5.3.1. Step 1: set up listening measures

One of the first difficulties related to the management of the customer’s voice online is to successfully set up systematic listening measures. After going through a social media image crisis in 2010 about the use of palm oil in its chocolate bars, the Nestlé Group set up the Digital Acceleration Team, a team dedicated to listening to customer conversations on social media. The content captured and analyzed comes from platforms such as Facebook, YouTube, Twitter and LinkedIn, and represents nearly 50 million sources in 17 different languages [LOR 13]. Unlike more traditional marketing measures for obtaining customer opinions, such as surveys and interviews, listening to online conversations provides access to spontaneous messages in real time and at a lower cost [TIR 14]. Tirunillai and Tellis [TIR 14] analyzed nearly 350,000 Internet users’ opinions in five business sectors to identify the criteria that a brand should take into account to measure the quality of an opinion, its importance and its ability to support decision-making.

Systematic data analysis is complex because textual data are presented in an unstructured form, which requires, for example, the elimination of characters or words unknown in the target language and which have no informational value. The systematic listening to the customer’s voice online must therefore be based on a set of criteria:

1. 1) Location: the interpretation and analysis of conversations must be contextualized because each platform has its own standards and formats.
2. 2) Valence: listening measures should allow for sentiment analysis, which consists of identifying emotional markers and the valence of adjectives to categorize positive and negative conversations.
3. 3) Volume: it is also important to analyze the frequency and intensity of conversations about the brand or product.
4. 4) Content: the systematic analysis of conversations should eventually make it possible to identify the different themes, attributes and characteristics of the brands and/or product mentioned in the message.

In their study, Tirunillai and Tellis [TIR 14] proposed the use of the Herfindahl index of concentration to analyze online conversations. This index represents an average concentration of quality dimensions among all Internet users’ conversations analyzed for the brand. For example, if the quality of a smartphone, represented by the durability of the battery, is the dimension most often mentioned by Internet users and other quality dimensions (i.e. design, reactivity) are mentioned only very rarely, then the index makes it possible to identify this concentration on a single quality criterion. Tirunillai and Tellis’ research highlights the difficulty of analyzing massive and unstructured data and emphasizes the need for upstream work to be able to exploit them.

However, the big data phenomenon has encouraged companies to invest more resources in developing their data processing and systematic analytical skills. The data scientist profession, or being an expert in big data analysis, represents a rare skill that is highly sought after by companies [DAV 12]. Nowadays, automated data analysis can also be done by using statistical approaches with “a vast set of tools for understanding data” [JAM 13]. Tools such as those offered by Monkeylearn (https://monkeylearn.com) allow systematic text analysis based on machine learning. Other customer listening measures are also deployed in marketing departments to monitor and measure their online reputation, including social monitoring software (e.g. Hootsuite, Buffer, Buzzsumo, Linkfluence). These tools have become more accessible in recent years, both in terms of ergonomics and price, which has accelerated their adoption within companies. Their functionalities include the ability to build indicators related to Internet user engagement and the impact of digital marketing campaigns, the ability to detect trends in conversations, identify influencers or detractors and create user or community profiles.

In order to better understand the relationships and interdependencies between online customer conversations, social network analysis is another methodological approach to visualize communications in the form of networks, nodes and links between more or less influential Internet users. Figure 5.1 illustrates a social network analysis applied to the Volkswagen scandal, following the manipulation of polluting emission values of its cars. Via the NodeXL tool for Microsoft Excel [SMI 10], tweets using the #VWGate hashtag were exported and analyzed using a classification approach. The graphical visualization of Twitter communications as networks shows that a large number of users mentioned the hashtag #VWGate, but these tweets elicited very few reactions such as retweets for example (these are all the points in the lower right-hand corner of Figure 5.1). This figure also shows more or less important groupings (to the left of Figure 5.1) with a few main actors whose tweets have been widely used by other users. In the case of #VWGate, these include the official accounts of the major national and international media.

Smith et al. [SMI 14] propose a typology of six possible types of social networks (see Figure 5.2).

The first type of social network, the polarized crowd, consists of two large and dense groups with many interactions, but connections between the two groups are almost non-existent. This type of network often manifests itself in discussions on highly controversial topics and often involves political discussions. The second type, the tight crowd, is similar to the polarized crowd, but the subgroups are strongly linked and interact with each other through intermediaries or connectors who act as bridges. Brand clusters, the third type, are very frequently observed among companies that have implemented a marketing strategy based on social media: the network is therefore not very dense and presents a large number of isolated individuals. Due to the notoriety of large companies, many Internet users express themselves by mentioning the brand, but these users are only rarely connected to other people mentioning the same brand. In the configuration of community groups, medium-sized subgroups tend to emerge and interact with each other. The broadcast network successfully illustrates the case of Volkswagen Gate: several groups exist, but form a star network with a central actor in the middle (the hub) – which very often corresponds to the mainstream media or a major influencer – and users who repeat (or retweet) messages from the central hub (the spokes). Users thus serve as an audience and very often have few interactions with each other. In the case of a support network, relationships are bidirectional. This is the case for companies, particularly in the service sector, that have a strong online presence and use social media to listen to and answer customers’ questions and complaints. Box 5.1 presents the value of using social media, and in particular the microblogging Twitter site, as an after-sales service tool in the banking sector.

