The reality of social media updates
We have been spending a lot of time lately working with customers who are trying to get a sense of what can be expected from systematically measuring social media. One of the challenges that comes up consistently is a realistic expectation of how often these metrics need to be updated. The current assumption is that social media commentary is streamed into the system, and the scoring algorithm is continuously updating consumer sentiment.
It’s a nice idea, but the reality is a bit different. One of the issues is having a correct set of expectations; people are familiar with social media as a consumer—there’s over a billion people on Facebook alone and that many users talking at the same time generates a staggering amount of content. So as a user, I am used to a non-stop stream of social data from my friends, their friends, advertisers, etc. And that is the first problem.
Once you apply a business filter to social data, it immediately becomes a non-stream. What Facebook (as the most obvious example) has is a highly imbalanced signal to noise ratio. A billion folks rattling around a site generate a lot of noise, and business is looking for signal. Now of course, the signal will come from a subset of the noise, and when you have a source of a billion, even a small percentage relevancy will be significant. So businesses are looking for information (signal) about what consumers are thinking about their brands, products, services, etc. and they expect it in a steady stream, because that is how they are used to dealing with social media.
That is not, however, how filtered social media works. You can’t expect data to come in on a regular basis (“we want daily updates on what consumers are thinking”), it doesn’t work that way. A good example is the airline industry, which is a very consumer-facing domain, and tends to have very vocal customers. People do not tweet or blog constantly about issues with the airlines, unless something bad has happened (e.g. stuck on a tarmac for seven hours), and then it’s a Tweetstorm. Tweetstorms are a great source of sentiment-laden data about business issues, but by their nature, Tweetstorms are non-predictable. It’s not like every Tuesday JetBlue makes the 2pm New York to Boston flight sit on the tarmac for seven hours. Tweetstorms and the associated media effects are arbitrary, unpredictable, and event driven, and social media analytics customers are looking for non-arbitrary, predictable data. Data comes in when it comes in, and rather than setting a data update schedule that has no real connection to what is going on in the social media space, we would be much better off setting thresholds and importing against that. If the threshold is triggered, you can assume something significant has happened, and you’ll be the first to see the bigger picture.
So the problem is this will work great for some industries such as consumer facing companies that manage to piss off their customers (banks and airlines are good at this), and perhaps less so for others (an ERP company looking for sentiment on a new operational feature). That kind of input needs to come from user groups, and no user group is anywhere near the scale of a Facebook.
This is an interesting area to be working in these days. There is definitely signal out in the social media space, but getting to it in a statistically meaningful way that can be understood by a non-technical audience is turning in to a bit of a challenge. The whole social media analytics space (at least those of use focused on trying to help businesses get a grip on this) needs to set the right level of expectation, and show the value that a well thought through solution can bring to bear.