academy of Marketing

Science, 24, 4 (September 1996), 350–365.

34. Smith, A.K.; Bolton, R.N.; and Wagner, J. A model of customer satisfaction with

service encounters involving failure and recovery. Journal of Marketing Research, 36

(August 1999), 356–372.

CUSTOMER COMPLAINTS ON SOCIAL MEDIA 339

35. Söderlund, M.; Liljander, V.; Gummerus, J.; Hellman, P.; Lipkin, M.; Oikarinen, E.L.;

Sepp, M.; and Liljedal, K.T. Preferential treatment in the service encounter. Journal of Service

Management, 25, 4 (August 2014), 512–530.

36. Stauss, B. The dimensions of complaint satisfaction: Process and outcome complaint

satisfaction versus cold fact and warm act complaint satisfaction. Managing Service Quality:

An International Journal, 12, 3 (June 2002), 173–183.

37. Stephens, N., and Gwinner, K.P. Why don’t some people complain? A cognitive-

emotive process model of consumer complaint behavior. Journal of the Academy of

Marketing Science, 26, 3 (July 1998), 172–189.

38. Stieglitz, S., and Dang-Xuan, L. Emotions and information diffusion in social media-

sentiment of microblogs and sharing behavior. Journal of Management Information Systems,

29, 4 (April 2013), 217–248.

39. Strauss, J., and Hill, D.J. Consumer complaints by e-mail: An exploratory investigation

of corporate responses and customer reactions. Journal of Interactive Marketing, 15,1

(January 2001), 63–73.

40. Susarla, A.; Oh, J.H.; and Tan, Y. Influentials, imitables, or susceptibles? Virality and

word-of-mouth conversations in online social networks. Journal of Management Information

Systems, 33, 1 (January 2016), 139–170.

41. Tax, S.S.; Brown, S.W.; and Chandrashekaran, M. Customer evaluations of service

complaint experiences: Implications for relationship marketing. Journal of Marketing, 62

(April 1998), 60–76.

42. Van Der Linden, D.; Scholte, R.H.; Cillessen, A.H.; te Nijenhuis, J.; and Segers, E.

Classroom ratings of likeability and popularity are related to the Big Five and the general

factor of personality. Journal of Research in Personality, 44, 5 (October 2010), 669–672.

43. Walster, E.; Berscheid, E.; and Walster, G.W. New directions in equity research. Journal

of Personality and Social Psychology, 25, 2 (February 1973), 151–176.

44. Xia, L. Effects of companies’ responses to consumer criticism in social media.

International Journal of Electronic Commerce, 17, 4 (July 2013), 73–100.

340 GUNARATHNE, RUI, AND SEIDMANN

CopyrightofJournalofManagementInformationSystemsisthepropertyofTaylor&Francis

Ltdanditscontentmaynotbecopiedoremailedtomultiplesitesorpostedtoalistserv

withoutthecopyrightholder'sexpresswrittenpermission.However,usersmayprint,

download,oremailarticlesforindividualuse.