We use cookies and other similar technologies (Cookies) to enhance your experience and to provide you with relevant content and ads. By using our website, you are agreeing to the use of Cookies. You can change your settings at any time. Cookie Policy.

ebizQ's Business Agility Watch


The Competitive Advantage of Predictive Analytics: Talking With Information Builders

Vote 0 Votes

Listen to my podcast with Dr. Rado Kotorov, Chief Innovation Officer for Information Builders. Rado is responsible for emerging reporting analytic and visualization technologies. In this podcast, we cover the exciting field of predictive analytics, how police departments are using it to head-off crime, and how companies are using it for competitive advantage.

Listen to or download the 13:10 minute podcast below:

Download file


PS: So now at a base level, exactly how does predictive analytics work?

RK: It is a very natural to our way of living and thinking activities so let me give you an example with a base activity that has contributed in our evolution. Let's take hunting, for example, what people do with hunting. They observe an animal, they observe the patterns of behavior, they start learning about how it moves, how it hides and etcetera. And based on these patterns, a hunter actually develops knowledge about how to setup the traps. So that's throughout life in every type of activity we have similar desires to recognize patterns and to base our actions on those patterns.

In medical science for example, we recognize which treatment algorithms help patients and we begin to identify side effects, improvement effects and etcetera and these allow us to essentially predict which treatments are suitable for what diseases. On another level, sitting in a street café in Paris and watching by passers, after 30 or 40 minutes a person would start recognizing the different colors and different fashion trends. So all of this, in all these three examples essentially what the brain does is it extracts patterns from historical data, what we've observed, what we have seen and we use the knowledge of these patters to apply to predict what is going to happen next.

So predictive analytics now using software allows us to expend dramatically the scope of this ability so we can take a lot more variables into account. So for example, if we're looking at fashion trends, we can take 200 various attributes and make very detailed predictions about what is a person likely to buy and fit their personal tastes. A pure observation won't give us that accuracy. So technology in essence is helping our natural activities.

PS: Very interesting. Now what are some of the more unique applications of predictive analytics that you've come across?

RK: The most probably unique one that I've come across is in child welfare and that's for matching foster parents with kids. The key problem in this area is there are about - let's say take a particular state like Michigan has 18,000 children in a foster care program. And the key goal is to maximize the length of stay with the foster parents. If a child drops out and changes foster parents too quickly, there are at risk at falling out of the program and become a permanent burden on society.

So again, how does a caseworker decide which parents, which foster parents, will be the best for a child? There are hundreds of variables like income, age, education, ethnicity, and etcetera that are associated with each foster parent candidate. So a predictive modeling comes here and essentially uses historical data to recognize which children match better with certain parents. So now, we can provide an application so that when a child comes into the program, the algorithm is deployed and it estimates which of the candidate foster parents would be best for this particular child.

Again, the goal here is to assist and guide the caseworker and provide them with more insight and means to evaluate more factors in matching children and foster parents.

PS: And that sounds very useful. Now, how do police departments use predictive analytics to track crime patterns and can they really --

RK: Going back to my first analogy with hunting, I mean how do - do you think police officers make decisions today when they go on a shift. A police department and an officer typically they have a lot of historical data on crimes so you can imagine it with many variables. Typically, they will have five years of history that would span many worksheets. About 200 different variables of weather conditions, events that are happening in the community and etcetera. So a police officer comes and they have to decide where are they going, which areas, which beats are they going to police.

We don't expect them - I don't think that they sit and analyze these Excel sheets and these data to make a decision where to send cars that would be too time-consuming. So naturally, experienced police officers use their gut feeling and their historic knowledge, observed crime patterns to allocate resources to particular beats. So where predictive analytics comes, so we're helping them to make better decisions then deciding on a gut feeling and on their limited historical knowledge. So the model takes all these historical variables, takes this gigantic Excel sheet with 200 columns and hundreds of thousands of rows of them and it analyzes it, detects the patterns.

And when an officer comes on a shift and they keying in a simple webform, the time on the shift and the weather conditions for the day and certain events that are happening in the communities, the software generates automatic predictions about the likelihood of crimes to occur in particular areas. So it predicts that within a certain beat, within a certain area there is a 50% probability of a crime occurring or 60%, or a 70% probability. And that helps the officer to actually decide in which beats they would send the patrol cars to prevent crimes. That again, the value for the officer is the larger amount of data that is being analyzed to detect the patterns, the refinement of the prediction compared to what we humanly can process based on our abilities.

PS: How about financial services firms. How do they use predictive analytics and could it possibility prevent another mortgage meltdown?

