This information helps with forecasting. Based on past demand, you know how many products you can expect to sell in the same time frame in the future. However, forecasting only goes so far. Factoring in other variables – market forces, political upheaval, changing customer preferences, etc. – is trickier, yet this is what the brave new world of predictive analytics promises to do.
The big business intelligence software vendors have begun rolling predictive analytics features into their main software suites. The same organizations leading the BI market – SAP, SAS Institute, IBM and Oracle – are also the ones paving the way for predictive analytics.
Yesterday, for instance, IBM opened a predictive analytics lab in China. This is just the latest in an estimated $12-billion commitment to build out IBM’s analytics portfolio.
What Exactly is Predictive Analytics?
Everyone makes predictions, but what makes predictive analytics worthy of big-dollar investments?
To put it simply, key elements of predictive analytics have already been proven. Take traditional business intelligence, combine it with data mining and add on statistical analysis and you have predictive analytics. Math geeks will squabble over the nuances, say, whether a specific model is a predictive, descriptive or decision-making one, but for most organizations this boils down to using historical data along with probabilities to better assess the future.
Organizations already do plenty of forecasting, obviously, but predictive analytics attempts to put hard numbers beneath what are typically little more than educated guesses.
Tonya Balan, manager of the analytics product management team at SAS Institute, offers an example of how predictive modeling is different from simple forecasting. Forecasting will tell you that you’ll sell more ice cream cones in July than other months of the year.
Predictive modeling, on the other hand, will tell you the characteristics of ideal ice cream customers, the flavors they prefer and what sorts of marketing efforts will resonate with them.
Predictive Analytics – Roots in Behavior
Arguably, humans make their way through the world on the basis of predictive analytics. Skipping the interstate for surface roads because of heavy traffic is a simple version of predictive analytics. Watching a baseball game is an ongoing lesson in predictive analytics. In a hitter’s count, you expect the pitcher to throw a fastball. A lot of information actually goes into that prediction, and the statistics back it up.
Baseball is known as a stat geek sport for a reason. The trouble is that the game of baseball has many restrictions on it. It isn’t an open-ended system. Trying to figure out what exactly a hitter’s count is when it comes to, say, selling IT software during a recession is a much more difficult proposition.
That doesn’t stop organizations from trying. Think of lending, for instance. Your credit score, which looks at your past behavior to predict whether or not you are a good credit risk, is an industry standard. The inputs are small, however, and the outcomes are limited.
Most organizations need more sophisticated models for such complicated things as customer retention, supply chain management and the development of new product lines. And the more complicated a prediction becomes, the more risk there is for garbage-in-garbage-out statistics.
Using Social Media to Glean Predictive Analytics
According to Gartner, one of the ways to get better inputs is to leverage social media. Social media can deliver all sorts of real-time data from employees, partners and customers.
“Social software allows users to tag assumptions made in the decision-making process to the BI framework,” said Kurt Schlegel, research VP at Gartner.
“For example, in deciding how much to invest in marketing a new product, users can tag the assumptions they made about the future sales of that product to a key performance indicator (KPI) that measures product sales. The BI platform could then send alerts to the user when the KPI surpassed a threshold so that the decision makers know when an assumption made in the decision-making process no longer holds true. This approach dramatically improves the business value of BI because it ties all the good stuff BI delivers (e.g. analytical insights, KPIs) directly to decisions made in the business.”
With business-class collaborative software, this makes perfect sense. While information still needs to move from one application to another, it’s a rather straightforward migration. However, what happens when you want to get broader insight from blogs, user boards, Facebook, etc.?
So-called text mining may well be the next frontier of business intelligence software, and the U.S. government is sinking a ton of money into it. But it’s a technology lurching forward with mixed results.
Obstacles to Predictive Analytics
This problem simply highlights one of the most pressing obstacles facing predictive analytics: data lock. It’s all well and good to say you’ll leverage social media, but when it comes time to pull out the data and actually do something with it, today’s BI suites aren’t quite up to the task.
Today, the big chore is getting data out of data warehouses and BI suites and into forecasting software. The big BI vendors are all working on integrating predictive features into their core BI suites, but it’ll take longer before they can pull in information from wherever an organization deems valuable.
The wisdom-of-crowds insights gleaned from the real-time web will have to wait.
Another serious obstacle is making sense of the data itself. Most knowledge workers aren’t statisticians. The trick here is to represent data as something other than raw numbers and formulas.
Visualizing complex data is critical if predictive analytics is to ever catch on. BI vendors all claim to be working on better representations, but the jury is still out on what will effectively engage average knowledge workers.
Warning from Recent Events: Quants on Wall Street
One of the dangers of any sort of forecasting is getting too enamored with models that truncate reality, limit inputs and outputs and then claim to accurately predict the future.
Wall Street “quants,” those doing predictive analysis for financial firms, were blindsided by the economic collapse. Few saw it coming, and few of their models showed even hints of problems to come.
Financial models underestimated risk, overestimated growth, and greatly overestimated the “rational” behavior of investors. Will organizations using predictive analytics learn from epic failures like this one? It remains to be seen.
What I’ll predict is that as predictive analytics catches on, smart organizations will work hard to verify the “inputs” to their formulas, constantly questioning basic assumptions. These organizations may seem cautious and conservative, and they may catch some flak from Wall Street. But in the long run, they’ll be a much safer bet than those predicting nothing but smooth sailing ahead.