Decoding Social Influence: The Marriage of Marketing and Data Science

When I mention the term 'social marketing,' most people assume a connection to Facebook, Twitter, LinkedIn, etc. Your friend 'likes' a page or an offer and it's recommended to you. You "like" a brand or mention the brand in a post, and your post is surfaced and promoted to others. Someone in your network reads an article and it's recommended to you. The fundamental premise comes from the idea that people trust and have a higher propensity to act on recommendations from friends.

Everyone has heard the age-old question: "If your friend jumped off a bridge, would you do the same?" Many would be surprised by just how true that message rings -- even in our adult lives.

For example, let's consider the contacts you keep via mobile. Did you know that for every person in your mobile social network (i.e. your circle of contacts) who churns from your current service provider, your chance of following suit more than doubles ? And if you're both part of a clique, (i.e. a social group within which everyone communicates with everyone else), that effect intensifies even more?

From a very early age, our decisions are subject to the influence of others. And although we all shrugged our shoulders when our mothers asked that question to prove her point, the same question is continually asked by marketers "If one customer behaves in a certain way, will others follow suit?" And more specifically "If one customer leaves, will her connections do the same?"

As I describe above, there is robust evidence for social influence in multiple marketing contexts (healthcare, telecom, social media, etc.), where it has been shown that customers who are close to each other are likely to exhibit similar behavior. But until recently, marketers have relied on standard practices of combining customer characteristics (e.g., demographics, past usage behavior) and past interactions between the customer and the firm to compute a customer score used for targeted retention efforts. The weakness in this approach is the inability to calculate the impact of influence by a customer's connections.

In attempts to redefine retention marketing, marketers are turning to data science to help (1) identify layers of social graphs throughout their customer base, (2) understand current and predicted impacts of those social graphs, and (3) translate those insights from abstract to action.

And for data scientists, the customer base of mobile operators, which has long been associated with chronically high churn rates, represents an ideal environment to quickly test, optimize and prove the impacts of social influence on churn behavior.

To their advantage, mobile operators have usage interaction data that easily connects the dots between individuals to map out social graphs. On the flipside, that usage data is being generated by millions of dynamic customers. The reality is that it's not humanly possible to manually identify social graphs, analyze their impacts and determine required actions based on those derived insights -- especially when behaviors are continually changing and data is continually flowing at this scale.

Big data technologies allow us to completely change the game as it relates to leveraging the power of social influence. We can determine and count at scale the pairwise linkages that underlie a social graph, and even make it possible to compute important graph-level metrics like PageRank and Betweenness. The granularity of that data allows us to separate things like professional and social connections, work colleagues vs. college buddies, and analyze those graphs to take action on them separately.

One specific point of focus is on advancing the scoring mechanism to reflect the 'social risk' versus the 'churn risk.' Instead of determining a score solely by the characteristics of the focal customer, we can also account for the risk factor of her close connections. With this approach, social influence is accounted for in both directions: (1) the focal customer being exposed to potential churners, therefore having higher risk of churn, and (2) the focal customer influencing her connections in the event of her churning.

Another area of focus is on post-churn campaigns targeting the connections of those who have already churned. In many cases, it is difficult for the company to detect early signs of churn, particularly if customers simply allow their accounts to lapse. Even if signs are identified and the customer is contacted, the marketing efforts may not successfully retain the focal customer. In such cases, the company may be able to "stop the spread of the fire" by identifying and targeting the most influential connections of the churned individual to prevent continued negative influence.

Now some of you may be asking -- how do you know that the influence is resulting in the change in behavior versus someone making that decision on their own? We refer to this debate as contagion versus homophily. Stay tuned for my next post where I'll dig a bit deeper on how science is helping marketers separate these two effects -- and ensuring that marketing efforts are aligned to opportunities of influence versus preconceived decisions.