Predictive modeling is on the rise for property & casualtyinsurers in virtually every line of business over the last year,according to Towers Watson's fifth annual Predictive Modeling survey, butmost insurers do not have a comprehensive, company-wide approachfor using predictive modeling for all core functions.

Results show that there is a lack of data-driven analyticsuniformity in most enterprises while overall usage fluctuatessignificantly depending on the company size or by the line ofbusiness.

Carriers have a willingness “to embrace predictive modelingprograms, but in many instances, the actual investment in, orexecution to establish these frameworks, has been incomplete ortargeted to specific business lines or operational needs,” saidBrian Stoll, director, P&C practice, Towers Watson.

According to Stoll, there are several possible reasons why. Thefinancial crisis could be causing insurers to put investments onhold and instead are focusing on expenses, or Stoll suggests thatthere could be a narrow vision of predictive modeling'sapplications and potential.

“Perhaps data, people or cultural challenges are a factor, orsome are only applying data-driven analytics when an area isunderperforming. Whatever the reasons, a compelling case can bemade that well-executed predictive modeling provides better pricingguidance to underwriters,” Stoll said.

The fifth annual Predictive Modeling survey explored the waysinsurers are applying predictive models in the industry, includingresponses from 59 U.S. P&C insurance executives. Click throughthe following slides for the survey's key findings.

A Solid Base forImplementation

P&C insurers increasingly have the basic tools andcapabilities they need to pursue data-driven analytics integrationthroughout the organization, suggesting consistent enthusiasm forpredictive modeling.

The survey reveals that more than 75% of personal lines andsmall to mid-market commercial lines respondents view predictivemodeling as very important to their business. All personal linesrespondents indicated that predictive modeling has at least somedegree of importance for sophisticated underwriting and riskselection techniques for rating and pricing, and 83% of personallines respondents cited predictive modeling as “essential.”

Similarly, 79% of commercial lines respondents agreed thatpredictive modeling functions were “essential” or “very important,”as did 56% of large account and specialty lines carriers.

Specialty lines carriers also expressed interest in predictivemodeling applications, with 45% indicating they have future plansfor implementation.

The value of predictive modeling is reflected by increasedimplementation. For all lines of business, the use of predictivemodeling has increased. In particular, usage in personal automobilegrew five percentage points from 2012, bringing total implantationto 80% among respondents, while homeowners increased to 62%. Amongthe largest gains over 2012 were workers' compensation (9percentage points) and general liability, commercial multiple periland business owners' policy (7 percentage points).

Favorable BottomLines

The survey reports that nearly all carriers experiencedfavorable bottom-line results because of predictive modeling,creating positive impacts on rate accuracy, profitability and lossratio improvement.

Results showed that benefits were less significant, however, torespondents' top-line results, such as modeling's impact on theexpansion of their underwriting appetite and on market share.

“Respondents are really tailoring their programs to focus onspecific market realities,” said Towers Watson senior consultantKlayton Southwood. “Personal lines carriers operate in a highlycompetitive, mature market, so it's not surprising a highpercentage have adopted many aspects of modeling. On the otherhand, commercial lines carriers face less intense pricing pressurein some segments, in part due to heterogeneous risks and theheightened reliance on individual risk underwriting expertise,particularly in large risk/specialty lines.”

In terms of renewal retention, there was a greater disparitybetween personal and commercial lines carriers. 25% of personallines carriers reported a positive impact, while 52% of commerciallines respondents reported a positive impact.

The survey suggests that the disparity can be attributed to thelack of product differentiation and keen price competition inpersonal lines, as well as customers' willingness to switchcarriers. Commercial lines carriers' positive retention may be dueto the potential for predictive modeling for improvement ofunderwriting and rating accuracy, allowing carriers to attractcustomers with more targeted premium rates. Commercialpolicyholders, in general, have broader insurance relationshipswith carriers, and heavier dependence on customized/differentiatedservices, reducing price sensitivity and willingness to switchcarriers.

Data and Communications

Survey findings reveal that a commitment to predictive modelingis more than just building and applying technical models. Improvedcommunications concerning the measurable value that strategicapplication that predictive modeling offers to all key stakeholdersis imperative for effective implementation and integration.

While insurers say they are fostering greater agentunderstanding and approval of predictive modeling, the surveyfindings reveal a communication gap between carriers andagents.

Less than half of survey respondents provide relevant insight toagents related to modeling efforts, and only a small fraction ofrespondents involve their agents in the model-building process,explain their pricing models to agents or communicate model changesto agents in advance.

However, the majority of carriers are using internal andexternal data to improve the sophistication and power of theirmodels. Commercial lines carriers, in particular, tend to focus onmore external risk-specific variables and socio-demographic data.Personal lines carriers, on the other hand, stress variableinteractions to strengthen their models and are more likely toapply modeling in the form of rating plan adjustments by creatingor revising rating/tier variables and relativities.

Survey results from the past two years show the increasedwillingness to embrace predictive modeling programs, but actualinvestment or execution is often tentative, or falls short. TowersWatson suggests that a more comprehensive, strategized andaggressive approach to predictive modeling implementation isnecessary for effective implementation.