To conclude, social network analysis makes it possible to identify and classify interactions in a visual form while restoring the dynamics of discussion between actors.

5.3.2. Step 2: respond to online customers

The customer’s voice raises another challenge for companies, namely the response to comments made online, whether in terms of speed, content to be transmitted or engagement to be demonstrated. For example, communications that have a direct mention (@username) should normally be processed quickly because Internet users who address companies in this way generally want their problem resolved. The customer listening system includes the analysis of direct messages, but also the mentions of brand’s or the company’s product names as well as the mention of events specific to product categories. Figure 5.4 provides a framework for analyzing response strategies along two axes: (1) the extent of customer listening and (2) the main objectives of interaction.

When a user directly messages a company’s account (brand or after-sales service), they often expect a quick, simple and efficient resolution (Figure 5.4, (1). The interaction then focuses on problem solving. According to a study conducted by Lithium Technologies in 2013, more than half of Twitter users expect a response to their tweet in less than 1 h [LIT 11]. However, in an analysis of the response behaviors of large U.S. companies, Einwiller and Steilen [EIN 15] show that companies respond to only 47% of the 5,000 Twitter complaints examined in their study. Moreover, in the event of a response, companies react on average after 7 hours; and only a quarter of these responses were published within 18 min of the Internet user’s first tweet [EIN 15]. In a study with more than 800 respondents, Sparks and colleagues [SPA 16] demonstrate the importance of responsiveness in digital environments: the faster the company responds, the more the Internet user will consider the company trustworthy and capable of caring about the well-being of its customers. On the other hand, while the slowness of response to negative messages can lead to dissatisfaction, the speed of response seems to be a main factor because it does not automatically increase Internet user satisfaction [MIN 15]. It is, therefore, in the interest of companies to respond quickly to complaints in order to avoid impatient Internet users talking about them to their entourage or looking for other communication channels to share their frustration [LIT 11]. At the second level (see Figure 5.4, (2), it is a matter of examining all conversations that include the names of the company’s brands and/or products. In this case, Internet users do not address the company directly to solve their problem, but will tend to use branded hashtags to improve the visibility of their message. The response strategy to be implemented must then be proactive: the company must respond to messages posted on the Internet spontaneously [NPV 12].

Finally, the approach may also include listening to the customer’s voice at events that are related to the product category or of interest to the company and its products (Figure 5.4, (3). In this situation, the response must allow the anticipation of market trends and the detection of prospects and/or opportunities. The objective is to link the brand to potentially positive events for the company. Thus, it is important for a company to respond to positive feedback in order to increase the future engagement of satisfied customers [CHS 15].

Figure 5.4 presents three levels of response, but does not specify the strategy and communication style to be adopted to respond to customers. English marketing literature has recently developed the concept of customer care on the Internet (Webcare) to emphasize the attention to be paid when interacting with Internet users [DEN 15, SCH 15]. For van Noort and Willemsen [VAN 12, p. 133], Webcare refers to “the act of engaging in online interactions with (complaining) consumers, by actively searching the web to address consumer feedback (e.g. questions, concerns and complaints)”. In the event of a complaint on social media, the best strategy is to respond to the Internet user [BOB 05] because the latter is still waiting for a response after having formulated and published an online complaint [BRA 09, LEE 10]. The advantage of having a response strategy is twofold: first, complaint management prevents the customer from expressing themself to other individuals via eWOM and prevents them from changing companies. Second, it allows the company to protect its reputation and maintain a good image toward future customers who consult and read content generated by other users before making their decisions. According to Scott and Lyman [SCO 68, p. 46], the company can then respond by providing explanations to their customers, i.e. developing a communication method to “explain unanticipated or untoward behavior”. Depending on the type of explanations and the stages of the complaints process [FEL 81], companies therefore have different responses they can choose from (see Figure 5.5).

According to the proposed typology of response strategies [IBE 87, BOB 05], two approaches can be implemented. The first strategy combines the company’s responses to challenge the occurrence of a negative event. The company then refuses to accept that a problem has occurred (in the case of denial) and tries to legitimize its actions [BRA 09] while minimizing the seriousness of the problem [BOB 05] in the case of justification. When the company challenges the seriousness of the incident for the claimant by demonstrating that other individuals have suffered more serious consequences, it implements a reference-based response strategy by making a social comparison with another group of individuals [TEB 87]. This first group of responses often presents an attempt by the company to mitigate the severity of an incident and thus modify the individual’s perceptions.

