Secrets to successful commercial segmentation: commercial insurers are beginning to develop customized predictive risk models, operationalize the models and harvest the business benefits.

When James Watson and Francis Crick discovered that differing sequences of amino acids create the DNA molecule and hence the basis that all life is built upon, science had effectively unraveled what makes each organism unique. Similarly, since insurance was first sold, underwriters have searched for the perfect combination of information that would help them segment the good from the bad risk and then help price such risks commensurately. Predictive models based on consumer credit data and other relevant data have been proven to provide significant risk segmentation for personal lines carriers. For small to midsized commercial insurance, however, the development of predictive models has been more challenging. Disparate product and data standards, heterogeneous policyholders, significantly varying premium sizes, multiple technology platforms, slow computing power and business inertia have made the progression of predictive modeling for commercial insurance less rapid--until now.

Over the past seven years, innovative insurers and professional service firms have worked hard to develop predictive models for commercial lines insurance. This progress has provided early adopters with the ability to select, manage and price risks better for a variety of products including business owners policies, commercial package, commercial automobile, general liability, commercial property, workers' compensation, umbrella, errors and omissions, directors and officers, employment practices liability and medical malpractice.

By coupling advances in high speed computing, actuarial and statistical modeling methods and new industry intellectual capital, leading companies have made significant advances in better risk segmentation and portfolio management by using predictive models. They knew it was essential to uncover not only the best combination of risk characteristics but also the ideal weighting of each risk characteristic relative to the others. Furthermore, they thought that these insights needed to be derived with a well defined statistical and actuarial rigor so that the results were usable, repeatable and executable. To do so would provide underwriters with a powerful new underwriting tool to more accurately price each and every risk in varying market conditions. In fact, they knew that segmentation was the name of the game and they wanted to be an early entrant to gain a competitive advantage.

This ability is particularly important now because the consensus is that the rates of the hard market of the early 2000s are gradually but decisively moving downward. The market is softening. Companies want to grow their market share, their premium base and their product diversification. Some say that a perfect storm is brewing for the return of the irrational soft market of the 1990s.

Leading companies know that if you can price more accurately than your competition (not necessarily price more aggressively), you will always be better off, despite market cycles.

Models on the Move

Previously in personal lines, with the use of credit-based predictive models, early adopters outperformed the competition. But over time, the majority of personal lines companies took advantage of the models. Numerous mistakes were made, including creating the concept as an industrywide black box, but those who implemented the models most effectively were able to emerge from the pack. Today such credit-based predictive models are table stakes, and the generic industry models that are widely used provide little competitive advantage except for two potential key differentiators--effective business implementation and the company's commitment to continue to innovate. The early adopters of the technology were able to remix their portfolio of risks and improve the quality of the book of business. Once optimized, the benefits of this action are recurring and compounding--the improvement of a book of business continues to produce financial benefits as the good risk outweighs the bad risk.

The race has begun in commercial lines with approximately 30% of small commercial premium being scored by predictive models today. The competitive landscape is becoming broad and varied with a mixture of national companies, super-regionals and regionals as the early adopters. The smaller companies have been able to move faster, implement the models more effectively and bring their underwriters and agency forces on board more quickly with these new underwriting tools. Here, being small and nimble has been an advantage.

There are clearly the market leaders who not only have been able to develop customized models (leverage their data) but who also have excelled at implementation both on their technology platforms and, more importantly, within their underwriting selection, pricing and work flow processes. In other words, these market leaders have learned how to operationalize the models and harvest the business benefits.

Leading companies are using models to help achieve a significant reduction in their loss and expense ratios through right pricing and higher levels of automated renewal processing.

Steps to Success

The companies who have begun to travel this path, and are excelling on it, have concluded that developing customized models (using a combination of internal, external and synthetic data) would allow them the best opportunity to reflect the uniqueness of their book of business and distribution system. They have concluded that generic industry models provide little competitive advantage and that the initial step is usually a very challenging process because, historically, data quality and integrity were not considered to be high priority. However, going through this process has enlightened many companies to conclude that their own data may be their largest off-the-balance sheet asset.

Applying the appropriate actuarial rigor and choosing the right statistical approach are the next key steps. Companies have found that this step goes far beyond actuaries choosing a modeling software package and curling up next to the fireplace with their statistical books. It requires years of experience, a deep understanding of the industry, and a realization that the findings must be actionable by the business units.

