Predictive Modeling Success Requires an Enterprisewide Effort
By BRIAN STOLL, senior consultant at Towers Watson
The availability of electronic data has exploded in recent years, and that information presents analytical opportunities never before contemplated. As a result, predictive models have become a key component of success for many diverse ventures and businesses across the globe, and property/casualty insurance is no exception. Having seen the benefits realized by personal-lines carriers, commercial-lines carriers, particularly national carriers, are now investing in predictive modeling. The early adopters in commercial lines are posting better loss ratios and experiencing more favorable pricing changes than their competitors.
Successful implementation of predictive-modeling applications requires significant involvement from across the entire enterprise. To get the most out of predictive modeling, companies need a clear understanding of its benefits--from the bottom-line opportunities of more accurate pricing, loss-ratio improvement and profitability to top-line priorities such as expansion of underwriting appetite, better renewal retention and increased market share.
Taken together with recognition of the pace at which marketplace competition is adopting predictive modeling, the initial decision to invest is not overly difficult. Achieving the desired returns on those investments, however, can be more challenging.
NATIONALS RAISING THE BAR
Clearly, national carriers with a large market share have a distinct advantage in utilizing predictive modeling, given their access to significant volumes of internal data. They have already set the standard high for personal lines. For instance, within personal auto, they have created tens, or even hundreds, of thousands of price points. Having seen the benefits in personal auto, most national carriers have already extended these techniques to homeowners insurance.
The success of these companies has also clearly raised the bar for standard and specialty commercial lines. National carriers have already committed to investing in the modeling staff, technology and implementation support necessary to become market leaders. Regional and smaller commercial-lines carriers have responded by analyzing competitor filings and rating/tier plans in an effort to remain price competitive. They can then pursue their traditional strategies and attempt to differentiate themselves on product and service.
The national carriers are aggressively pursuing core profitable business of smaller competitors (this has been playing out for years in personal lines). The effectiveness with which the rest of the marketplace can respond by combining internal pricing capabilities, consulting expertise to identify and address gaps, and superior modeling software technology will separate the winners from the losers within their peer groups.
Once the foundation has been established, the optimal application of predictive modeling requires a long-term commitment from senior management. Company leadership buy-in is vital to the success of predictive modeling, not only by providing funding and giving it credibility throughout the enterprise, but also by giving visible advocacy and aligning the corporate culture behind the initiative to ensure execution. Senior management can provide the necessary support to sustain predictive modeling as it rolls out across the enterprise.
Of course, there are many challenges to successfully building and implementing predictive models--including data and technology, workflow and process, and people and cultural issues, such as change resistance, degree of underwriting autonomy and buy-in, to name a few. The ongoing support of senior management is also critical to working through obstacles and achieving increasingly sophisticated milestones in successive iterations.
Another key step in the predictive modeling process is establishing clear and achievable short-term goals and continually re-evaluating those goals. This involves determining whether the short-term focus is on the bottom or top line, and aligning people behind that objective. This also requires clarity on whether to emphasize improvement in technical price accuracy, perform competitive market analysis or engage in price optimization.
While there are many aspects of predictive modeling to address, picking a few at any given time (e.g., data quality, redefining territories, introducing GLM for a new coverage, implementation or monitoring results) and making sure to get them right works better than trying to address too many issues too quickly. This means effectively balancing the desire to expand to different coverages or geographies with enhancing the quality of the technical models and addressing implementation or data/technology challenges.
Once the issues have been identified and senior management is fully on board, building and maintaining the models moves to the forefront. Internal and external data must be painstakingly scrubbed, transformed where necessary, validated and ultimately molded into a viable predictive-modeling database. The modeling approach (e.g., frequency/severity, pure premium or loss ratio) and methodology must be selected. This is followed by univariate and multivariate analyses; identification of potentially new and innovative rating variables and variable interactions; and, ultimately, validation, implementation and monitoring of results.
Predictive modeling software can greatly facilitate the speed and effectiveness of this process, providing discipline and structure for data preparation, enabling quick and efficient iterations in far less time than traditional direct coding, as well as delivering many and varied visualizations of results to support both analysis of the models and communication of results. Actuarial/statistical software has been developed specifically to assist with this process, and the right software can ensure that the highest standards are used throughout the business. The associated benefits of reliability, repeatability and auditability can help an organization achieve its risk management objectives in conjunction with pricing goals.
REACHING FULL POTENTIAL
Finally, once the models have been built and validated, the challenges of effective implementation must be faced to ensure that the full model potential is reached. Only when a quality predictive model is effectively implemented can it achieve the desired results. This means communicating the results and implementation approach to all key stakeholders, training the underwriters and agents effectively, and ensuring execution of the new work processes. Aligning the organization behind the initiative and communicating expectations ensure that the model will perform to its potential and deliver the targeted results.
Companies that follow this path will continue to enhance their models and performance results, and consistently match or outperform the competition. There is a clear correlation between predictive modeling sophistication and superior growth and profitability. Given that the market is always moving forward, relaxing or failing to sustain momentum means losing competitive position and risking the erosion of results.
While predictive modeling alone cannot guarantee that a company will succeed in the marketplace, failure to effectively compete in predictive modeling can virtually ensure that a company will not succeed. Given the momentum behind the use of data and analytics today, it is clear that a commitment to predictive modeling is key to outperforming the competition.
May 1, 2010
Copyright 2010© LRP Publications