Due to the public nature of the communications, contesting either the seriousness or the occurrence of an incident can be risky and aggravate the situation, particularly when other consumers concerned by the incident also express themselves and thus amplify the scope of the conflict [BOB 05]. Therefore, when the company accepts the incident (asks for forgiveness or apologizes) or refuses responsibility for the incident (but does not dispute its existence), this represents a strategy that may delay the conflict. This type of response allows the company to express its regret about the inconvenience encountered by the customer. The company then hopes that the customer will forgive them [BOB 05]. Studies on this subject show, in particular, that the response strategy must correspond to the seriousness of the problem encountered by the customer: the more serious the incident, the more exhaustive the response provided by the company will have to be and possibly include (material) compensation [DEN 15]. Various studies also show that it is better for a company to accept its responsibility than to deny it [MUN 13, MUN 14]. In addition, when there is a strong customer community surrounding the brand, other Internet users can also support the company and respond to negative feedback. Taking into account community reactions is therefore essential to increase the effectiveness of the response strategy [KUN 12, LEE 10].

Recent literature raises and addresses the question of communication style, particularly from the perspective of whether or not it should be adapted to the specific context of a platform. This gives the company two response options: (1) maintain the same tone, regardless of the context and platform-specific communication codes, or (2) adapt its language to each environment. Communication Accommodation Theory (CAT) provides a framework for interpersonal interactions [GAL 05]. While the strategy of maintaining institutional communication can strengthen brand consistency, especially when it is present on many platforms, adjusting language according to context and people can also be useful to demonstrate the company’s proximity to its customers. For example, Jakic et al. [JAK 17] show through an experimental study that adjusting the tone of a message according to the person is the preferred strategy to promote customer trust in the brand because it reflects the company’s goodwill and quality of interaction, an effort generally well perceived by Internet users.

5.4.1. Challenge 1: when the customer’s voice is manipulated (the case of deceptive reviews)

Listening to the customer’s voice offers companies the opportunity to have a more reactive and/or proactive management of incidents reported by customers. However, because of the potential impact of these messages on consumer attitudes and purchasing decisions, companies also perceive online conversations as potential risks that can affect the credibility of their brand and the effectiveness of their marketing communication. Therefore, this freedom of consumer expression encourages companies to seek ways to regain control over their communication [MUN 15]. To do this, some companies do not hesitate to publish deceptive reviews [LUC 16, MUN 15, MUN 16]. The practice of deceptive reviews refers to “the deliberate attempt, whether successful or not, to conceal, fabricate, and/or manipulate in any other way factual and/or emotional information, by verbal and/or nonverbal means, in order to create or maintain in another or in others a belief that the communicator himself or herself considers false” [MAS 04, p. 148]. In accordance with this definition, Xiao and Benbasat [XIA 11] emphasize three main characteristics of misleading communication: (1) deception is an intentional and deliberate act and is thus distinguished from disinformation (i.e. an unintentional distortion of information); (2) the attempt at deception involves the manipulation of information and (3) the actor uses deception for the purpose of exploitation. In order to appear authentic, marketers and reputation specialists generally demonstrate professionalism in the production of these opinions by imitating the customer’s voice to perfection. Ott et al. [OTT 12, p. 201] stress, moreover, that these “fictitious opinions have been deliberately written to sound authentic, in order to deceive the reader”. These deception attempts, which represent nearly one-third of the reviews published on online platforms, are not without consequences because they also tend to erode consumer trust [MUN 12]. Even if measures to combat fake reviews such as the French law for the digital republic and the AFNOR standard have been put in place, consumers are increasingly sceptical about opinion platforms and are turning to their nearest and dearest again to obtain offline WOM [MUN 12].

5.4.2. Challenge 2: when the internal customer – the employee – expresses himself online

A systematic listening approach should not neglect another stakeholder as important as the company’s customers, namely the company’s employees. While customer voice and customer service via platforms and social media have been the subject of an increasing amount of research [KIN 14], studies on employee voice, via websites such as Glassdoor or on social media evoking a company’s reputation, are still few in number [KÖN 18]. Opitz et al. [OPI 18] point out that negative reviews posted by employees disproportionately damage the company’s reputation compared to customer reviews. Despite these results, a study conducted by the Glassdoor website [GLA 18] also shows that 62% of job seekers’ perceptions of a company improve after reading an employer’s response to an online notice. Another recent analysis indicates that only 12% of companies listed on Glassdoor interact with site visitors by creating a free account in order to respond to negative opinions left by employees or candidates [ADA 16]. In conclusion, the systematic listening to the customer’s voice should take into account the multiplicity of existing platforms as well as all stakeholders likely to express themselves through these digital channels.