Next-level risk modeling for the insurance industry
Developing a predictive model that keeps problems at the center establishes a holistic approach to solving issues.
Most insurers have adopted risk modeling in their business to one degree or another by now, but despite early successes, many are failing to adjust and evolve and are missing out on significant savings and value.
Early in the adoption of risk models, insurance companies tended to build in-house and often focused on pricing models for a large line of business. At this point, risk modeling was new to insurance, and the stakes were high. As such, considerable effort was put into identifying and acquiring data scientists and analytics experts to develop these models. Understanding this was not their area of expertise, insurance leaders created internal research and development (R&D) groups to develop and apply these risk models.
The data scientists and analytics experts needed to develop predictive models can be costly, but when you are starting out, and there is a knowledge gap, it is a necessary investment. However, once the knowledge and experts are available in-house, further applications of predictive analytics become less costly, and programs are implemented more smoothly. Once you have the expertise and in-house talent, your perspective naturally turns to other applications for other business problems and opportunities.
However, senior business leaders within the business tended to divest themselves of the oversight and responsibility for guiding their company’s analytics program, thinking they would leave it to the experts. This almost always proves to be a miscalculation, as the activities of the R&D group become misaligned with the overall business priorities.
Most data scientists and analytics experts thrive in creating cutting-edge risk models but are not always as adept at or passionate about the tedious work of implementing and monitoring models. These meticulously crafted risk models create little business value if they never get implemented or end up being one-time efforts that grow stale over time.
When a risk model is created to solve a business problem, it should always be viewed as an ongoing program. It should always be regarded as a business problem that needs to be addressed rather than simply the development of a risk model. Predictive models can be a perfect solution to certain business problems, but a model is finite, whereas problems tend to persist over time.
Factors to ponder
When thinking about a business problem that requires a predictive or risk model, consider the following steps/process:
- Addressing a new business problem: When you address a new business problem, expertise and resources are necessary to come up with an intelligent solution. Using data scientists and analytics experts, although costly, is appropriate for scoping the problem, gathering data, and creating a predictive model.
- Implementing a solution: Implementing predictive models has become easier over time, and there are now standard approaches insurers can leverage, but the application and maintenance of such risk models should not be an afterthought. Know how the predictive model will be implemented before you begin and consider how best to supply the needed data and integrate the results with your core systems and business.
- Monitoring performance: It is imprudent to assume a predictive model will work exactly as expected in the design phase. Surprises and unintended consequences are the norms in all business undertakings. Well-performing models may drift or change over time, particularly as your business evolves. Expect changes and find technologies and resources to support your ongoing program.
- Refreshing the business solution: Monitor your progress, and when the monitoring indicates, refresh your data and predictive model. Even if comprehensive re-working is not required, revisions still create a new predictive model. Insurers need to expect and plan for refreshes of predictive models whenever they tackle a business problem.
- Reconsidering the business problem: Eventually, it may be decided that the approach taken to build the original predictive model could be substantially improved if reconsidering the problem from scratch. Here the data scientists and analytics experts on the R&D team can be brought back in to exercise their specific skills.
Once you consider all these steps, you see that the predictive model is just one piece of a holistic effort to solve a problem. Centering on the business problem, not just the model, is the right approach.
There is also the issue of wanting analytics applied to more and more business purposes. It is imprudent to expect to keep expanding the R&D team for each new problem addressed if we also are asking them to manage the entire process. It is not cost-efficient, nor are the specialists in these teams the appropriate resources to be used for monitoring or refreshing models. The proper expertise and resources should be assigned for different steps in the process, allowing insurers to be more cost-effective and successful in addressing each problem. This may require software and tools that encourage teamwork and transparency.
For those who want to build a data-driven organization, you should start thinking about technology platforms that allow you to scale the predictive modeling process. Evaluate tools that can integrate with core systems to streamline and automate data collection and integration and remove implementation pain points.
The insurance industry has come to a tipping point where executives should be asking: How are we managing our risk modeling program? And can we be doing things more efficiently?
Taking burdens off your data scientists and analytics experts will free them up to address other problems and opportunities in your business. Applied thoughtfully and efficiently, predictive analytics can continue to transform your business in new ways and new areas.
Chris Cooksey is the head actuary for data and analytics at Guidewire Software, Inc.
Related: