Recommendation
After analyzing Twitter data, we recommend Amazon focus on improving the voice recognition capabilities of Alexa. This will boost Amazon sales through Alexa speakers and position Amazon as the market leader in the smart devices category.
Planning and Execution
Initially, we expected to find tweets about specific features that customers are unhappy with. We presumed consumers might complain about Amazon devices’ ability to hear and understand user commands, especially in noisy environments. We collected roughly 5,000 tweets about Alexa speakers and the primary competitor, Google Home. We assembled seven datasets using Tweepy based on several keywords, including “Amazon Alexa,” “Google Home,” “Echo,” “Dot,” “Tap,” “Domino’s,” and “Burger King.”
Due to a recent, widely-criticized Burger King television advertisement attempting to activate Google Home, we included “Burger King” in conjunction with searches for Google Home (1). To compare this to a similar but accidental hijacking of Alexa, we included “Domino’s” in conjunction with searches for Amazon devices. Next, we passed the CSV through a Word Frequency analyzer. This led to another set of challenges. To analyze relevant word frequencies, we used stopwords to eliminate articles such as “the,” “an,” “a,” and others. We imported the Natural Language Toolkit (NLTK) to assist with language processing but ran into errors when using the library.
After debugging, we discovered the Anaconda package manager had downloaded the NLTK library into its development environment, yet we were trying to run NLTK elsewhere. Once the NLTK was downloaded to the correct environment, we calculated word frequency. After sorting, we identified frequently used words unrelated to Alexa speakers or Google Home. Then, we removed irrelevant tweets from our datasets and cleaned the data to account for any possible lowercase/capital case differences. Irrelevant tweets were common since people tweeted about another person called Alexa, not the voice assistant.
We also made clarifications so that #AmazonAlexa would be considered the same as #Alexa in our search terms. Next, we calculated lift ratios, which show how correlated word pairings are. The Lift Calculations script outputted a CSV with a table of lift values (see figure below). Once we compared the values, we were able to narrow down traits eligible for sentiment analysis. Sentiment analysis is the process of computationally identifying and categorizing the attitudes and opinions in a piece of text, whether positive, neutral, or negative.
We used SentiStrength software for analyzing sentiment. SentiStrength counts each word in a tweet as positive, negative, or neutral. Then the counts of overall positive and negative tweets were added together and averaged to find the overall sentiment of the dataset. We considered scores over 0 to be positive sentiment and scores less than 0 to be negative sentiment. This method allowed us to infer customer attitudes.
Data Analysis and Insight Development
The outcomes of the data analysis suggest a generally neutral-slightly positive (+.037) sentiment towards Alexa speakers. There are not many complaints, and a vast number of tweets are neutral announcements and news articles about Amazon Alexa from tech reviewers. The second outcome is that sentiment towards Google Home is neutral-more positive (+.193). Google Home seems to have slightly more positive tweets about its usage, but the dataset also suffers from the same challenges of having many announcements, promotions, and news articles rather than consumer testimony.
Google Home’s positive sentiments might also be due to its advantage in machine learning over Amazon Alexa (2). We directly compared Google Home and Burger King to Amazon Alexa and Domino’s. Both were situations where a food commercial triggered the activity of the home smart speaker. However, Burger King’s intentional hijacking led to a sentiment score of -0.583, where users voiced their concern over the security of Google Home. In contrast, Domino’s accidental triggering of Amazon Alexa led to users expressing humor in a positive sentiment score of 0.385 for the dataset.
By targeting Google Home, Burger King has created negative sentiment towards its own brand and the smart speaker whereas Alexa and Domino’s enjoy a now fruitful relationship where users frequently tweet about enjoying the ability to order pizzas through Alexa.
This methodology, relying on Twitter data, however helpful, is imperfect. There are five principal limitations. First, casual Twitter users do not use the platform to constructively critique products. Second, many tweets in our data were retweets for Alexa speaker giveaways. Third, these are relatively new products, so the dataset is limited; indeed, after filtering for relevant data, the dataset shrank to approximately 2,000 tweets. Fourth, we found that sentiment was broad rather than specific (i.e. “I like the Alexa” vs. “I like feature X of the Alexa”). Finally, we discovered that at 140 characters, tweets do not transmit much information. Most of our sentiment searches resulted in inconclusive results most of the time. Refer to Exhibit 1.
