What happens when you combine Moneyball and Draft Day?

And don't worry I don't talk here about Brad Costner or Kevin Pitt. But more about the impact of analytics on drafting decisions. Today and tomorrow are draft days and more than 200 college football prospects hope on these two days to be drafted by the 32 NFL teams. While the popular movie with Kevin Costner surprisingly tells us little about which criteria and analytics professional football teams use to select players, Moneyball with Brad Pitt (being anchored in the Baseball league instead) gives us a glimpse at how much data and analytics help these days teams, managers, and coaches to select the right players based on (objective) field data.

Did you know that some of these models actually rely on classic Attribution models (like the popular Markov Chain models) we commonly rely in marketing to understand how much different touchpoints along the customer journey contribute to purchase decision? Basically the idea is the same. You have a set of players that interact to reach a goal (in the NFL that`s moving the ball forward or making a touchdown). While the player who is placing the egg in the competitors goal area is certainly the one that gets (together with the Quarterback) most of the spotlight, we must agree, that these two people usually can't shine much if the rest of the team does not provide a playing field that blocks the opponents from interacting, or opens the opportunity to run forward. The question is thus similar to marketing: How much does everyone's action on the field contribute to the outcome. So how much do I need Youtube to get consumers attention and start the journey, how much does an email ensure that consumers have enough information, how much does my SEA contribute to forming the right believes and how much is then the real impact of my last touchpoint (e.g. a banner ad the customer clicks on right before coming to my e-store and buying my product).

In Sports Analytics the game looks exactly the same. Without some people blocking and conquering the ball, there is no offensive. Here we start with the next offensive move (i.e. we open a customer journey). Some players block the other team, some players position themselves, the quarterback decides who to pass the ball to, and this player hopefully makes the touchdown.

Markov Chain models can help us with measuring the contribution of each player. To make things a bit easier to understand, let me really on another popular US sport: Basketball. If you are my age, there is of course only one Basketball team that is worth looking at. The Chicago Bull with his Royal Airness: Michael Jordan. But you will remember that while Jordan was scoring most points, he had a loyal team around him, with players such as Scott Pippen and Denis Rodman. So how much did they contribute to a games final outcome.

Let us assume the team scored 60 points and we can look at all game passes and their outcome in the game.

In 3/4 of the moves Pippen starts, in 1/4 of the cases Rodman starts. If Pippen has the ball, he passes in 2/3 of the cases to Rodman and only passes in 1/3 of the cases the ball directly to Jordan. In case Rodman got a ball, he passes in 2/3 of the cases the ball to Jordan and in 1/3 of the cases he tries to score but fails. If Jordan has the ball the scores in 50% (1/2) of his attempts. Of course this is purely hypothetical, as his airness was much better ;)

So what is the total scoring probability? We can easily calculate it by going down all paths that end with the ball in the net.

So the total probability to score in our example is 0.375. The Markov Model does now try to figure out what happens if we take players from the field (or in marketing words we do not use a specific touchpoint in the chain). Let us assume we do not have Pippen. We then can only use moves that go through Rodman or Jordan.

The scoring probability goes down to 8%. In case we take of Rodman the corresponding probability is 0.75*0.33*0.5 which equals 12.5%.

We can now calculate the contribution of each player with the following scheme.

We first calculate for each player the removal effect (which is 1 - the scoring probability w/o this player divided by the total scoring probability). We subsequently sum up the removal effects and divide the player specific removal effect by the sum of effects (which is just the relative impact weight). Finally we can apply this weight then to the game's final outcome. Remember that we said the game ended with 60 points for the Bulls? So how much did every player contribute now? The last column shows how much of these points go to each player's contribution and gives us a fair chance to evaluate how much everyone on the field contributed to the game's outcome.

Interested in understanding more how Attribution Models and Analytics work and how you can leverage data to come to better decisions? Check my book with Gokhan Yildirim "Applied Marketing Analytics Using R" that provides you not only with plenty of background but also with code snippets and data sets to directly train your analytical skills and get your ready to meaningfully contribute to your organisation's success. Or just follow me on Linkedin!