With the rise of new media, customers have assumed a more active role as market players and the ways in which they collect and distribute or exchange information are constantly changing (Hennig-Thurau et al. 2010:331). These media platforms are a perceived threat to long-standing business models and corporate strategies, but simultaneously provide brands with growth opportunities and the potential to maintain their relationship with the public through online communication (Hennig-Thurau et al. 2010:331, Rybalko & Seltzer 2010:336). As a result, social networking sites (SNSs) serve significantly in environments where negotiation between individuals and business organisations may occur (Kent & Taylor 1998:332). It has not only attracted the attention of “youngsters,” but also adults aged 35-44 (Kaplan & Haenlein 2010:59). The free micro- blogging platform Twitter, in particular, is praised for having altered the realm of marketing in business and stipulation of dialogue on a global level (Bruns & Stieglitz 2013:100, Sevin 2013). However, communication on Twitter can become problematic for a brand since users are able to generate and share content instantaneously to public audiences without the supervision of traditional gatekeepers (Jansen et al. 2009:2169). This study attempts to understand ever-changing interactions between brands and consumers.
The purpose of this project is to gain insight into the use of social media platforms, particularly Twitter, by two different business brands to communicate. It will do so through a comparative analysis of two food ordering and delivery platforms, of which one is American-based, and the other Australian-based. The comparative analysis will cover the ways in which both businesses manage their own brand perception and engage with its users online in public relations and customer relation management. Compared to products, services are characterised as intangible and the quality provided varies (Hornikx & Hendriks 2015:179), therefore the information surrounding the industry is more unpredictable and interesting to study.
Menulog is Australia's largest online food ordering platform and has been operating for over ten years. By contrast, the American-based UberEats is relatively new as it was launched by Uber in 2014 and partners with restaurants in many cities around the world. Through critical analyses of the types of messages these brands send out, it is revealed that both brands employ different promotional methods, as well as strategies to suppress the voice of users who share complaints or
question their quality of service. User voice is further interpreted through public sentiment !3
expressed about the brands. Although Twitter provides a platform for connectivity and discussion, it is ultimately up to the brand to improve their online presences by facilitating a space of safe and constructive communication (Rybalko & Seltzer 2010:337).
This report argues that Menulog is more susceptible to negative sentiments on Twitter than UberEats because it operates nationally instead of internationally. It also offers the possibility that UberEats manages negative sentiments more effectively, and therefore has a stronger customer service position when interacting with customers. This hypothesis will be tested on a randomly selected sample of tweets.
This section describes and justifies the methods used in the report, which are content analysis and sentiment analysis. Any apparent ethical concerns will also be raised.
1. Content Analysis (CA)
CA is a research technique for objective, systematic and qualitative analysis of communication content (Berelson 1952:18). The structure adopted for this study is similar to that of Hornikx and Hendrik's (2015), where the goods and services of 24 brands were analysed on a smaller scale. The tweets were collected by building a scraper, using an Application Program Interface (API). Random sampling was used to reduce the sample size fairly, and then two rounds of coding was applied manually by two individuals. A reliability test was conducted to determine what percentage the two results matched.
This study uses NodeXL, which is a Microsoft Excel open-source template. Through the import tool, 300 tweets were collected “From Twitter User's Network,” specific to the Twitter handles @Menulog and @UberEats. This ensured the tweets collected were only ones authorised by the brands' accounts. The sample size was reduced by half through the RAND (=rand()) function in excel, adjacent to the tweets column. The RAND numbers were then copied and pasted as real
1. What types of messages do brands send out on Twitter? How do they compare?
2. How does sentiment expressed about the brands compare?
3. How do the social network maps for the chosen brands compare?
values in a separate column, and the initial RAND column deleted. After sorting the values from A to Z, smallest to largest, the first 150 tweets were selected as the final sample.
The coding frame above was then applied manually (see Appendix A for full manual). According to Krippendorff (2007:330), coding frames are used to interpret observable data with minimised judgement. Still, subjective categories can cause discrepancies hence a second observer re-codes the sample tweets before it is tested for reliability using ReCal2 by dfreelon.org (see Appendix B).
2. Sentiment Analysis (SA)
If brands are to protect public perception of their online image, it is crucial to monitor consumers' opinions on Twitter. This allows them to respond to attacks directly and also foresee future harm (Hornikx & Hendriks 2015:177).
SA was collected using a lexicon-based program called SentiStrength, which estimates sentiment using a five-point scale. It reports the number “5” as being the most positive and “-5” as the most negative. It is important to note that the numbers “1” and “-1” are considered to be neutral sentiments.
300 tweets were collected using the import tool “From Twitter Search Network” on NodeXL. The results were cleaned to eliminate tweets by the brands themselves, as well tweets in another language besides English. The .txt file was uploaded onto SentiStrength and analysed. Using functions on excel like =AVERAGE and = COUNTIF, the averages of positive and negative sentiments were calculated, alongside the amount scored on each number of the five-point scale.
This method was adapted by the SA conducted in Jansen et al. (2009). Because not all nuances can be accounted for, SentiStrength must be combined with other methods.
