Sramana Mitra: What are the trends in the data you are processing?
Rick Kieser: It is difficult to say, because we process more than 300 million comments a year. Since we have such a large platform, we span the major industries – pharmaceutical, transportation, energy, financial services.
I can’t make comments on trends that are happening. But I can say that the ability to manage unsolicited feedback and use different technologies to do it on an automated basis is a huge trend. The volume is increasing. A social media feed, for example, can bring in hundreds of thousands of comments. In an extensive survey coming from the market research world, you might be dealing with 50,000 comments.
[The rise of] unsolicited feedback is causing the industry to figure out how to leverage technology to be more effective in manually reading these comments. A lot of firms that work on this are applying a “one technology suits all” approach. There are a number of firms out there that say that they can code and classify virtually every comment by using a rules-based NLP approach – with a simple two by two matrix. On the left side you have interpretation. On the bottom you have “easy to interpret” and on the top side you have “difficult to interpret.” Then you have volume – low volume and high volume. If you are going to map survey responses on that two by two matrix, it is in the lower end of that matrix – low volume and easy to interpret. By easy I mean that people are answering questions in text boxes or off their cell phones, and they are fairly structured in terms of what question they are answering.
In contrast, in the upper right hand corner there is social media – very difficult to interpret and at high volume. It is interesting to see how Twitter feeds are becoming increasingly difficult to read. They are just a series of characters and hashes. With that, if you overlay technology onto this matrix, semi-automated coding can cover everything from low volume to high volume and very difficult things to interpret, because in essence it is manually based.
NLP resides in the lower right hand corner – good for high volume but not good on low volume, because it is cost prohibitive to get [the technology] trained well. It is also not good on things that are difficult to interpret because you have to write customized rules. As I said, many firms are applying a “one technology approach.” That is the wrong answer. You need to use a blend of technologies, and you need to have an environment that allows one to seamlessly use one technology based on the characteristics of feedback.
SM: If you look at the landscape out there, I am sure you are aware that there are various firms doing analytics and categorization of social media feedback at scale. Who are your top competitors? Who is doing interesting stuff?
RK: The ones that come to mind are Attensity, Clarabridge, and Semantria. They are all different in what they do, but they tend to focus on leveraging one core technology as a solution. We are very different. We use multiple technologies. We believe that is the right way.