Making Sure the Price is Right

Challenges are growing as many commercial lines insurers face economic pressures and increased regulatory requirements. In a business world where many factors are out of insurers' control, in order to remain competitive, they are well advised to understand the types of risks most appropriate for a specific product set, and price those risks accordingly. In some cases, this results in a carrier turning down business that is not profitable, or even exiting an entire market that is not performing as expected.

So how do they know what they know? How do they identify, segment and price risks in such a way as to consistently improve loss ratios?

The short answer is applying predictive models to the underwriting process, say experts, and insurers of all sizes are starting to realize the benefits of their use.

"Insurers are saying 'we are in a defensive posture with this, and we have seen adverse effects based on doing what we normally do,'" notes Mark Gorman, principle with Minneapolis-based Mark Gorman & Associates. "The market leaders already have invested in this technology as a core competency. But this is no longer a debate about whether or not to use predictive analytics. Now carriers of all sizes, representing all lines of business, are confirming that it's a matter of when, not if."

One line of business where predictive analytic models have already started to offer big paybacks is workers' compensation.

RELYING ON EFFECTIVE PRICING

Bob Dove, CEO of Honolulu-based Hawaii Employers' Mutual Insurance Co., (HEMIC) relates that in 1996, HEMIC replaced the existing workers' comp assigned risk pool to provide workers' comp coverage for Hawaii employers, including those who unsuccessfully sought workers' comp insurance in the voluntary market. HEMIC currently writes 6,500 employers, making it the largest workers' comp insurer in Hawaii. As the insurer of last resort, however, HEMIC does not have the luxury of turning down business that is not profitable - instead, it must rely on effective pricing models.

"We anticipated moving into a soft market, and we knew margins would be tested," says Dove. "We wanted to do a better job right-pricing the product."

When the company began building out its predictive models a year and a half ago, it did so to replace tools that measured only two metrics at a time.

Dove says the evaluation process became a matter of looking at account underwriting, which examines the characteristics, specs and experience of that account, as well as traditional underwriting, which takes into account broader classifications, such as NCCI data (National Council on Compensation Insurance Inc., Boca Raton, Fla., the nation's largest provider of workers' comp data).

"By using predictive models, commercial lines underwriters can review an even larger number of more specific underwriting variables to identify those policies that are most likely to produce losses over a period of time," notes John Lucker, principal in the insurance practice at Deloitte Consulting LLP, Hartford, Conn.

Dove agrees. "We are guaranteed market, so the question is not 'do we want to quote this?' It's 'how much do we want to charge?' With predictive modeling, we are able to string data elements together to find enhanced predictability," he says. "We are still doing classic underwriting, but nine different data elements now go into the scoring system. This helps us price the risks we are taking on."

This philosophy appears to be working: In July, HEMIC declared a $3.25 million dividend payable to qualifying policyholders.

A MOVING TARGET

In the highly state-controlled, heavily price-regulated environment of workers' comp insurance, the likelihood is great that premiums charged on payroll volume and job classifications will fluctuate throughout the policy term and, as a result, will influence the amount of premium due.

Already charged with ensuring rate integrity, and with selecting the most appropriate risks (that best match the profile of its book of business), the workers' comp underwriter also faces state regulations that govern which risk characteristics may be considered and/or allowed, and which may promote unfair discrimination. And, depending upon the state in which the business is underwritten, the regulations affecting rates are a moving target.

The state of California is a case in point: In 1994 the state passed The Competitive Rate Law in California, and for the first time in more than 80 years, created open competition in the California workers' compensation insurance market. Since then, the "open rating environment" has been an explosive one, with the insurance bureau making multiple recommendations for two-digit rate increases that offset increased medical and other costs.

"There are just so many things we can control, and the competitive nature of the business, along with the political environment in California, is extremely volatile," notes Nora Greathouse, SVP of Underwriting, West, at Majestic Insurance Co., San Francisco.

Greathouse points to the California WCAB (Workers' Compensation Appeals Board) as an example of an industry in flux.

"The commissioner could not approve or disapprove in time for insurance companies to make timely decisions about July business," she says. "Just recently, the commissioner disapproved a rate increase, and issued a mandate for us to prove, via data, why an increase was justified."

As a result, Majestic, a $150-million company that provides coverage for 2,500 employers, is more focused than ever on its data set and, already using the theories behind predictive analytics, says Greathouse, even though it has not yet built a predictive model.

"If the commission does not believe the data," she asks "where does that leave the insurance companies? We are a boutique operation, yet we are held to the same requirements."

DISCIPLINED, REPLICABLE

Greathouse asserts that the knowledge base each insurer develops in-house must support decisions to either stay pat, or file for a rate increase or reduction. "You should be able to proceed as long as you can prove that your data is accurate," she says.