3. Ethical Considerations
There are ethical debates surrounding the appropriateness of online research and anonymously collecting data without gaining formal approval to do so (Kadushin, 2005). However, the topic of this study does not touch on sensitive topics or threaten the safety of Twitter users therefore the process of gaining informed consent was deemed impractical.
Research Findings and Discussion
Message Typology and Content
It is valuable for brands to first understand the terrains of their social media before they can effectively respond to feedback, queries and positive or negative comments that are expressed by customers (Kietzmann et al. 2011). In an attempt to understand the presence of brands on social media platforms, this report will first analyse the type of messages sent out by both brands.
Notably, the majority of tweets distributed by both brands are categorised as replies. This observation may be linked to the concept of the “dialogical loop” where the Internet is characterised as a space which facilitates feedback, queries and most importantly responses between
organisations and the general public (Kent & Taylor 1998:334). However, the charts above demonstrate a considerable difference in the frequency of replies used by each brand. Menulog's use of replies is over double that of UberEats at 98 percent, therefore displaying high levels of engagement with customers. This indicates that the brand's primary use of Twitter is a responsive mechanism. On the other hand, UberEats appears to preference seed messages almost just as much as replies, followed by standalone statements. Perhaps by incorporating questions, news and facts into their messages, UberEats is attempting to spark multiple conversational threads. That being said, no definitive conclusions can be drawn since the size of the data set is not representative of the whole.
To understand brand messages more thoroughly, the content of each tweet is examined in the next section. Problems arise when sorting the data by its subject matter because some tweets are ambiguous in nature and offer no indication of what is being referred to. For example, roughly 22 percent of UberEats' tweets have been marked as unclear. Regardless, assumptions can be drawn based on the remaining data. The distribution of results across content categories vary quite significantly between Menulog and UberEats. Remarkably, messages addressing issues with customer orders are extremely skewed, with Menulog at 36.5 percent and zero percent for UberEats. This result may be the product of suppression implemented by the brand as a strategy to create the illusion of a smooth-running and problem-free choice for food delivery in comparison to its competitor.
Furthermore, Menulog displays greater empathy towards customers who encounter issues ranging from the application and ordering, to payment, reviews and locating restaurants. The brand often begins the response with “Hey [name], we're so sorry to hear this!” followed by “Please DM us...
so we can look into this for you.” Menulog's strategy in dealing with customer issues seems to be an interplay of empathy and informational cues so that customers who are personally addressed feel validated, but are also reassured of the brand's problem-solving capacity at the same time. Research has demonstrated that this strategy of combing informational and an extent of emotional elements are popular among consumers (Araujo et al. 2015).
Despite the fact that messages addressing order issues or other queries by UberEats are almost non- existent compared to Menulog, the opposite can be said about its use of promotional material, general replies and news-related tweets. In the column chart above, this visual comparison is evident as UberEats surpasses Menulog tweets in promotional material by over 10 percent, in general replies by 25 percent, and in news by about 19 percent.
With regards to promotional material and general replies or mentions, further trends are apparent. Although Menulog seems more directly involved with customers online, UberEats employs a larger mix of marketing aspects in order to engage with broader audiences. Tweets by Menulog include some use of emojis, images, gifs, links but hardly any use of hashtags to personalise the brand, and zero use of events. By contrast, UberEats uses all three strategies more frequently. For instance, the
brand's hashtags are incredibly specific to the information shared within the tweet like “UberEATSHyd” (launch in Hyderabad, India), “UberTACOS” and “UberChopper.” UberEats also coordinates some of their restaurant's food in promotion with certain movies showings such as Lady Bird (croissants, chocolate cake, caffeine) and The Shape of Water (key lime pie). The distinctive promotional methods used by each brand highlights gaps for improvement in the other.
Power of Sentiments Expressed
Tweets contain immediate sentiments which allow reactions towards the brands to be analysed (Jansen et al. 2009:2170). By organising the sentiment analyses into a table, some trends are made apparent.
Positive sentiment surrounding UberEats is only slightly greater than Menulog, however negative sentiments surrounding Menulog is significantly higher than that surrounding UberEats. Compared to UberEats, Menulog's percentage of negative sentiments doubles itself at level-2 and -4, and triples itself at level-3. However, it does score higher in positive sentiment at level+3.
According to previous studies, tweets concerning organisations are often positive rather than negative but Jansen et al. (2009:2173) claims this does not prove true for services. Existing literature suggests that tweets on services provided by brands are especially vulnerable to negative reviews (Hornikx & Hendriks 2015:17).
Close examination of Menulog's most negative tweets reveals key themes of delivery quality; specifically waiting time and miscommunication. Triggers of emotive tweets are linked to trust and personal experience with delivery drivers. Customers' negative experiences were identified as being expressed through patterns of brief and detached messages. For instance, the expressions “not 2 be trusted” and “incompetence is clearly a KPI there” (key performance
indicator) were tweeted without further explanation. According to Hornikx and Hendriks (2015:179), such stimuli get noticed and therefore negatively impact service evaluations.
