Wednesday, October 24, 2012

Business decisions based on sentiment

Photo credit: Paul Ekman's facial expressions,
Every executive I've spoken to since 2005 has been tantalized by the promise of using social media sentiment as a KPI. Sentiment was hyped to be the most actionable of all social media data, a tidy shortcut to business decisions, not too unrelated from perception-based metrics of old with massive scale.

This continues to be a promise, with examples cropping up of campaigns being tweaked, products reintroduced or discontinued, all based on sentiment. Real business decisions.

At the same time, more scrutiny is being placed on sentiment-- how it's calculated, whose sentiment it reflects, what data it needs to be validated by, and why it fluctuates, in the context of vertical norms and benchmarks.

The market appears to be maturing rapidly about the different ways it's calculated, as witnessed in part by listening platform differentiation based on sentiment technique. Often it's a make-or-break decision to have the possibility of customized sentiment rules in a platform. Academia mirrors this trajectory with papers in psychology, computer science, and communications optimizing algorithms and language processing techniques.

We've fixated on how closely we can mirror sophisticated human judgment. And incremental improvements arise frequently. So much so, it's made me wonder whether accuracy is the most important part of this equation. What about the relationship between sentiment (emotion, incl.) and behavior? Or more precisely, how much of (the variance in) purchase behavior can be predicted by sentiment? In order for executive decisions to be based on sentiment, we need to know which behaviors are reliably tied to sentiment. The academic literature runs sparse (and/or wild) here-- as do publicized business cases. 

Structure (amidst big data) does not necessarily beget (revenue yielding) behavior.  Our hopes are only partially fulfilled; our work, only partially done-- in order to enable business decisions based on sentiment, we need more research on the behavioral relationship.

If you can make it to San Francisco next week, I'll be discussing this topic at the Sentiment Symposium. Until then, I look forward to hearing from challengers here. Executives, tell me your greatest examples of business decisions based on sentiment. Academics, share your research on the relationship between sentiment and behavior. Together we'll derive hypotheses as to when we can and cannot act on sentiment.