Guest Post by Susan Etlinger, Industry Analyst at Altimeter
There’s a new trend taking the analytics and business intelligence world by storm these days: the idea of “predictiveness.” It’s sounds so magical, doesn’t it? Pull some data, analyze it and it will tell you the future. Cue happy customers, rainbows and pots of gold (or at least recurring revenue).
Except that it isn’t that simple.
Insight isn’t a rabbit that an analyst pulls out of a hat. The practice of data science is iterative and unglamorous. It can also be transformative under certain circumstances.
What makes the idea of predictiveness in social analytics so alluring in theory and so hard to pull off in practice is that social data—whether it’s based on natural language or images—is a form of human expression. And we human beings are complicated.
For that reason, social data is quite different from what we have come to expect in the world of business intelligence. In business intelligence, we learn about what is happening based on a set of structured signals—in the supply chain, product development cycle, sales process or elsewhere. It’s not a perfect system—the world is full of variables—but the fact that the we have access to structured data, defined business cases and algorithms that we have tuned over time lends it authority. Which leads me to point #1:
Human language and human preoccupations are far more complex to analyze than structured business signals.
One day, we’re collectively obsessed with the NBA finals, the next with a Supreme Court decision or a celebrity’s latest foibles. We use jargon, slang, sarcasm, emoticons, emoji, abbreviations and images, GIFs—even video—to convey our attitudes. And we do this in hundreds (if not thousands) of languages, sometimes in as few as 140 characters, and all in real time.
Taking all of this data and mining insights from it is very challenging. It requires the ability to understand the nuances of written language, the subject and context of images, and how all of that works together to tell us something meaningful about the attitudes and possible behavior of the user. And here we come to point #2:
Social data is part attitudinal (what people express) and part behavioral (what people do (share, retweet, like, pin, etc.)
In social analytics, at least today, the signals are less conclusive than in traditional business intelligence. We’re not looking at data on conversions, inventory management, turnover rates and errors per 1,000 lines of code. We’re looking at data on how many people discussed the latest death on Game of Thrones, or how the Republican slate is shaping up, or how cute Steph Curry’s daughter is, and what it all might mean to our current marketing campaign or sales pipeline or strategic roadmap. It’s a much more diffuse, dynamic and sometimes even fast-moving system.
But, when all of this rich data comes together with a clear business case and research question, social data can provide unprecedented insight. It can sometimes be a leading indicator of a pending crisis, or an unanticipated opportunity, a product defect or a new revenue opportunity. And if we do a really good job at analyzing what people say, and take into account the common themes and intensity and recency and acceleration and how that has affected business conditions in the past, we can start to better understand the relationship between what people say and what they actually do. That starts to sound a lot like predictiveness.
But the way we collect social data is fundamentally different from the way we collect business data such as purchases, miles flown or customer support calls. Each piece of data represents a human being’s action or statement, hope, dream, question, problem, suggestion or desire. It may come in the form of a tweet, a Facebook post or a photo on Instagram. Sometimes she may be speaking directly to us (@VirginAmerica) or just airing a grievance to friends and followers. Sometimes it’s impossible to know. So we need to account for the origin of the post, the context in which it was composed and what we can ethically determine about it and the user. This brings us to point #3:
Context is critical to making sense of social data.
When it comes to social data, which encompasses the complexity of human expression, maybe the phrase “predictive intelligence” is too simplistic. Maybe we should start with the idea of “contextual intelligence” instead. That puts the onus on us not only to define the proposition—what we are trying to solve for—but consider the context in which it exists.
To be sure, contextual intelligence isn’t going to be easy. But in data science, one thing is certain: you get what you pay for.
Click here to schedule a demo with us today.