I was sorting through some statistics textbooks when I found my SAS Survival Data Mining course notes. I began to wonder how useful survival analysis could be for social businesses…
What is survival analysis?
Survival analysis involves statistical models that predict when a particular event or outcome will occur, like when patients will die or customers will churn (e.g. switching from AT&T to Verizon). Analysts use survival analysis to predict when a discrete event will occur. Another predictive modeling technique, logistic regression, can be used to predict if an event will occur, but not when. Survival analysis involves time-dependent outcomes or events.
How is this useful for a social business?
Many companies already have statistical models that tell them what predicts employee turnover. Variables like challenging work, the perceived lack of career opportunities, dissatisfaction with managers or management, and so on, may predict employee turnover (source: Wikipedia).
So what? What kinds of actionable recommendations can we make based on these statistical models? Perhaps the HR department will improve their intranet and make career paths and opportunities more visible, or HR could investigate managers when entire teams have below average engagement scores on the annual employee survey.
Imagine how much more useful it would be to predict which employees are at risk of leaving and when they may leave. Why not use survival analysis to predict when individuals employees may leave and design real-time interventions to prevent the loss of a company’s most valuable resource, its talent?
More often than not, survival analysis is not useful because of the limited data involved. Workforce and Human Capital Analytics teams have only demographic and survey data available to them. The models generated using this limited data can identify what variables predict employee turnover (e.g., lack of career opportunities), but not much else. Employee surveys are anonymous, preventing employee-level interventions. These teams can make high level suggestions, but can’t provide the business with the real-time insights that could identify and prevent valuable employees from leaving, before they leave.
What if Enterprise 2.0 data could help? Employees actively contribute to, and passively learn from, activity streams, blogs, discussion forums, and wikis. This activity is identifiable, in real time, and stored over time in a database. Could changes in employee usage patterns over time predict when employees are going to leave? Mining Enterprise 2.0 data using survival analysis and other statistical techniques could generate real-time analytics, allowing managers and HR professionals to intervene and prevent employee turnover, before it’s too late.
Marketing can also benefit from survival analysis and social data. The primary application of survival analysis is in telecommunications. The wireless carriers analyze customer history data to predict when you are going to switch. In theory, the marketing departments should be able to intervene with promotions and other upgrades to prevent customers from switching, before it’s too late.
I’m sure negative tweets and comments could be used to predict customer attrition or churn. But what about “unliking” in Facebook? Or “unfollowing” a company’s twitter account? What about changes in a customer’s usage patterns in a customer community? Could these changes be used to predict when a customer is going to stop shopping a retailer? Retailers can already leverage their existing customer satisfaction and transaction data to predict the variables associated with customer attrition. Real-time predictive analytics would allow the marketing department to re-engage with customers and prevent attrition.
My real motivation in posting this blog is to point out the usefulness of the underlying social data for predictive modeling. In the midst of all our whining about the absence of social analytics or ROI, we don’t really think about what we’re trying to accomplish. There are many potential uses for the underlying data, above and beyond the out-of-box web analytics and social network analyses. You say you want social analytics? Then start generating the business objectives and use cases to make it happen.