While it's impossible to know what you don't know about the massive stores of data insurers accumulate, it's also possible to not understand what you know. That's where predictive analytics comes in. Vince Franz, VP and Chief Actuary, Celina Insurance Group, talked with Insurance Networking News about actuarial practices, the role of predictive analytics and how to justify the cost of such a project.
INN: How did you first get interested in using predictive analytics, and how long ago was that?
VF: When I came to Celina seven years ago, there was nothing in the way of predictive analytics, which is probably the case for most companies our size. So we did a couple in-house projects that were somewhat crude and unwieldy and thought there had to be a better way. So in 2009, I went to the Casualty Actuarial Society rate-making seminar. That's where I learned how prevalent predictive analytics was becoming. It's gone from a nice-to-have to a must-have.
We wanted something we could get going relatively quickly, that was user friendly. I liked the twist that EagleEye Analytics has; it uses a machine learning approach, which takes the data you have, and finds the combinations of risk attributes that show which risk segments are most and least profitable. That was very attractive to us because it was using our book of business and I thought it would give us a different approach than that of some of the big companies that we can't go toe-to-toe with. If we use a different approach, maybe we could find our competitive niche.
INN: How are you using it? What does it do for you?
VF: We have a thousand different rating variables in different combinations. It's overwhelming. And this tool helps you find what combinations of characteristics you are overpricing, and underpricing. That's what the EagleEye tool does. It finds all those combinations and puts them into risk segments ranked from the most profitable, which you are probably overpriced on, to the least profitable risks that you may not be recognizing in your pricing, and are costing you lots of money. It helps you refine your pricing structure so you are matching the premium you're charging more appropriately to the exposure.
INN: Is it paying for itself? How can you tell?
VF: We implemented the changes into the rating structure just last year, so it's too early to have quantitative analysis on those revisions. One thing we did do was make it a priority to complete an expense-reduction project. That way we could demonstrate to our senior management that this tool is expected to pay for itself in X-number of months.
We refined the criteria for when we order reports for our underwriters. They use reports to evaluate the quality of the business coming in, and we found that a lot of the time the reports didn't add value when the quality of the risk was so good. So then we tried to focus on where we really need these reports, whether we should write the risk and how much we should charge for it. And that was a substantial savings per month, so that helped us pay for the product rather quickly.
INN: What kind of surprises did you find in the data?
VF: It takes a while to put together the data set, massage it and clean it up. And, it's not always intuitive. In our personal auto analysis, we found that cars made before 2000, when combined with certain other risk attributes, were just really profitable for us. You might think they are not well maintained and are more susceptible to an accident. But we reasoned that families may have a newer car that gets driven more, while the older one sits there, so there's more exposure to the newer one. We were able to recognize that in our pricing structure, and hopefully hold on to those types of risks.
INN: Knowing what you do now, what would you have done differently?
VF: I would start as soon as possible. We probably lost business from our most profitable accounts to companies that were further down the predictive analytics path and refining their rating structure. The other thing, and we did try to do this, was to involve other areas of the company.
INN: What did you ask? What did the other areas tell you?
VF: We asked where they see a need in light of corporate and business objectives, and to categorize those needs into something you 'need now,' 'need soon' or 'need later,' just for an idea about priorities and urgency. We wanted to say, 'here's the business need, now which vendor helps us accomplish those objectives the best?' Expense reduction was a popular item. And everyone put refined pricing as a 'need now' because other companies are using predictive analytics and putting us at a disadvantage. We came out of those interviews with a nice list and also a lot of consensus. And we did that before we talked to vendors. Our marketing and underwriting folks had their needs, so that helped direct us. The hardest part was just getting it started.
