How Can We Measure Influence? The Coffee Shop Analogy
Let's say we have someone who goes and gets coffee. Imagine that the coffee shop has perfect records of who comes in and buys what. So we know that this woman--we'll call her Susan--comes in periodically and buys a coffee and spends $10. But we also know that sometimes she comes in with another women--we'll call her Brenda. And when she comes in with Brenda, she stays longer and spends $20.
So what we have here are two sets of data, and we can compare them. This is exactly what an experiment is. Everything is the same between two cases except one thing. And in this case, that thing is Brenda. Brenda being there causes Susan to spend an extra $10. Notice that I said cause, not correlation. If we have enough data and enough times they are in the coffee shop with and without each other, we can be very certain about this. So we have an actual measurement of influence, in dollars, that we can assign to a person. And for Susan, we now have part of her spend which we know is caused by someone else. We call that her receptivity.
Now in the back of your mind you are probably asking "hey, how can we look at sales receipts and know that Susan and Brenda are friends?" That's the right question, and the answer is we don't. So what we do is we look at this possible pattern among every pair of people we have. We look at Susan and customers 3 and 4 and 5 and so on, and we find that there is no relationship, no influence, no causality. For the vast majority of these pairs of people we look at, we get noise. We waste some computing cycles, and that's OK. They are cheap. But if there is a real relationship there, that emerges from the noise and with enough data, we can start to get confident about it.
So that's how we measure impact and receptivity for every person in a dataset. You may be wondering whether this works outside of a coffee shop and the answer is yes.
We've now done this with over 3 billion consumer data points in industries as varied as charitable donations, streaming video, retail and video games. And what we find is a remarkable pattern. The amount of socially driven spend or activity is around 30 to 40 percent everywhere. It's just what people do. And so now we move from "isn't that interesting" to "how can we make that actionable?"