Those companies who approach predictive modeling solely as a technical or actuarial exercise typically do not succeed. Experience has shown there are three drivers for business success--implementation, implementation and implementation. The implementation of predictive models is not about scoring; rather, it is about how scores, the interpretation and explanation of the scores, and the associated business rules can be deployed to help drive key business decisions. (See "How to Realize the Business Benefits of Predictive Modeling" on page 86.)

Companies who view predictive modeling as a business initiative are able to better define a set of business applications that are enabled by predictive modeling and realize a significant business benefit. In fact, companies typically realize a benefit that represents a five-to-10-times rate of return on their initial investment. Typical applications that companies focus on initially are shown in "Applying Modeling Results" on page 87.

Make It Actionable

The critical success factor for deploying the models is that the key business decisions must be actionable, defensible and measurable, and individuals must be accountable for producing the intended business benefit described above. Successful companies focus on developing a "decile management" approach that links business actions to the indications that are produced by the predictive model. Decile management involves dividing risks into rank-ordered buckets where each bucket represents 10% of risks and then based on the positional ranking and profitability of each decile (bucket), business actions can be assigned to each decile. In other words, individual risk decisions with respect to non-renewal, renewal retention, and pricing are tailored based upon the loss ratio estimation for the subsequent policy term. This is the approach companies use to realize the significant business value that is enabled through the use of predictive models.

Effectively developing and implementing predictive models into a commercial insurance business requires a significant investment of time, money, human capital and intellectual property. The latter is an important consideration because the process to create and leverage models is unique for each company.

Beyond Decile Management

As models for individual lines of business are deployed, companies have learned a number of other considerations could help to produce an evolutionary change in business processes. Initially the focus is learning where the predictive modeling footprint could be expanded in a manner consistent with the company's underwriting philosophy and culture. The process begins with account underwriting and extends to service, expanding pricing detail and determining the key performance measures to validate that the organization is achieving the desired results. Companies ask themselves a number of questions to better understand the options that might be available. For example:

* What will be the account vs. line of business strategy as respects profitability? In other words, will each line of business stand on its own or will line subsidiaries be allowed based on the total account profitability?

* Will there be different levels of service provided in areas such as billing, loss control, audit, claims and customer service based upon the decile ranking?

* Since predictive models enable more detailed pricing for the individual risk, will an expansion of pricing tiers be a likely outcome?

* What are the performance metrics that can be measured by decile and who in the organization will be accountable for the specific actions?

Finally, the last step is to develop a commitment to search for new data sources and risk characteristics so the company can extend its competitive advantage through predictive models.

The Next Generation

Up to this point, companies have taken a very tactical approach to the use of predictive models with sound business reasons for doing so. Now the early adopters are posing the question: How can I leverage my investment in predictive models to gain a strategic advantage over my competitors?

Some companies have expanded the use of predictive models from a product to a client view. Others have used models only to help underwrite the midsize commercial risk, but they also have been used to enter new markets and new products. How the models are implemented might change but they have been deployed very effectively by several companies. They have expanded their underwriting appetite, and now they are poised to grow profitably during a soft pricing cycle.

During the next few years, the winning strategy will be to link predictive models with supporting technology to bring insurance company functionality to the point of sale, independent of the channel of distribution. This means that companies will be able to price, underwrite and complete the entire policy transaction in real time using predictive models and download the completed transaction at the point of sale. For small commercial business, the company that can implement this strategy will likely gain significant market share and achieve a level of profitable growth that outpaces its competitors.

Key Points

* Implementation of predictive models is about how scores, the interpretation of the scores and the associated business rules can be deployed to help drive key business decisions.

* The critical success factor for deploying the models is that the key business decisions must be actionable, defensible and measurable.

* Successful companies develop a "decile management" approach that links business actions to the indications that are produced by the predictive model.

Contributors: Rebecca Amoroso and John Lucker are principals at Deloitte Consulting LLP; James Marino is director, Deloitte Services LLP, and Frank Zizzamia is director, Deloitte Consulting LLP. They can be reached at [e-mail].

Applying Modeling Results

Successful companies define a set of business applications in which
modeling results can be used. Typical targets and expected results are
shown below.

      Business Applications                      Expected Results

Target Non-Renewals                        Loss Ratio Savings--4 point
* Automatic cancel worst-of-the-worst      minimum
Renewal Pricing                            Expense Ratio Savings--2
* More accurate pricing upon renewal       point minimum
* Improve retention
* Fine-tuning pricing to influence
  retention
* Customer service retention programs
New Business Growth                        Automated Renewals--90%
* Apply new insights to target the best
  business
Automate Underwriting Process              Profitable Growth--Better
* Enhance underwriting efficiencies        than the industry
* Disciplined and consistent approach