Michael Koukounas, senior vice president of global scoring and analytics leader at Equifax Inc., will be presenting at GlobalDirections 2013 in Washington, D.C., this month. In this Q&A with SearchCIO, he previews his talk on leveraging real-time data analysis and the six steps he has formulated to build a powerful analytics strategy.
Michael Koukounas: You have to set up analytics to drive toward the business issue you have. In many cases, you have to start at the back end of the process and work your way to what are you're trying to solve. What are the business issues? What are the touch points? What are the constraints around the touch points, and work your way back to, what data do I want to use and get access to?
That's the reverse of how most people think about it. Most think there's this huge pile of data and I'm going to dig into it to find these big nuggets of gold. What I've learned over the years of doing analytics is that's not the way it works. You really have to understand the business problem you're trying to solve. Then you have to understand what the touch points are. What are the touch points by which you are going to impact the revenue -- impact from the perspectives of what people are buying, what decisions get made, where you're going to spend money?
Once you understand the business issue and the touch points, then you think about the constraints around the touch points, which could be a system, a regulation or a reputational constraint. In many cases, constraints are real-time information. The ability to get information to the right point in time to make decisions is critical, because you only get so many bites at the apple.
I would rather have analytics that is not as complex but still moves the needle and works within the environment I have today.
senior vice president of global scoring and analytics leader, Equifax Inc.
Do you have an example of how real-time data analysis can be used to address a business issue or opportunity?
Koukounas: I worked with a home improvement company that only closed one in 10 deals. Customers would come in to price out carpeting or windows or appliances and, nine times out of 10, they walked out without buying. If [the company] can move that decision to one out of seven, that's billions of dollars in revenue. A lot of those companies use instant credit to get people to make a decision, to get them to upgrade from maple to cherry cabinets, for example. It's the ability to communicate with that customer though their smartphone while they are in the store, to let them know then that they have access to this credit or a promotional offer, versus in-store advertising or through the mail.
How are real-time data analysis and big data strategies complementary?
Koukounas: A frozen food company that delivers food to your home has 3,000 sales agents in the field. What they want to do is increase their penetration and customer base. To do that, they want to know who their best customers are and what the next best offer to make to that customer would be. That's the business case. Then, they have to think about the point of contact with the customer, which are the 3,000 salespeople that are out doing their rounds.
If I'm going to find or solve for [the best customers and the best offer to make], I have to look at historical customer information -- what do they buy historically? -- and then synthesize that information every day and push it out to the salespeople. [I have to] push it out in a way that identifies the route the salesperson will be on that day and then create the promotional offer that targets specific customers on that route based on their buying history.
Are the analytics technologies available today suited for real-time data delivery?
Koukounas: The problem isn't that there isn't existing technology; the problem is there's tons of legacy technologies that become a constraint. I could build the best analytics [platform], but it would be like building the pyramids. I would rather have analytics that is not as complex but still moves the needle and works within the environment I have today. Most times, [the solution] is a combination of new and old. A company does something offline in the cloud with these new high-performance environments, and then they're feeding the touch points that are, in many cases, legacy systems.
You will be talking about six steps to building a powerful analytics strategy at the conference. You mentioned three already: identifying the business issue you are trying to solve for, identifying the touch points and resolving the constraints around those touch points. What are the three other steps to building an analytics solution?
More on real-time data strategies
Real-time data analytics in sync with the online music industry
Know thy customer, real-time
Analyzing data in real-time calls for business and tech savvy
For more information on the conference, see GlobalDirections '13
Koukounas: The fourth step is understanding the data -- what data will help you solve that business problem? -- and starting to think about the right data. The next step is designing the analytic [solution] and how you're going to set it up. The last step is executing.
How do you begin to set up an analytics solution?
Koukounas: The 80/20 rule always works: I can get 80% of the answers with 20% of the work, but it takes a lot of work to get to that last 20%.
There are a lot of different types of data. You're solving for a problem that could be solved with structured data, for example. Then someone wants to add unstructured sentiment data from Twitter. The introduction of unstructured data will increase the complexity of the analytics significantly, expand the amount of time, and you don't know how much added value it will bring.
Is that the right thing to do? Do you want to add that complexity by going after structured and unstructured data like Twitter feeds? In certain analytics, that type of unstructured or sentiment data can be very useful but, in other cases, it adds implementation risk and duration to the project.
So when you think about structuring your data, you're getting analytics. You're trying to figure out, "How much research and development do I want to do?" and "What's the right combination given the impacts of the analytics and what [I'm] trying to solve for?"