Get smart with predictive analytics

More and more, insurers are looking to incorporate predictive analytics to achieve business results.

"Over the last two years the pace of change has increased," says Steve Kauderer, partner in Bain & Co.'s Financial Services practice in its New York office. "The identification of both internal and external data to use in analytics has increased, and the ability to link and bring disparate groups of information together has grown."

ACUITY's Ed Felchner maintains that having an effective analytics strategy has evolved from competitive advantage just a few years ago to market necessity.

"You have to have at least some capabilities in predictive analytics today or you are going to fall behind," says Felchner, ACUITY's vice president of personal lines and marketing. "We're fortunate to not be playing catch-up and to have established a foundation for analytics we can build on to increase competitive advantage."

Analytic Goals

Predictive analytics differs from data mining and reporting because of its forward-looking objective of predicting outcomes. P&C insurers are using analytics to target many strategic objectives, but two areas of focus stand out from the rest.

"We see insurers predominantly making investment in underwriting and claims," says Arunashish Majumdar, chief architect and leader of the insurance practice in North America for TCS consultancy. 

William Dibble led Infinity Property and Casualty Corp.'s national claims operation until his retirement in 2013, spearheading the company's efforts to implement a real-time predictive analytics solution to detect potential fraud and improve claims processes and outcomes. He now serves on the advisory board of n2uitive, a provider of recorded statement products and services, and as an independent consultant for insurers, advising on the use of analytics in claims operations.

"Companies are getting a lot more sophisticated in gathering and analyzing data and using it to predict claims outcomes and detect fraud," Dibble says. At Infinity, analytics helped the company improve its special investigative unit success rate from 60% to 90%; however, Dibble stresses that carriers need to continue to refine their analytics to keep ahead in the cat-and-mouse game that is the insurance fraud fight.

Learn new strategies for properly leveraging data at the 19th annual America's Claims Event, June 17-19 in Austin. For more information, click here. Use Code PC360AE & Save $100.

"As carriers get more sophisticated with detecting organized crime and connecting the dots to identify groups of people who are going into organized fraud, they see that those groups refine their approaches to avoid detection. Insurers need to have adaptive analytic models and continue to use what they learn to stay a step ahead," Dibble says.

Risk selection and pricing remains a key focus for insurers because of its impact in terms of lift. Based on its work with P&C carriers, Bain estimates that companies can improve their combined ratio by 3 to 5 percentage points and their premium growth by 5%-15% above baseline expectations by using predictive analytics to actively target high-profit potential customers with attractive pricing and drive the worst risks to competitors.

"We consistently see the best carriers accomplish those goals," Kauderer says.

"For us, targeting greater precision, being more competitive on new business, and getting the right price on existing business was well worth the investment in analytics," says Pat Tures, vice president of actuarial at ACUITY.

"Our goal is to have everyone pay what they owe—no more no less—and to not have some customers subsidize others. The more accurately you can price, the more fair you will be, and the more competitive you will be overall," Felchner says.

ACUITY pursues a "precision pricing" strategy in personal lines that incorporates different analytic tools and a continually evolving array data and information sources. A key factor in the company's strategy is the use of AudaExplore's Insight products. In 2013, ACUITY began using Property Location Insight (PLI) from AudaExplore and, in 2014, it began using Auto Location Insight (ALI).

Both PLI and ALI are location-based predictive models that look at a wide range of variables, including weather, geography, topography, road conditions, traffic volume and infrastructure. Each model creates an account-specific score that accurately reflects the probability of loss from fire, theft, liability, weather and other events.

"We can insure two customers living next to each other who have completely different scores because of different exposures. It allows us to rate by peril, and within each peril have very narrow bands of pricing," Felchner says.

For instance, a home adjacent to a commercial area may face a higher risk of crime than one a few blocks away. Or, vehicles garaged at the end of a cul-de-sac could be less susceptible to collision loss than those parked at the end of the road near the intersection. The result of this analysis is individual risk-based pricing.

"Years ago we had wide rating bands and perhaps a couple of hundred pricing points, and the whole universe of risk had to fit in there. Now we have literally billions of possibilities, with each account developing a unique price that is right for it," says Tures.

