Although identifying and leveraging influencers is one of the fundamental strategies in social media marketing, a recent article in the Harvard Business Review (“What Would Ashton Do – And Does it Matter” by Sinan Aral) shows it is not as simple as many think. For those not familiar with the term “Influencer,” it refers to someone who has significant influence with either a niche or the mass market due to social media presence. It could be someone with two million Twitter followers or somebody whose blog is read by virtually every doctor (and thus influences the medical community). There are third-party services, such as Klout, that create scores that attempt to show how much leverage somebody has in social media.
In social media marketing, the tactics often revolve around identifying influencers and getting them to promote your game or product. The belief is that if some of these influencers are promoting your game or product, you will hit a threshold at which everyone is talking about it (and playing it). Thus, every marketer’s top goal is to get Ashton Kutcher, who has 13.7 million twitter followers, to tweet about them.
The article shows, however, that such a strategy is potentially misguided and may not drive many new players. He uses the example of Kutcher, relating an anecdote from a lecture he gave. First, he asks the audience how many people follow Kutcher, and most raise their hands. Then he asks how many have done something because Kutcher suggested it and nobody raises their hand. This example points to the paradox regarding influencers: Many who are followed by millions are listened to by only a few.
The main reason it is hard to understand influence is that people confuse correlation and causation. Behavior tends to cluster among friends; friends often take up activities about the same time but not because their friends are doing it (there are other external factors that get them all to want to do it at the same time). This process is called homophily. We eat at the same restaurants, go to the same gyms, and follow similar commuting patterns. All these things mimic social influence but may have nothing to do with it. Researchers call them “confounding factors.” This finding is consistent with one of the problems I identified last week in determining LTV (and predictions in general), the over-fitting of data. In that post, I argued that predictions could be less reliable because factors that fit the data are not actually causing the activity.
The author of the HBR article showed a great example of the importance of understanding this dynamic. “These findings have dramatic implications for marketing strategy. Imagine you’re the chief marketing officer for a brand that’s planning a product launch. Your company’s data scientist presents evidence that predicts that when someone adopts your product, many of his friends will start using it too. If your data scientist demonstrates that 90% of the correlation has to do with influence, you’ll probably allocate a lot of the marketing budget to peer-to-peer and word-of-mouth strategies—for instance, by offering a ‘friends and family’ promotion that allows users to share a discount code. But if she suggests that 90% of the correlation has to do with homophily, you’ll realize that a peer-to-peer campaign is unlikely to work and that you’d be better off segmenting the market into discrete demographics and targeting likely adopters (regardless of who their friends are) through traditional ads and promotions.”
The way to mitigate this issue is to account for confounding factors so you can differentiate behavioral tendencies from behavioral changes. To determine how much you may have influenced my decision to play a game, you need to know the effect of your behavior or recommendation over and above the prior probability of playing. Your success in social media marketing will depend on your ability to use robust analytics to increase your understanding of what is driving behavioral change, and not simply a correlation.