The ultimate goal is to apply a disciplined, replicable approach to the underwriting experience, notes Gorman.

"Once insurers have made the decision to build out predictive models, they have to balance what's theoretically predictive and what's practical from a business standpoint," he says.

No one would argue that working with the best possible set of underwriting data is key to creating the most effective predictive model. But the big-picture view of what it takes - making the decision, deciding on the most appropriate data to include, extracting the data, building the model and finally, implementing it - can be overwhelming at best.

"It really requires a leap of faith," says Dove. The company is currently using Denver-based Valen Technololgies' UnderRight solution, and Valen integrates third-party data to HEMIC's own.

"Companies often struggle with how to effectively use the models that have been developed, and benefit from being able to translate model predictions into practical business rules and decisions - integrated into their workflow - that align with the company's strategic business goals," notes Valen's president, CEO and founder Dax Craig.

HEMIC's strategic business goals continue to take priority throughout the insurer's predictive modeling initiative. And although HEMIC outsources its actuarial functions, the insurer involved its seven underwriters in the process from the beginning.

"There was no real pushback," he says.

According to Gail McGiffin, global head of underwriting at Bermuda-based Accenture, when pushback does occur, it's usually due to cultural, historical bias among underwriters.

"This will change over time as we get a newer generation of underwriters," she says. "To be fair, the technology advancements in the world of commercial insurance have been slow and point-based. So there is still skepticism, not just about the model but also the technology enablement of the model."

CONTROLLING YOUR DESTINY

While Majestic Insurance evaluates its potential predictive modeling initiatives, the company is taking other proactive measures to improve its underwriting and rating. Working with iPartners LLC, Alpharetta, Ga., Majestic is able to identify trends such as sudden changes in the frequency, duration and outcome of medical claims in a specific industry and/or class.

"We wanted the ability to look at overall exposure in a particular category, and its affect on current loss ratios," Greathouse says. "This tool enables us to look at a 60,000-foot level, use that view to make informed decisions and create reports back to the regulatory bodies."

Like Greathouse, Sharon Lane, VP, underwriting, commercial, at Penn National Insurance, Harrisburg Pa., faces a host of state-based regulatory ups and downs.

"There is not a lot of flexibility to control your own destiny," Lane says.

Playing off the success of predictive models used by underwriters in its personal auto line, the $500 million mutual company plans to have its workers' comp model built, and scorecard in place, by the end of the year.

This means combining data sets from the Philadelphia-based Pennsylvania Compensation Rating Bureau (PCRB), NCCI, third-party credit, demographic and supplemental data.

"Because workers' comp is so highly regulated, the model will help us better understand what non-traditional attributes or characteristics can help guide underwriters in their process," Lane says.

HEMIC's Dove predicts similar results. "You'd like to be able to predict the experience of each and every risk, but the smaller a risk or account gets, the less predictable it becomes," he explains.

Dove offers as an example the assumption that a good, small risk should only have a claim every five to seven years. If an account had a claim last year, it doesn't necessarily make it a bad risk, because they may not have had a claim previous to that, he points out.

"You have to be able to lump them into a larger group of like accounts, then break them out in a more sophisticated way," Dove says. "Predictive models allow you to do that."

PREDICTIVE MODELING FINDS ITS NICHE

The hallmark of predictive modeling - its ability to lend credible intelligence to decision-making-is being appreciated in more areas than just underwriting.

"As insurers look to decrease costs and improve operational efficiency in today's economic climate, they seeing the biggest benefit for predictive modeling is in the claims process," notes Stuart Rose, global insurance marketing manager, at SAS Institute Inc., Cary, N.C. "Not only are insurers using predictive modeling for fraud detection, they are also creating models to assess the propensity for claims litigation and improve salvage and subrogation recovery income."

Rose asserts that insurers are just scratching the surface in terms of where predictive analytics can be used within their organizations.

Taking into account the current economic landscape, Rose says that one potential area where analytics is under-used is within the distribution channel.

"Predicting the most efficient distribution channel in which insurers sell their products to customers, and using the huge amount of data available to create distribution efficiency models can be integrated with a social network analysis techniques to bolster an insurer's distribution strategy," he says.

Rose suggests that insurers consider creating a center of excellence that is responsible for analytics throughout the organization.

"This center of excellence would ensure accurate data as well as consistent standards for data integration, analytics and reporting," he says. "Additionally, this department would be able to ensure that the limited analytical resources can concentrate on projects that crate the biggest benefit to the organization."

(c) 2009 Insurance Networking News and SourceMedia, Inc. All Rights Reserved.

For reprint and licensing requests for this article, click here.
Analytics Policy adminstration Data and information management
MORE FROM DIGITAL INSURANCE