Although Felchner says there are several factors behind the company's growth in its personal lines division, individual risk-based pricing has been essential. The company achieved a 32% growth in new business personal lines premium in 2013, is on track to exceed that in 2014 while running a profitable combined ratio.

Targeting Business Outcomes

In evaluating predictive analytics, companies may be tempted to start with the data. That's the wrong approach, Majumdar says.

"Start with the business problem," he explains. "What are you trying to solve? What is the business outcome you want to achieve?"

At Falls Lake National (formerly named Stonewood National Insurance Co.), the objective was improving underwriting results. In the years following the onset of the 2008 recession, the insurer began to experience significant profitability problems with its book of Workers' Compensation business.

"We were a beneficiary of the good times before the recession of 2008, but suffered through the bad times after it," says Steve Hartman, CEO and president at Falls Lake National.

Those bad times culminated in significant underwriting losses from 2010-2012, leading the insurer to undertake a reanalysis of its entire book of business. That analysis resulted in updated rate filings, updated deviation protocols on scheduled credits and debits, and refocused underwriting.

The book re-underwrite produced a 35-point improvement in the company's accident year loss ratio between 2012 and 2013. The insurer wanted to incorporate what it learned from the review into an underwriting and pricing model that would help avoid mistakes of the past and capitalize on future opportunity.

"That's where predictive analytics come in," Hartman says. "We wanted to start systematizing a underwriting process that has been solely a judgmental process. We wanted to give underwriters analyses and information to augment their judgment, relying on best-in-class tools to differentiate among different risks and opportunities, and better match exposure to the price charged on risks they wrote."

With 27,000 policies written and 14,000 claims incurred since the insurer's inception in 2004, Falls Lake lacked the amount of internal data needed to develop its own credible predictive model. The company chose the InsureRight platform from Valen Analytics for risk scoring and underwriting decision support in part because of the platform's incorporation of a large amount of industry data.

"Valen matched our policy and claims data to the millions of data points they have and returned a tool that is customized to fit our business," Hartman says.

Insufficient internal data to develop a credible predictive model is a common problem, particularly among small- and mid-sized carriers. "Although personal and small commercial carriers tend to have a good volume of data for traditional business intelligence needs, they often need to bring in outside sources for analytics," Kauderer says.  

Falls Lake began implementing the Valen platform in September 2014 and plans to complete the rollout by early 2015. The objective is for the tool to supplement, not replace, underwriter judgment.

"Some companies look for an analytics tool to be a black box that provides a yes/no response. We didn't want to do that," says Hartman.

"Our underwriters average between 15-20 years of experience, which is irreplaceable," he adds. "However, they are limited by time and resource factors. We wanted to provide a framework for discretionary pricing and a tool that provides more data, and more analyses, than they have ever seen before, then let them apply their experience."

A question that often arises with the use of third-party data sets and scores is that, because the same scores are available to multiple carriers, where does the competitive differentiation exist? The answer lies in how insurers layer their own analytic approach on top of third party data to create a model unique to their needs.

"Although it's true that some basic data variables are in common use by almost all insurance carriers, there is still immense variation in the number, type, and interpretation of variables used when pricing insurance policies," says Tom Eggenberger, managing director of the driver behavior group at AudaExplore.

"Carriers still need to make many complex decisions about how to interpret each data variable and the relationships between the variables. Some carriers have more sophisticated analytical capabilities, which allow them to identify more subtle trends in the data and to set more accurate prices," Eggenberger says.

"By itself, the [AudaExplore] score doesn't mean anything," says Felchner. "Let's say you have a wind score of 30. What is that worth? You need to take your data and look at how those scores respond to your outcomes—that's what we had to figure out."

Figuring that out involved ACUITY's actuaries taking years of underwriting and claim data from its data warehouse and running it against the PLI and ALI scores to determine how factors related to business outcomes, using Towers Watson's Emblem software as the company's predictive modeling platform. Any new factor that ACUITY considers for rating undergoes a similar analysis.

Advance of Analytics

Expect insurers to continue to apply what they've learned to analytics and find ways for it to advance business outcomes. "Even with all the progress they've made, insurers have only scratched the surface of what analytics can do," Dibble says. 

"More and more, the better business will go to companies that are priced right," says ACUITY's Tures. "Those who use predictive analytics to guide pricing precision and decision making will win, and those who don't will fall further and further behind."