RK: Two aspects here. The financial industry have been the earlier adopters of predictive analytics compared to anyone else. So taking from the direct mailing campaigns, those offers that come in the mail for credit cards and different loans, the challenge that a financial institution faces is they acquire customers list. So let's say they would acquire a 500,000 people mailing list. So they don't want to mail to all these 500,000 people. They want to determine which are the people that are most likely to accept the offer.

And usually with direct mail, the response rate is about 2%. So it makes a dramatic cost difference whether you're mailing to 2% or to 500,000 people. So they use predictive analytics to identify the 10 to 5% of the most likely people from a list to accept an offer, and they mail it, and that reduces the cost of their direct mailing campaigns.

Another level at which use predictive analytics is from detection, very widely deployed area predictive analytics. They use historical patterns of transactions and they immediately analyze whether a particular transaction is likely to be fraudulent or not. Believe it or not, it happened to me when I bought my wife's engagement ring in Europe and came to the states. The transaction was declined because the bank didn't expect that I'm buying frequently jewelry. There's a classical case of they analyzing on a micro level my patterns and determining whether something is a transaction that I typically do or might be a potentially fraudulent transaction.

Whether it comes to the mortgage and finance and whether it will prevent the meltdown, I think that the mortgage crisis is probably one of the easiest areas to apply predictive analytics. And traditionally, it has been applied. And if we look historically at what happened is the mortgage crisis originated not because banks have not been applying predictive analytics to credit and mortgage applications. I mean that's why we all have a credit score. What has happened is that the banks have relaxed and pretty much ignored the credit score, which has allowed people who traditionally would have not received a mortgage to actually obtain a mortgage. So there was an inflation in the credit score and a blatant ignoring of the credit score.

If we look what's happening today, we're actually seeing the banks being conservative. And there was an article about it in the New York Times over the weekend. They're being conservative and actually raising internally the credit score so that they can make obtaining a mortgage much more difficult. So if my credit score comes let's say, at a 500 level, the banks would add additional 10% to hedge against weak credit scores. So yes, to answer your question directly, mortgage is probably the easiest area in which your predictive analytics can minimize risks to a great extent.

PS: Gotcha. Now, let's look ahead and what do you see for the future of predictive analytics?

RK: I think that it's going to have a much wider deployment than up until now and there are a number of contributing factors. Number one, the education of analytics and predictive analytics is changing. DePaul University just announced the first master's degree in predictive analytics and they recognize that this is an interdisciplinary work, which requires knowledge of BI, marketing, statistics, economics, and computer science so it's a very interesting development.

But what it also does is it recognizes that statistics and the way it taught in business schools nowadays is not teaching the students so much about the math which was kind of a deterrent for a lot of people to enter the field because it was very esoteric and very complex. But actually, teaching them about how to interpret the models. The math is done behind the scenes by the computers and all we need from this analysis is the interpretation of what it really tells us, the business value, the analytic value that the person should understand and apply in a real situation.

That has started very much with General Electric applying Six Sigma across all the areas from engineering to marketing and actually taking exactly this approach and saying we don't want you to know how to manually create the regression equation, we just want you to know how to interpret the results. And so that trend and allowing people to understand the business value and taking kind of the esoteric side of it away from the business user has created a tremendous appreciation of the insights that can be brought by the economic analysis and predictive modeling and etcetera.

The second contributing factor, if we look at it, is BI's a very established discipline. It has been around for 20 years so most of the companies that have competed on reporting and traditional BI, looking at historical data. That type of competitive advantage has equalized everybody the systems. So now the shift of management is how can we look a little bit ahead and act proactively against the acting reactively and that ensures the competitive advantage. So if we take an industry, for example, like retail. In retail, the sales cycles have shrunk from nine weeks to six weeks. So if you don't sale the merchandise in six weeks, it essentially goes back to discounters like TJ Maxx, Marshalls and etcetera.

So anybody who can predict trends better and liquidate the merchandise naturally has a competitive advantage so that's driving the push to creating in organizations the culture of competing on analytics, deploying predictive analytics to better gauge the future, and to reap these benefits like traditionally has been done in the stock market. A broker that had a better insight had a better return on the investment.

ebizQ’s expert blog team covers a broad range of BPM, business integration, business analytics/monitoring, collaboration, content and related issues.

Peter Schooff

Peter Schooff is Contributing Editor at ebizQ, and manager of the ebizQ Forum. Contact him at pschooff@techtarget.com

Recently Commented On

Monthly Archives