If Twitter isn’t driving visits or conversions, why bother?

Twitter’s direct impact on sales and traffic is still fairly small for most companies. For retailers, Twitter visits make up 0.02% of the total, with a relatively low 0.5% conversion rate. But most social channels would look similarly pathetic if measured so myopically.

Social’s most incredible attribute is its ability to answer questions that can help businesses perform better. We took on some of the tougher questions in The Conversation Index Volume 5, with a focus on Twitter – at 340 million tweets per day, Twitter is like a social seismograph for the entire world. Every analysis in this volume gets us closer to answering two fundamental questions:

  1. Where does social intersect with our everyday experiences, in life and in business?
  2. How do social and the “real world” affect and reflect one another on a larger scale?

We needed a ton of data to even begin to ponder this question, so we chose 13 brands from the BrandZ Global 100 list, acquired 26 million tweets that mention them from Gnip, pulled over 8,000 TV and radio mentions, 17 months of stock price data, more than a year and a half of Google search interest data, and 270,000 pieces of consumer-written content from online reviews. Here’s a taste of what we discovered.

Stock prices move with Twitter mentions

Back in 2011, we shared with you how UGC reflected the economy at a macro level (reviews mention price more when consumer confidence is low, and when the Dow falls). This time, we were curious to see if the volume of tweets about the brands in our analysis reflected their stock prices. We found that regardless of currency type, Twitter mention volume positively correlates to each brands’ stock price. At a macro level, this correlation of .91 was initially very surprising. It seems that the same things that make stocks move upward (like positive news) tend to make social chatter spike. Yet the things that can sink a stock don’t have the same power to incite tweets.

Twitter users’ influence up – and down

We looked at basic influence metrics of the users tweeting about brands, and saw a marked increase in the average user’s follower count over time. In the first half of 2011, the average user had 1082 followers. By the first half of 2012, that number had jumped to 1300. Further analysis might reveal one of two things:

  1. As Twitter use increases, the average user’s influence will inflate with it, or…
  2. The conversation about brands is shifting into the domain of more influential users

However, we noticed something that contrasted either of these hypotheses: the average number of lists that users had been placed in by others is decreasing over time. In the first half of 2011, our average user was listed 33 times. By the first half of 2012, this average had plummeted to 23. Is this an indication of Twitter’s list feature falling out of use, or is this an actual reflection of influence that is on the decline for users that mention brands? As usual, answering one question begs two more…

Twitter sentiment: A different animal

By the end of our analysis for this Conversation Index, we had solidly convinced ourselves that Twitter was a completely different species than onsite UGC (reviews, Q&A, etc.). The two channels are used so differently it’s difficult to find fair standards for comparison. But we were determined to compare the sentiment of tweets to ratings and reviews.

First, we used text analytics to “score” the tweets about the brands in our study. Then, since we had two different scoring systems, we standardized both scores and ended up with a score that reflected how far a piece of content’s sentiment was away from the mean of its population, negatively or positively (this is called standard deviation).

The results were mixed and only reinforced that the two channels are apples and oranges in some ways. Three of the brands we looked at had higher Twitter sentiment than in UGC; three others had higher sentiment in UGC than Twitter.

The overall sentiment score for our Twitter sample was a positive .047 (standard deviations from the mean) higher than UGC. Yet the scores appear to be meaningless when the context and themes are factored in. The most telling example may be from our Pampers analysis: In UGC, the most prevalent positive themes were “great product,” “dry diaper,” and “great value.” The common negative themes were around “diaper rash” and “sensitive wipes.” When we look at the tweets, it’s a whole other story. The positive themes – strangely enough – are around “pampers lol,” “wearing pampers,” and “printable coupons.” Aside from the coupons, these were mainly jokes made about others still wearing diapers. The negative themes focused on “changing pampers” and “potty training.” These were more straightforward – users sharing with the Twittersphere just how much they disliked changing diapers—which does not mean they dislike Pampers.

The net here is that for social sentiment analysis to be meaningful to businesses, they need to be able to analyze content about their brand or products, not just content that happens to mention them. With most social content, for brands that have more than just a few social interactions a day, it’s very difficult to separate the signal from the noise.