Seeing is Believing: Power of Observation in Predictive Analytics

Actions Speak Louder than Words: Why a Virtual Nose Smudge is Better Than a Tweet

The typical influence score used in marketing is based on speech, as in what people post. Today, this is usually measured in Tweets: the person who has the most followers is thought of as the most influential. On the surface, this makes sense, but, let me make the case that it’s far, far better to know what people do than what they say.

If you’re an avid movie goer, for example, it’s great for the movie producers that you Tweet. How great? That’s just the thing--no one really knows. We assume it’s probably good, but is it $5 good, or $5,000 good? How can you plan a media strategy when you can’t confidently measure the impact of the action?

The better question is, if you bought a ticket, how many more tickets will your friends now buy? Imagine if you know that answer, and suddenly you really can say that the impact is $5 or $5,000. Now you start to realize that the value of a particular movie goer may be their own spend plus some number of friends’ tickets that were not going to have been bought. The person who is able to bring five friends with him to every movie is much more influential than the one who sees the movie by himself -- and you can assign an actual dollar figure to that influencer, versus an abstract number or a guess.

There’s a long precedent for this in social science, where observation--not speech--is the gold standard in research. In the 1950s, for example, a museum owner approached a local university to help them learn more about their exhibits and their patrons. What they wanted to know was which exhibits were the most popular so they could promote them and invest more in making similar exhibits. The first team used a survey and asked everyone what they liked. Oh, we loved the exhibit on Ancient Chinese Art and we loved the one about the Birds, people said.

One of the investigators had been walking around the museum and these answers didn’t ring true to him. He didn’t actually see a lot of people at those exhibits and he did see a lot of people in other places, like the exhibit with the topless women of the Serengeti. He guessed, correctly, that the people weren’t giving accurate answers. And to find the truth, he started taking measurements that the patrons couldn’t see. What he did was to go exhibit by exhibit and count the number of nose smudges on the glass and the amount of wear on the floor tiles. These were a much more accurate gauge of what people actually did. And it turns out they liked the dinosaurs and the exhibits with the topless natives running around.

This story is a pretty good analogy to games and data. Developers run the museum. They see all the actions. They shouldn’t just rely on what players say -- though, of course, they should listen. But how do they know what’s really going on?

Good analytics fill this gap for developers, but great analytics tell them what they players are going to do next, why and how they impact each other. This is where predictive analytics meet Social Value: algorithms learn the behaviors of players and calculate the Social Value based on their actions.

The possibilities for these analytics reach beyond the world of games. Obviously, this approach works best for companies with larger amounts of users who are relatively social. If no one is playing/watching/buying/listening together, or not sharing their experiences, then this social influence is going to be a lower portion of bottom-line revenue. But, where users tend to influence each other more around a given product on a somewhat regular basis, the algorithms will show more, and are able to learn and predict more quickly and accurately.

Instead of focusing on what the players are saying, targeting the players who matter most based on their actions -- the leaders versus the followers -- is how analytics will boost ROI.

Standard influence scores (like Klout, etc) deal with what people say (or what is said about them), but instead of what they talk about, Social Value looks at what people do.  The Social Value algorithm looks at the relationship between people within a system and sees if there is a ripple effect: when one person does this, do their friends do it? In the end, if it isn’t transparent and provable, it’s not trustworthy.