A Decision Science Lens:
Predictive Analytics
A Practical View
Predictive analytics is all the rage. If you believe some sources, we are on the precipice of day when everyday Citizen Scientists will be capable of generating endless streams of predictive analytics. This, of course, will change the world. Or… is a ridiculous pipe dream that woefully over-promises a large volume of outcomes that have little practical value.
How can you say that?
This series of articles, this being the first of many to come, uses the lens of decision science to examine other concepts in the Data Science & Analytics field. When you consider predictive analytics from the point of view of a decision-maker, things can have a very different perspective.
From an academic standpoint, predictive analytics seems to be near the pinnacle of analytic excellence. Only prescriptive analytics rates higher (although academics might disagree). It also rates quite high on the science continuum. It is hard to think of a better validation than predictions. They bring analytics “outside the lab” as close to the scientific method as possible.
So why does a decision science lens lead me to call predictive analytics a pipe dream? Why would it have little practical value? Let’s be clear. I called the idea that armies of Citizen Scientists would be producing endless predictive analytics a pipe dream. Some predictive analytics will prove quite valuable. This article is calling these bold predictions to task. That still deserves further explanation.
What’s in a number?
Numbers are not the Holy Grail of analytics, no matter how popular that concept is. I don’t need to know the price of Apple stock in 36 months to make money. So predicting the price is not the right objective, from a business perspective.
From an academic or purists point of view, predicting the price allows me to far better quantify the models and science I used. That is surely worth something, right! Well, at least until my model’s predictive power wanes and I need to re-calibrate, re-validate, or re-engineer.
It is not just the direction either!
Some in business would say they only need to know the direction. I have been known to overuse the term “directionally sound”. It is WRONG or at least it doesn’t mean just getting the direction right. Simply knowing if a stock price (or any other thing) will rise or fall is rarely ever enough to make a solid decision. In our example, just consider the value in knowing Apple will rise $0.01 over the next 36 months… vs maybe $100? While hopefully a solid illustration, even those two orders of magnitude may be indiscriminate. Now there is a fun term.
So — How is predictive analytics best used to help business or society?
Now that is the right question! It is a tool for decision science. Part of decision science is developing the logic, the process, and the framework to apply to any problem. It starts with a great question. It is tied to measurable outcomes. It often includes both numbers and directions, but it is also tied to ranges, scales, segments, and trigger points. Show me the off-the-shelf predictive software that allows you to do all that! But I digress… often.
So — predictive analytics provides the most value when it is applied to a well-framed set of questions/decision points.
Does that make predictive analytics a lesser discipline?
NOT IN THE LEAST. People really need to get over their personal attachments to analytic disciplines, techniques, and buzzwords. You can’t execute data-driven decision-making without predict analytics. Or decision science. Or data transformation. Or any number of equivalent (not ancillary) functions. They are all important! Stop fighting over which is better and work on proper integration and control.
But won’t Artificial Intelligence or Machine Learning automate all of this?
OH, HELL NO! We are struggling to teach computers to analyze. While there has been plenty of progress in the last several decades, breaking things apart is a far different exercise than putting new things together.
In areas of forecasting, which is part of the process of synthesis, success is limited to only the most tightly defined and artificially limited exercises. Even then, plenty of groups like the Sante Fe Institute have long shown that humans still win. Actually, I believe they stopped trying.
Computers are faster. They are more accurate. If given a tightly defined task, they excel. Few important decisions fit that description. When they do, it is only after they have been developed through decision science.
What about Deep Learning?
Now that is a good question. Not quite the right one… but we will need to address that in a later article.
For now, recognize that all analytic disciplines have their limits. None is better than another unless the task is clearly and tightly defined. Some are certainly more hyped than others. So don’t believe the hype. And for Peet’s sake stop conflating everything!
And one more thing… thanks for reading!
