Prescriptive Analytics Is Easier And More Profitable Than Predictive Analytics
JeanFrancoisPuget 2700028FGP Comments (4) Visits (21436)
When you hear about algorithms these days, chances are that you hear about machine learning or predictive analytics. (Some make a distinction between machine learning and predictive analytics, but the distinction is not material for this post. I'll use both interchangeably here). A quick search returns recent discussion in the news of machine learning algorithms: Usin
These are mere examples of a general trend: predictive analytics is becoming mainstream. People start fantasying about its power and possible danger. Yet, there is little discussion on its limits. A good recent example of the latter is Lear
The difficulty of predictive analytics has been captured by Gartner in this picture. (See How Does Cognitive Computing Relate To Analytics? for a description of descriptive, diagnostic, predictive, and prescriptive analytics.)
I like this chart except for one thing, its horizontal axis label. While I definitely agree that predictive analytics is harder than descriptive analytics, I strongly disagree that prescriptive analytics is even harder. When I tweeted my opinion, I got an interesting reaction.
That's the point on which I disagree.
Prescription does not necessarily require prediction.
Let me give some examples I was exposed to.
I could go on and on. None of these examples require prediction. They are about optimizing the use of resources to meet known demand. Moreover they yield large return on investment, see for instance Extreme ROI with Optimization .
This said, it makes sense to combine prediction and prescription in some cases, see Price Optimization for an example. See the comments for more examples. I do believe that this combination is a largely untapped opportunity in general. But that should not lead people to think of optimization only after they have successfully deployed a predictive analytics application. It should be the other way round IMHO.
Indeed, you often get shorter return on investment by focusing on optimizing what you know before trying to predict what you don't know yet.