You have been so nice to respond to my publishing of “old” work that was never shown that I want to continue doing that.
What follows is the first of a three part series (when you are breaking down a 3-4 page writeup into pieces three parts seems to be about ideal).
I was tasked last year to focus on how to make predictive better. I was never a fan of how predictive analytics was implemented (I am fine with the concept, but I don’t think anyone cares about the concept – instead deteriorating it into a parrot-act of repetitiveness with no good results).
In conducting the research came upon long-forgotten concepts and ideas, and mulling through them gave me a new idea: the one you will read about in next week’s installment
First, let me set the playing field.
Predictive analytics is finally changing.
An art form of sorts, revived by the recent interest in Big Data and analytics shown by global corporations in the past decade, predictive was never intended for the current uses.
The definition of predictive analytics puts it at odd with the current usage.
Predicting a behavior was not intended to be used as a harbinger of a customer’s intention to purchase but rather as a lagging indicator of an occurrence or event so that the knowledge could be used to build better analytics models.
The thought that an occurrence will repeat many times over because of past data points indicating a similar setup is criticized by many analytics experts – even when adopted by most organizations.
The difference is the narrowness of what predictive can do today. We are simply focused on one path, one way to get from point A to point B. If last time we were at point A we took a bus to get to point B, we will do the same today. The complexity of today’s world makes those “guesses” just about impossible. What if, for example, it is raining heavily and I am in a rush? Could I take a taxi instead? Or, what f I have time and it is a beautiful day? Could I walk? Or, what if I am with someone who owns a motorcycle? As you can see, the many variables that are traditionally ignored by predictive (we look for a pattern, and then try to repeat it when similar data points are recognized in the same sequence) are what make the new models far more interesting.
Keep in mind, this is not what predictive intended to be – but what became from the poor implementations along the way.
A successful bad implementation will be repeated. A failed good implementation never sees the light of day again. This is how Twitter came be used for Customer Service (but I digress)
Instead of trying to predict behavior step-by-step as most predictive applications do, why not use the pattern as a loose guideline of a sought outcome, break down the steps, and consider the many options available at each step. What yonder could’ve been a monumental step in calculating and analytics is very possible today thanks to advances in data capture, storage, management, and analysis.
The “Big Data” era brought the capabilities to analyze just about any data set in real-time and add many more variables as part of the analysis yielding far more interesting insights.
And it is within this new approach that we find not predictive analytics – but anticipatory analytics: the ability to dynamically and actively generate insights at each step of the way based on previously impossible to include variables and elements: intent, decision-making by users in real time, and untold goals and objectives.
As a result, my phone may hail a taxi for me (and maybe offer me a discount) if it detects rain or nudge me towards walking on a nice day – not because I did it before, but because I am about to do it. This is where predictive transforms to become the art of the possible.
What does anticipatory analytics look like? Come back next week to see…
(Cross-posted @ thinkJar)