The term “predictive analytics” may conjure thoughts of an over-hyped promise by technology to forecast future events. The phrase itself sounds like a boardroom buzzword designed to mystify and dazzle. The fact, however, is that predictive analytics is not hype; it really works.
It may be unfortunate that predictive analytics is called “predictive analytics.” A more accurate and less buzzwordy description would be “applied statistics.” After all, statistics are easy to understand, and better convey the concept behind the predictions. Take a look at an obtusely simple example.
Jim owns a gym. The gym offers weight machines, cardio machines, group classes, and some personal training. Jim’s Gym also has showers with complimentary towels for after a workout, as well as Smoothie Bar. Jim’s Gym wants to retain their customers year-round, and add new customers. Unfortunately, Jim’s Gym plays a game of attracting new customers around the beginning of the year and the beginning summer, and sees high churn towards the end of Q1 and as summer ends the holiday season approaches. This is the most basic form of predictive analytics, or “applied statistics.” Using a Linear Regression model, we can predict with a high degree of accuracy when customers at Jim’s Gym will drop their weights and walk out.
The previous example was a 2-dimensional linear regression model (a fancy way of saying a ‘classic line chart’). However, business and life are rarely lived in 2-dimensions. Instead, we find we work with tens of dimensions, and sometimes hundreds. For example, what if Jim decides to personally call each and every one of his customers in February and August to check on their progress, in an effort to reduce churn?
On a 3-dimensional plot, we compare the effects of calling each customer on churn with the base rate, and the corresponding month for each. We can see that the campaign reduced churn in Q1, but actually increased churn in the latter part of the year! What gives? Well, it turns out that there were several other dimensions involved that were not accounted for in this simple 3D plot. Jim’s Smoothie Bar reduced their fruit offerings in July…so that must explain the increased churn. But wait, Jim also provided free 1:1 personal training in July, so that should have reduced churn. However, this was a particularly hot summer, so some people canceled their memberships to stay indoors…but gyms are indoors…but Jim’s Gym has fans and swamp coolers, not air conditioning. Also, around this summer period, Jim added 4 additional Zumba classes, cut 2 yoga classes, added more protein options at the Smoothie Bar, reduced the hours of the Smoothie Bar, added more personnel, stopped the free towel program, and twelve other factors changed.
Understanding these changes, we now have 24 dimensions to analyze. And while 24 dimensions may seem an insurmountable problem, at the core, it’s still a statistical analysis…albeit a complicated one. I don’t know about you, but my head starts to hurt after the third dimension. Humans are good with two dimensions and sometimes three, but not many more. Computers, however, can handle hundreds of dimensions. This is where predictive analytics works well: To utilize the power of computers applied to statistical algorithms to analyze multidimensional data with the goal of finding a pattern. That’s it! It’s just that easy. No magic. Just plain ol’ statistical algorithms.
In the above example, a predictive analytics model may have discovered that the calling of the customers has no effect whatsoever on churn, and the increased churn was the result of a combination of decreased yoga classes and a reduction of the hours of the Smoothie Bar. It may have also predicted that 1:1 personal training classes also reduced churn. That’s the power of applying computer power to statistical algorithms to analyze multidimensional data with the goal of finding a pattern. Simply put: It’s been proven time and time again to work by the leading Fortune 100 companies, like Google, Amazon, Microsoft, Apple, and more.
There are many statistical formulas and algorithms utilized when creating predictive analytics. Some are relatively new, having been invested in the 21st century, and some are hundreds of years old, classically applied to two- or three-dimensional problems. Some of the algorithms are surprisingly simple in concept, and some are a bit more complicated in nature. Yet they all share a common thread: They have all been proven by large corporations to work with a surprisingly high degree of accuracy.
Google, Amazon, Microsoft, Apple, IBM, HP, Netflix, Chase, Ford, Morgan Stanley, Tesla, and dozens and dozens of others have successfully integrated predictive analytics into their core business model. Not just recently, but over the past 20 years or more. When these corporations realized that classical statistical methods could be used to predict future events, they hired teams of PhD’s and scientists to write the algorithms necessary to implement the promise of predictive analytics. And, after years of research, millions of lines of computer codes, and hundreds of millions of dollars, the dream became a reality. Netflix uses predictive analytics to predict what shows you will like. Amazon uses predictive analytics to predict what products you will buy. Morgan Stanley uses predictive analytics to predict what stocks will rise (and fall). Chase uses predictive analytics to predict your credit worthiness.
The heavy lifting is over. The large corporations that have invested hundreds of millions of dollars in research and time have produced a toolbox of algorithms that can be utilized by small businesses and entrepreneurs. The millions of lines of code are all encapsulated behind the scenes. In fact, a powerful predictor can be written in as little as 3 lines of code:
regr = linear_model.LinearRegression() regr.fit(X_train, y_train) prediction = regr.predict(X_test)
Of course, a trained Machine Learning Engineer who understands how the various algorithms work at the core and how to apply them to data, as well as a trained Data Analyst who can recognize which data to use and which to discard is needed and essential to producing effective predictions. After all, while there are machines to perform laser surgery, it’s best utilized by a trained surgeon to ensure accurate results.
What used to take years and millions of dollars now takes weeks to months at most, and at a cost that’s appealing to every business owner. Utilizing the same algorithms used by the Googles and Amazons of the world, with the same great results, but at a drastically reduced timeline and cost, small businesses can now harness the power of predictive analytics.
How many dimensions of complexity occur in your business? What trends in sales would you like to observe? What patterns and hidden messages are waiting in your data to be discovered? Predictive Analytics is the solution to reveal the patterns in your business that will greatly aid in informing future decisions and predict trends.
Here is a prediction: Within 3-5 years, every small business with utilize predictive analytics to remain competitive and relevant, much like they currently utilize websites for advertising, accounting software for accounting, and email for communication. Those ahead of the curve that adopt the competitive advantage earlier will reap the rewards earlier. The Machine Learning Engineers and Data Analysts at SerpicoDEV understand predictive analytics at its core, and have successfully applied it for satisfied clients. Contact us for a no-charge “data checkup” to analyze your data to determine which predictive models are the best fit for your small business.