Going Exponential? The Difference Between Predictive And Prescriptive Analytics
I was recently approached by one of our most respected account executives to perform a half hour class at the Texas Government Data Forum, upcoming in late June. I am to deliver a breakout session entitled “Optimizing Program Enrollment with Machine Learning “.
In preparing the session, the State listed desirable topics to cover, including ‘prescriptive analytics’. I was assured by lay people less familiar with that term that they must have meant ‘predictive analytics’ instead.
In fact, they are different things, and prescriptive analytics are highly sought. A greater level of business domain knowledge is required to deliver prescriptive analytics, and while I am blessed to have an MBA on top of my Master of Science degree, most data scientists do not.
I thought little else of this interaction until the next time I donned my company t-shirt. We produce a t-shirt every year, complete with an annual slogan designed to inspire and motivate all the hard working folks here toward growing the business. This year’s theme is “Going Exponential in ’18”:
This, of course, presumes that exponential growth is desirable. This assumption, in turn, reminded me of a presentation I attended at Data Day Texas by University of Chicago Professor Kristian Hammond. Professor Hammond proposed a thought experiment that goes like this:
I have a friend, Bob, who is excellent at fruit arbitrage. If I hand Bob an apple, the next day, he will double that to two apples, total. He will do this consistently. If I strike a deal with Bob, where we split the take, should I harvest from Bob every day, or at the end of the week?
This was not a trick question, and Professor Hammond went on to declare that daily harvesting leads to linear growth of one apple per day, and weekly harvesting leads to exponential growth.
The point of this exercise is that many people, particularly ambitious people with less business acumen, will assume that exponential growth is universally superior. However, Professor Hammond points out:
Do you have a deal with Marie Callender to make pies?
Do you have a warehouse filled with nitrogen to keep them from spoilage?
Do you have a way to breathe as you enter said warehouse?
The point here being that, unlike with money, an increase of tangible goods comes the increase of costs associated with their storage, and the cost/benefit ratio must be considered. Professor Hammond then asks us to consider the linear growth alternative:
Have you ever heard the phrase ‘An apple a day keeps the doctor away?’
Well put, Professor.
Predictive analytics answer how many apples are likely in your future. Prescriptive analytics answer whether so many apples should be in your future.