Evaluating Twitter as a Feedback Mechanism for Addressing Customer Grievances

This post is co-authored by Abhilash Maradwar, Ishan Sharma and Shubham Mallade, who are M.Tech students at IIIT Bangalore and Dr. Sachit Rao,who is an Assistant Professor at IIIT-Bangalore.

The immense popularity of social media platforms, such as Twitter, in diverse sections of society has led to them being used for interesting interactions between different groups. One such instance is between a customer and a business with a presence on Twitter. The motivation to study and analyse such an interaction came about in a conversation,involving one of the authors and a friend, about a parcel whose shipping information was not updated by a leading courier company, even after an inordinate delay. The friend was advised, by a third party, to post this issue in the publicly visible Twitter page of the company, as an immediate reply would be expected and even guaranteed as the company’s reputation would be at stake.

It was this anecdote that led to whether Twitter can be studied in the context of being an effective feedback mechanism by virtue of it being a public and online platform. In general, it has become understood that social media platforms have become an essential component of customer service strategies, as they now offer customers a choice of how and when they communicate with the company. Thus, by analysing such a conversation, the effectiveness of this platform can be gauged. This study can also be used to yield other insights, such as: the dynamics of companies in customer satisfaction and their relation to size; the effect of using software automation tools in interacting with potentially inconsistent human customers; the effectiveness of such platforms in altering the reputations of businesses, to name a few.

This problem will be addressed as follows: A company with a notable presence in popular media, such as those selling consumer products or services, will be selected. Next, the Twitter feed, the so-called Tweets, of the company’s consumer support handle, that are descriptive of company-customer interaction, will be downloaded and analysed offline. Collecting this data is a challenging task, primarily because there are limitations to the volume of tweets that can be extracted. Currently, a maximum of 100 concurrent (queued or processing) asynchronous job queries are permitted per account. In addition, result files expire 48 hours after the job is successful and queries taking over 60 minutes to process automatically time out. The tweets are extracted in reverse chronological order, so the most recent tweets by customers may have to be separated, by working under the assumption that the issues raised in the tweets may not be resolved so quickly.

There exist several additional challenges which have to be overcome, even with the availability of data. The first is the identification of a set of tweets that belong to a particular instance of conversation between a customer and the company, as this customer may have raised different complaints on different products. Moreover, on occasion, the companies also request the customer to send it a direct message, instead of tweeting. In these cases, it becomes difficult to determine whether the issue has been resolved or not as access to these modes of communication is not made public. This issue is perhaps the biggest hurdle in solving the problem of evaluating Twitter as a valid feedback mechanism. At this point, the ideas being applied to crossing this hurdle assume the following: The issue raised in the tweet is considered resolved if the customer does not retweet the same issue within a reasonable time frame or if the sentiment that describes the last tweet of either the company or the customer is found to be positive if this tweet does not explicitly mention that the problem is resolved. This problem leads to another challenge: sentiment analysis of the tweets with one immediately evident problem which is that of identifying sarcasm. In addition, it has been observed that in most of the cases, responses given by the companies are machine generated, so distinguishing the human element in the company’s response will require alternative approaches.

From access to the field site/participants perspective, the analysis will be conducted on the conversation recorded only in the public feed. Twitter provides an interface (API) to developers who can then access customers’ tweets which are explicitly made public by the customers themselves. This API also provides an access token to the developer that can be used to authenticate and restrict the developer on the number of tweets that can be accessed. This relieves the offline analysts from having to take consent of the customers as only public tweets are available for analysis. Moreover, the analysis will not influence the conversation between the customer and the company. These features also address the Ethics related issues that arise in such research efforts.

To conclude, the results of this study would indicate the effectiveness of Twitter, as a process that allows Internet Mediated Interactions, as a mechanism that permits feedback between companies and individuals toward resolving an issue that would enable both parties to reach consensus.

Disclaimer: The views expressed in this post are those of the authors and do not necessarily reflect the views of CITAPP and IIIT Bangalore.