Yesterday I concluded that “Existing theoretical frameworks assume too much, and demand too much cognition by the end user.”

The opposite of asking you to think about linear regression or support vector machines is Netflix.

Netflix uses a machine algorithm to suggest movies that you might like. They do this using a few sources:

  • When you first sign up, Netflix asks a few questions about you. 
  • They have a prior viewing history of all their subscribers before you, who also answered a few questions about themselves. Y
  • You tell them what you like by watching various movies and shows. 
  • You tell them more by rating them on a five star rating system.

By comparing your tastes to other people like you, in other parts of their library graph, they create a more relevant experience.

This is machine decision making at scale.

You benefit in three ways:

  • You don’t have to think by searching through a massive library of content (at least in the US).
  • You don’t have to think much about giving a movie 1 star or 5 stars.
  • You don’t really have to think, since starting a movie is a no-cost, low-risk procedure.

Netflix also benefits:

  • They don’t have to get into committee fights with their content providers about who gets prominence in the interface.
  • Humans don’t have to decide which groups see what, saving considerable resources.
  • The algorithm learns over time, increasing your satisfaction and generating lock-in.

There are huge benefits in delivering utility to people by not making them think.

People like easy. Netflix uses predictive analytics to make things easy.

Companies like Netflix, Google, and Amazon have big-n/little-thinking problems. They have a massive amount of content and need to figure out how to route the most relevant pieces to the right people. Predictive analytics at scale works very well for these companies.

Most companies have a small-n/little-thinking problems. They don’t have much content. They’re just fighting for scarce attention. Predictive analytics has a different application in this space. Same question – how do you reduce the amount of thinking that is required?

The routinization of business rule architecture is fairly well established in call-center and direct mail. That’s all done. If anything, the blanket application of rigid business rules has done more de-humanize and destroy customer relationships than anything else could have. This has been a colossal step backwards.

Customer centric predictive analytics systems are differentiated from their business centric cousins on that point.

In many ways, this is a solution hunting for a problem. It’s a different problem set. How do we turn this around on itself?

***

I’m Christopher Berry.
I tweet about analytics @cjpberry
I write at christopherberry.ca