(Image credit: Adobe Stock.)
The rise of data science has posed a challenge to actuaries as modelers-in-chief and as the guardians of data in the insurance industry. With both functions deriving their organizational value from the ability to turn data into insight, there is high potential for conflict between actuaries and data scientists. In some cases, they may regard each other as hidebound traditionalists and dangerous cowboys. But in the best of cases, they can work synergistically and provide a greater overall value to the company. To accomplish this, insurers have employed a number of successful approaches.
Perhaps the most obvious model for data science in an insurance company is to have an independent unit. The data scientists may report to any one particular executive: to an analytics head, to professional services, and of course to a Chief Data Officer (CDO) in companies who have that designation. The stand-alone model is likely the easiest way to integrate data science into an insurance company. Actuaries can easily—and will—keep doing what they trained to do and are good at: rate reserving and regulatory reporting. The data science world will start to manage itself, focusing on new uses for data. Within this scenario, conflict will be avoided because a clear division of labor.
Another way to mitigate conflict between actuarial and data scientists is to force them to work together. One interviewer in Novarica’s recent study on this topic explained the thinking: they don’t want different departments building separate data sources, creating redundancies, and other problems. So a predictive analytics team (situated within underwriting) creates data that everyone throughout the organization then uses. Making actuaries more dependent on data scientists ties them together—now the problem for both is not conflict, but finding the correct use for integrating the data. A possible pitfall to this approach is scrambled messages as clear ownership is crucial. Linking the data management to business goals and objectives may go a long way to ameliorate the discomfort caused a big change like this, especially if management creates feedback loops based on the insights obtained. Proving business value is a proven way of converting naysayers.
A third model is having data scientists embedded within the actuarial function. As this unit grows, it has the potential to become a true predictive analytics group within the actuarial department—these people can focus on 1) engineering—building pipelines, data mining, supporting existing models—and 2) extending the predictive modeling function and supporting actuaries. This model, where instituted with strong institutional support, has been effective at using data science tools. However, it may take longer to become effective than enforced collaboration and a separate data science unit.
In general, data science tends to be poorly understood and expectations may range from wildly optimistic to unduly negative. Insurers who have been most successful in leveraging data science within their organizations have focused on clarity in three key areas:
- Clear Leadership Structure: As with any new technology or department, clear demarcation is important. With something as potentially disruptive as data science encroaching on core actuarial processes, strong leadership is a necessity. The lack of strong guidance is a serious warning signal when data science enters the actuarial space.
- Clear Ownership of Data: Unless there’s a one hundred percent separation between data sources (for example, between unstructured and structured data), which is unlikely and rarely useful, then data scientists and actuaries may struggle over ownership of data. Insurers should make sure that each department/group owns data, and that both can have access when they need it.
- Clear Expectations: Some insurer executives think that data science is more flash than substance. Without buy-in, the support necessary for a flourishing data science program will be very difficult. Leaders must clearly set expectations that outline where data science will be used, and what types of potential value they expect to generate.
Tentative steps towards data science by most insurers should not conceal the fact that the proper integration and use of its methods could very well lead to great improvements in the gathering and use of data. Large hurdles, including institutional inertia, and possibly hostility face even those insurers forward-thinking enough to try this. The unhurried pace at which insurers have begun this process makes sense, but the solutions to these problems are achievable. Data offers meaningful improvements, and it’s only a matter of time before the industry moves forward.