Nonetheless, we conclude that these findings are applicable in a broader scenario because Twitter offers palpable insight into general sentiment with minimal costs other than employee hours spent on analysis. For example, this analysis could be beneficial in monitoring sentiment about new Alexa devices over time as new upgrades and features are released. To improve this analysis, Amazon requires more data collection; this means monitoring Twitter more often but also utilizing other modes of data collection, such as Amazon reviews and market research.
Application of Findings
Our primary recommendation is to improve voice recognition capabilities in Alexa speakers. This improvement needs to be coupled with a “user profile” feature so that permissions vary from person-to-person. This feature will prevent, for example, a young child from ordering an expensive toy that a parent does not intend to purchase (4). Currently, Alexa’s ability to discern voices is limited, and several compelling business objectives justify making substantial improvements. The first of these objectives is addressing security and privacy concerns. Our analysis showed that Twitter users are concerned about how easy it was for Burger King’s ad to “hack” the Google Home.
Twitter users have good reason to be concerned: Alexa contains sensitive user information and has the ability to make purchases from Amazon. Consumers trust Amazon and making these necessary improvements will reinforce trust and encourage more consumers to feel comfortable with making purchases from their Alexa speaker. Indeed, RBC Capital Market estimates that Amazon orders through Alexa could generate an additional $10 billion in sales by 2020 (6). The resulting reputational benefits will be long-lasting and play a role in cultivating a sustainable competitive advantage while serving as the impetus for Alexa proliferation. For a complete SWOT analysis, refer to Exhibit 2.
The benefits of this proposal outweigh the costs. Amazon spent $14.2 billion on research and development in 2016 (2). The technology for voice recognition exists, it simply must be improved (3, 5). We estimate implementing an improved voice recognition feature would not impose an unreasonable cost on the firm, and could be accomplished within the next two years to close the gap with Google. Acquisitions may be required to expedite the process. These estimates are conservative and imprecise, and Amazon will need to conduct a further cost-benefit analysis. In the future, voice recognition will be essential to smart products such as Alexa speakers, and implementing this plan will better prepare Amazon for that future.
Exhibits
Exhibit 1:
Exhibit 2:
Works Cited
- Maheshwari, S. (2017, April 12). Burger King ‘O.K. Google’ Ad Doesn’t Seem O.K. With Google. Retrieved April 28, 2017, from https://www.nytimes.com/2017/04/12/business/burger-king-tv-ad-google- home.html?_r=0
- Bhartiya, S. (2016, November 22). Amazon Echo vs. Google Home: The choice is obvious. Retrieved April 27, 2017, from http://www.cio.com/article/3143137/open- source-tools/amazon-echo-vs-google-home-the-choice-is-obvious.html
- Fox, J. (2016, August 09). Amazon and Google Change the R&D Race. Retrieved April 28, 2017, from https://www.bloomberg.com/view/articles/2016-08-09/amazon-and- google-change-the-r-d-race
- González, A. (2016, December 01). Amazon’s Alexa on journey to fully grasping natural speech. Retrieved April 27, 2017, from http://www.seattletimes.com/business/amazon/amazons-alexa-on-journey-to-fully- grasping-natural-speech/
- Liptak, A. (2017, January 07). Amazon’s Alexa started ordering people dollhouses after hearing its name on TV. Retrieved April 28, 2017, from https://www.theverge.com/2017/1/7/14200210/amazon-alexa-tech-news-anchor-order- dollhouse
- Voice Training. (n.d.). Retrieved April 28, 2017, from https://www.amazon.com/gp/help/customer/display.html?nodeId=201601940
- Vena, D. (1970, January 01). Amazon’s Alexa Just Took Another Step Toward World Domination. Retrieved April 28, 2017, from https://www.fool.com/investing/2017/04/25/amazons-alexa-just-took-another-step- toward-world.aspx