Folding a predictive analytics strategy into a business intelligence system poses challenges, especially for those companies operating on a tight budget. But, according to Eric King, president and founder of The Modeling Agency, a consulting company based in Pittsburgh, Pa., businesses will need to face those challenges eventually, if only to stay competitive.
King warns, though, that one of the worst analytics mistakes a business can make is plunging into the deep end of the predictive analytics pool.
“I hear so many people, when we have an introductory webinar, so fixated on software,” said King, whose firm specializes in data mining and predictive analytics. “And we try to very politely tell them, ‘Look, you’re on page 5; we’re back on page 1. We’ll get there.’”
SearchBusinessAnalytics.com sat down with King to discuss how businesses can get started with predictive analytics and how to overcome challenges such a deployment poses.
Should all businesses be using be predictive analytics?
Eric A. King: I believe all businesses will be utilizing predictive analytics and data mining as another standard process within their overall business intelligence operations. There are large organizations that have gotten there first and earlier and have pretty much operationalized it. Those that are either behind the curve or maybe smaller companies that have not progressed as far down what I would call the business intelligence path will get there at some point. They’ll need to.
Why will businesses need to?
King: One reason, of course, is to extend their competitive edge and not fall behind. But another is that there are so many cost centers within business intelligence that don’t really provide a reward to the organization. There’s substantial investment in just collecting data on every transaction and then storing it, structuring it, understanding how to access it and consolidating it. And then [there's investments in] ways in which to interpret the retrospective view of the queries they make on that data, ways to visualize it and more effectively interpret it. But until they start getting more of that perspective view to anticipate behaviors, that’s when they’ll be able to operate far more efficiently both in generating revenues as well as cost savings. All businesses are moving toward that area of practice, but it’s not what you begin with.
What makes putting together a business intelligence operation so difficult that some businesses are behind the curve? Where does it fall apart?
King: I don’t know that it falls apart. It just takes a lot to progress through. One of the things I think may have tripped up the timeline a little bit is the large data warehousing initiatives in the '90s and early 2000s and then they became legacy systems. There were so many disparate data marts and databases around the organization that had to be consolidated. So there’s a lot of reinventing that had to occur before they could continue their path down the business intelligence chain. That stalled things.
And what stalled predictive analytics, specifically?
King: I’ve been in this field exclusively for 20 years, and it’s been the slowest moving yet steadily moving trend I’ve experienced across any technology area. The forces for progress with predictive analytics outweigh those that detract, but they almost counterbalance. The forces for is that organizations continue to succeed in predictive analytics. Many of them write up compelling case studies, and organizations see there are substantial gains to be made if they properly apply predictive analytics.
The forces against are that it’s complex, cryptic, intimidating, not natural to know where to begin or how to properly traverse the process. Yet, the ironic thing is, there are a number of industry standard processes available at no cost. They’re phenomenal at expressing how to properly maneuver what is essentially a discovery process that must be managed in a way that’s different from many other IT projects or even other business intelligence initiatives.
For businesses interested in deploying predictive analytics but living on a tight budget, what would you say is a good first step?
King: It’s the same recommendation we would make with an organization that has set a large budget aside for predictive analytics. I’ll start out with what not to do. A majority of those who proclaim themselves data mining practitioners make the big mistake of starting by collecting data resources, buying some software, applying the data resources to the software and hoping for some compelling or tangible or impactful results. The majority of the time that just won’t happen.
What does happen?
King: Clients will build excellent models with modern software that end up answering the wrong questions, aren’t understood by executive leadership, aren’t adopted by users or supported by IT or integrate well with the environment. There are many other reasons that predictive analytics and data mining fall short at the strategic or project level, not really at the tactical level.
So, where should organizations begin?
King: If a client puts aside a quarter of a million dollars to start in with a data mining initiative, we’ll tell them to hold onto the money. Don’t go and invest in some expensive software package. Perhaps you’ll need it at the right time, but first begin with training. It’s important for people within the organization to have a solid understanding of predictive analytics and data mining standard processes, its capabilities, its limitations, its risks and rewards and how to traverse that discovery process under a formal approach.
You would say to get that education externally?
King: I don’t think that exists in most organizations. As a matter of fact, it’s rare even in the marketplace. We’re one of the few providers of vendor-neutral training for data mining and predictive analytics. But the other parts they need to look at are the strategic aspects: How do we develop a project plan? How do we perform a comprehensive assessment of our resources, of our environment, of the objectives of leadership? Does that align well with functional managers? Do the resources support the stated objectives? If not, how far are we from the starting line? How will the models being inserted on the other end be impactful? How are we going to measure it? Those are the things people gloss over and just dig into the project but that should also be included in the training. So, step number one is training to understand how this particular practice differs from others, and how to properly approach it. From there, proceed with the assessment and project definition.
How long does it take to put together a comprehensive assessment?
King: The Modeling Agency does it within two weeks of the on-site meetings. That industry standard process we follow, called the Cross Industry Standard Process for Data Mining, spells it all out; it’s a cheat sheet. We have the experience to know how to run through it and ask a lot of the questions to fulfill it. When that’s undertaken properly, we’ll be able to rank potential pilots that will have the greatest impact for organization.
How do you recommend businesses budget for a predictive analytics deployment?
King: They want to budget for this a year in advance. It is moderately scalable because it’s highly situational. My advice would be to set aside $20,000 to $30,000 for the comprehensive assessment to be guided through by that expert. That assessment is then going to give a much more accurate forecast of the remainder of their investment.