Why would anyone want to do this? The first obvious interest is in tracking mindshare. How many people are talking about your product versus its competitors.
But it's not enough just to count the mentions of your product. When Microsoft launched the Zune, this was almost universally compared with the Apple iPod, so within a day or two there were thousands of webpages mentioning both. But unsurprising information is of little value; what's potentially significant here is not the absolute numbers but the relative shifts.
There are some important questions here about the volatility of buzz data. If mindshare fluctuates, is this a significant movement, or just random noise? The challenge is to build up enough statistical history to be able to set realistic action thresholds, and to identify potentially important weak signals for further investigation.
It might seem useful to know exactly what people were saying about the two products - which one they preferred and why. Until recently it has been almost impossible for software (and not always easy for humans, especially in unfamiliar cultural settings) to distinguish an enthusiastic "brilliant" from a sarcastic "brilliant", but Israeli researchers are now claiming a 77% precision in detecting sarcasm.
Joe McKendrick, New algorithm spots sarcasm in customer testimonials (Smart Planet, May 2010)
MacGregor Campbell, Just what we need: sarcasm software (New Scientist, May 2010)
However, tagging mentions according to sentiment still looks a pretty inexact science. Some vendors operating in this space don't include automated sentiment analysis at all (e.g. ConMetrics ); others provide simple trends only, leaving humans to do the detailed analysis (e.g. Lexalytics).
But never mind the technical detail. The point of this kind of business intelligence is that it is actionable. Companies can get an early indication of the success of a marketing campaign, long before mindshare feeds through into sales.
Because we aren't just interested in product mentions - we can also track discussion of particular design features of the product. How many people are talking about battery life or screen size or capacity or cost? This kind of detailed information helps identify the features that the marketing campaigns should emphasize, and may also feed into product development. Obviously if battery life is the most talked-about feature of this class of product, then that's a valuable item of intelligence for product designers as well as for sales and marketing. (I wonder how easy it would be to integrate this kind of business intelligence with a requirements engineering tool/method such as Quality Function Deployment, or a statistical technique such as MaxDiff? See Eric Almquist and Jason Lee, What Do Customers Really Want?, Harvard Business Review, April 2009)
If you have enough high-quality data, with all the automatic replication and spam stripped out, then you can also track the influence paths across the Internet over time. Not only identifying the pages that talk about the Zune versus iPod, but which pages came out first, and which of the earlier pages are strongly referenced by later pages. Not just individual thought leaders but also communities or geographies - for example, a given buzz might start on university campuses before spreading to other demographic sectors. That tells you where you should conduct market trials if you want rapid dissemination, and also where you should go for a relatively isolated trial of some high-risk venture. It also tells you which websites to watch for potential trouble.
What interests me most about this kind of innovation is not the technical details but the potential for transforming the business process - to develop greater organizational intelligence. Two years ago, Onalytica founder Flemming Madsen laid out a vision in his blog Predicting Sales from Online Buzz (Jan 2008) and Predicting Sales from Online Buzz - 2 (April 2008).
- predicting sales, market share and other outcomes
- detect changes in competitors’ behaviour
- setting targets known as “influence budgets”
- using “influence budgets” to predict whether an organization is on track to meet its actual revenue or market share targets, and take remedial action if required
But here's the thing I found most exciting. If an organization can develop sufficient confidence in the reliability of the predictions resulting from this kind of business intelligence, then the visible growth of influence and mindshare may enable it to sustain longer-term programmes and campaigns, instead of cancelling projects that don't deliver an immediate commercial return. Some people might imagine that an organization driven by buzz would be excessively short-termist - but the champions of this approach insist that good use of buzz by a truly intelligent organization could have quite the opposite effect.
I have talked to one large organization using this technology, and I'm hoping to publish this as a case study in the near future. In the meantime, I should be delighted to talk to any other organizations, to see what is actually happening in practice.
See also Just Shut Up and Listen, by Kishore S. Swaminathan of Accenture.