The Business Analytics Knowledge Exchange recently had the opportunity to talk with Greg Hayworth, a scientist with health insurer Humana, about his use of advanced analytics to do everything from mine text to reduce calls seeking assistance to predicting which claims are duplicates.
ANNA BROWN: Tell us about your role at Humana.
GREG HAYWORTH: I’m a scientist, which sounds a little strange to some people. Why does an insurance company need a scientist? But I do advanced analytics and predictive modeling-type projects in our provider network operations area. One thing I do is look at claims, subrogation, where you have a car accident or worker’s comp situation, and someone else is responsible for paying or detecting fraud and abuse.
What do you see as the major trends in analytics?
HAYWORTH: Storage is cheap, and with parallel processing capability, getting to the data is less of problem. We’re in a sweet spot now where we can ask all kinds of questions of our petabyte or more of data. Before, the query took three hours to run and maybe I would get something, maybe not. Now the query runs in 20 seconds so I can ask the question 16 different ways and I still have time to go to lunch.
You talk about parallel processing, what are your thoughts on new technologies that deal with big data?
HAYWORTH: It’s not my area of expertise, but I know with parallel processing what took hours can be done in seconds. Once you can easily tap into your structured data, you’ll see a lot more people tapping into the unstructured data. I know people have been doing text mining stuff for many years, but now it is taking off.
Can you share an advanced analytics success?
HAYWORTH: Our biggest success story has been in the area of identifying duplicate claims. It’s a common problem in insurance. Physicians or hospitals submit their claim; it was delayed for one reason or another. They perceived that it was lost, so they send another one in. We just want to pay one, at the contracted rate, so pulling out and identifying those duplicate claims is something that every payer’s got to do. We were able to build a predictive model identifying the characteristics that identify duplicate claims vs. claims that look similar – such as for a person with a chronic illness who makes multiple visits for the same problem. We used to have a department that spent all day looking at pairs of similar claims. We increased the productivity of that department 450 percent.
Is there any advice you would give to companies working on projects like this?
HAYWORTH: It’s important to seek out a partner in the business unit that will be affected by what you discover. I now have an implementation readiness checklist. We ask, “What are you going to do with this?,” “Do you have people ready to change your software?” and “How will you reallocate resources?” If those questions don’t have solid answers, maybe it’s not time to do that project.
How have you used text analytics at Humana?
HAYWORTH: The structured data from our call center tells us that 80 percent of calls are about claim status or verification of benefits. We have online tools to answer those questions, so why do we keep getting calls? That’s where text analytics comes in. We looked at notes that were coded as “claim status calls,” they’re not really claim status calls; they’re things that are related to claims. The first thing that is on the drop-down list that starts with the word “claim” is claim status. But by understanding on a more granular level why providers are calling us, we can be innovative about having better answers online and proactively pushing information out via other sorts of electronic channels. We ended up with a 10 percent reduction in calls per claim.
How do you hire analytic talent to help with these projects?
HAYWORTH: There are a lot of people who know how to apply formulas and know how to use tools, but can’t influence their way out of a wet paper bag. The X factor is the ability to influence. You need to be able to convince an operations manager that what you discover might mean they will only need four people to do something, rather than 18.