Whether or not due to strategic suppression, the most negative tweet about UberEats was by “college kids” who felt that ordering food was too expensive. However, a noteworthy finding was that a few tweets marked negative under Menulog was actually addressing UberEats. For example:
Thus the tweets that were praising Menulog while negatively rating UberEats were inaccurately accredited. This highlights weaknesses and limitations in the method of using SentiStrength.
These findings are reflected below in the calculated averages of sentiment towards the two brands. On Twitter, Menulog has lower positive sentiment and higher negative sentiment than UberEats. This supports the hypothesis that Menulog is more susceptible to negativity, or at least would seem so considering it only operates in Australia. As a business that operates in hundreds of cities around the world, negative tweets about UberEats can easily be suppressed or drowned out by others, or can be written in a foreign language which cannot be analysed on SentiStrength.
This project attempts to understand how two business brands use Twitter to communicate. Research findings have demonstrated that both Menulog and UberEats use the Twitter platform primarily to respond to customers and avoid potential harm to the brands' reputation. However,
Menulog does so to a much greater extent considering it operates only in Australia and therefore
must maintain its current position in the market to continue alongside its international competitors like UberEats. In addition to replies, use of tweets by UberEats also include promotional and news- related material, as well as general replies to customers to display high levels of brand-to-customer engagement. This may also be viewed as a strategy to suppress negativity by the brand as other message content will drown out negative ones. Results show that sentiments collected for UberEats is more positive than Menulog, while highlighting the limitations of using SentiStrength. For example, a few tweets criticised or questioned the quality of UberEat's services, praising Menulog instead but SentiStrength categorised them as negative sentiments on Menulog's part. Therefore, the methods used in this study should be combined with other methods to produce more accurate and conclusive results.
Promotional methods employed by both brands were also observed. Menulog includes use of emojis, images, gifs and links to some extent but fails to promote their brand further through brand-specific hashtags and events. Whereas UberEats promotes food pairings in accordance with new movie releases. Moreover, their hashtags include “UberEATSHyd” “UberTACOS” and “UberChopper.” It is recommended that Menulog adopts similar promotional approaches in the future to engage with customers in a new and exciting way, which will strengthen its brand's identity. On the other hand, UberEats should be forward like Menulog when addressing issues and concerns so that customers know they are being listened to. Overall, Twitter proves to be an important platform for public relations management hence brands should understand their online presence and relationship with the public before initiating any from of engagement.
The research conducted in this study has several limitations in terms of methodology that must be acknowledged. This section will consider the challenges which come with content analysis and sentiment analysis and provide further suggestions.
The most problematic aspect of CA was designing coding frames. To begin with, categorising messages by typology and content was difficult because it had to encompass all integral elements of the tweet. If the message failed to fit the designated categories, it would be labeled as other “unclear.” Ultimately, it would have been beneficial to be able to closely examine
the content of these tweets as well. Considering 22.3 percent of UberEat's tweets were marked as unclear, ability to compare the two brands regarding subject matter was not entirely achieved. The process of categorisation is also subject to interpretation and bias (Krippendorff 2013:127). As a result, having a second observer who was unfamiliar with the project re-code the tweets required careful attention and time. Although the reliability test was successful, its success may have been attributed to the brief descriptions of message typology and content by the research to the observer. The two brands were also selected before close examination of the brands' tweets online. If this study was repeated, comparing food ordering and delivery services that were more similar in nature would perhaps have been more insightful. For example, UberEats and Deliveroo which are both non-native brands to Australia.
SA was able to provide a general overview of sentiments expressed by customers for each brand, however had some clear inaccuracies and inconsistencies. Namely, users have to be extremely descriptive and expressive in order for Sentistrength to be able to rate the sentiments as positive or negative. Furthermore, the context of curse words were not properly recognised as offensive. For example, one tweet addressed to Menulog read “Don't lie to customers. Don't lie to ‘restaurant partners.' Just fucking don't” only received a rating of -3 despite the implications of the statement being dire to the brand's reputation. This questions whether or not computerised software can comprehend language in the same way the humans can (Hornikx & Hendriks 2015). Irony and figurative language are prime examples. Moreover, SA runs the risk of reducing expressions of emotions to a numerical scale, obscuring nuance and the original context that the user tweeted. Future suggestions could include a complete re-coding and re-training of the classifier to improve accuracy (Pak and Paroubek 2010) or using a patterned analysis based on content and emotion to bypass the issues mentioned above (Nguyen & Jung 2017). SA cannot be ignored as it provides a valuable first layer of analysis on large datasets, however like CA, cannot be used alone.
This particular report exclusively focuses on the communication between two brands and their customers on Twitter. Consequently, it disregards social communities formed among other brands and or groups of consumers. Hence the social network mapping as a whole is not captured. Moreover, the scope of future studies could be expanded to address customers' level of satisfaction towards the brands' response as well as customers' relationships in terms of independent discussions that may affect brand reputation.
...(download the rest of the essay above)