Predictive Analytics: Looking to the Horizon
But traditional business intelligence (BI), such as end-user query and reporting, basic OLAP (on-line analytical processing), etc., provides mountains of data on what has already happened. In other words, the things we can no longer affect.
What has been missing for years is information about the things we can still change. It's an odd way to do business. BI seemingly gives the business only eyes in the back of its proverbial head.
This is changing, though. Predictive analytics software from vendors such as SPSS, a provider of predictive analytics software and solutions, and SAS, a BI vendor, is helping business managers spot commonalities, trends and associations among customers that they never would have seen before.
This, in turn, allows them to refine their selling efforts. And it's doing it in real time, which means companies can react quickly and take advantage of those trends rather than finding out about them after the fact.
The growing market for predictive analytics (as much as $3 billion by 2008, according to some analysts) is putting BI eyes in the front of the proverbial business head. And training those eyes on the horizon.
In other words, BI is becoming less about storing and regurgitating data, and more about creating knowledge. All it takes is the desire to learn more than you ever thought possible about your customers and the right tools to extract the information.
Peepers on the Past
Traditional BI started out as a means to use historical data collected over a period of time to predict trends. Analysts would spin through a mountain of data and use their business knowledge to determine future strategies. This methodology is still the most popular today.
While typically better than pure "seat -of-the-pants" guessing, this method is still limited by the knowledge and/or interests of the users. In some cases, it follows the adage "research proves what the researcher set out to prove." Business users start with an assumption and then use data gleaned through BI to support the conclusion.
It also smacks of a military truism: "Generals always fight the last war." Traditional BI allows business users to gain great insight into what worked (and didn't work) in the past in order to make decisions.
Unfortunately, it doesn't allow for a change in market conditions, customer base or other factors that might influence the future. Like the French digging trenches at the beginning of World War II, you may be setting a strategy for conditions that no longer exist.
The thing to keep in mind when looking at historical data is the farther away an object is from a viewer, the smaller it appears. In the telescope of history, the farther away your data is from the present, the less significant it is to what's happening today.
Venturing into the Unknown
The advantage predictive analytics provides is the ability to go beyond your assumptions and discover things you wouldn't otherwise know. The key is the ability to recognize patterns or associations between seemingly disparate things.
One famous example (that may or may not be apocryphal) is the association between diapers and beer.
According to the story, a convenience chain looking at sales data noticed that when men purchased diapers, often they would also purchase beer. The supposition was that, if men were sent to the store by their wives for diapers, they felt they should pick up something for themselves as well.
This led to the conclusion that placing diapers closer to beer in the store will increase beer sales.
Whether the story really happened isn't important. What's important is that diapers and beer are not an association most people would make based on pure business knowledge alone.
Baby food, diaper wipes, formula, yes, but beer? In this case it was probably luck. Predictive analytics will find that association, and hundreds of others like it, because it doesn't use human assumptions. It simply looks for statistical patterns and tells you what it finds.
Data Under the Microscope
In science, a microscope is used to look for very small things that can have a large impact on our lives. Predictive analytics does the same thing. While it can identify the big trends that affect your entire customer base, it can also highlight small but highly profitable sub-segments of your customers.
For example, let's say you're selling commemorative plates. How valuable would it be to know that people in Wisconsin whose family income is more than $60,000 per year buy three times as many plates as any other customer group? Or that customers in the Green Bay area almost always by a plate with a U.S.A. theme, but rarely purchase one with a movie theme? And that they are four times more likely to make a purchase in June than in September? Do you think that would have an effect on your direct marketing efforts?
With traditional BI, it would be difficult to dig out this kind of data unless you already knew to look for it and ran a query or built a model to pull it out. Predictive analytics finds it for you automatically.
A good way to think of predictive analytics is as what-if scenarios on steroids. It spins through millions of pieces of information, finds the associations, considers the variables and then predicts what is likely to happen if you take a particular course of action.
Amazon.com is probably the best known user of this type of thinking. Once you register and make a purchase at Amazon, its predictive analysis engine starts churning. It looks at what you've purchased and looks at what else other people, who've also purchased what you've purchased, have bought for themselves.
The next time you return to the site, Amazon presents you with a list of merchandise you'd probably be interested in. In my case, I know their predictions are pretty accurate.
All of this happens automatically. There isn't a person at the other end making the decision. They simply use statistical probabilities and all the data at their disposal to tempt you into disregarding your budget and spending more money at Amazon.com.
Your Thoughts Matter
Most of this discussion has focused on things that are relatively easily quantifiable. Every purchase has an SKU number, and while deriving patterns from millions of bits of hard data is a number crunching challenge, it's still fairly straightforward.
One of the values of the better predictive analytics tools, though, is their ability to identify patterns in soft data such as comment cards.
If the simple text is input properly, predictive analytics can be used to identify patterns that can have a huge impact on the business. For example, if the words "customer service" and "sucks" show a pattern of association, you know you have a major problem that needs to be addressed, preferably sooner rather than later.
With a little digging, you can even find out if that's an association that goes across the board or only pertains to a particular time period. The point is you're able to see there's a problem and do something about it before those customers abandon you and turn to a competitor.
Here's another example. If the word "blue" shows up often in regard to a product line that doesn't have any blue offerings, it may be an indicator that you need to offer the product in blue. At the very least, it's worth considering, and it's something you wouldn't have known before.
Of course, at the end of the day a human being is still required to take action and make decisions on how to improve on a situation, good or bad. But at least the information is there and available, not hidden away in a file cabinet.
In next month's column we will discuss some of the factors you need to consider when deciding between BI and predictive analysis tools and the presentation layer of each technology.
Mark Robinson is a Business Intelligence practice manager at Greenbrier & Russel, a business and technology services firm specializing in business intelligence, custom application development and enterprise solutions. He can be reached at firstname.lastname@example.org.