Over the past decade, the use of predictive modeling in commercial lines has gone from a powerful tool leveraged by a few strategic early adopters, to table stakes for most insurance companies. The industry has seen a tremendous evolution in the use of advanced analytics and predictive models in the underwriting process. On the claim predictive modeling side, we appear to be moving rapidly out of the early adopter stage, partially because of the advent of Software as a Service (SaaS) and cloud computing.
Unlike the early days of underwriting predictive models, claim predictive models can now leverage the cloud as a way of bringing analytics to life faster for insurance companies and self-insured entities that don't have the large scale or resources of some bigger companies. As a result of the changes in technology, many claim organizations are wrestling with the "how" and "when" elements of the deployment equation, versus the historical "if" side.
The firm's claim modeling then expanded to personal lines bodily injury, commercial auto bodily injury, and general liability. For each of these lines, there are a variety of ways to turn modeling insights into actionable business results. For example, use business rules to drive resource allocation based on factors such as the potential claim severity and duration, propensity for fraud or litigation, subrogation opportunity, or other business applications.
Similar to the early adopters on the underwriting side, the early claim adopters have primarily chosen to develop their own in-house customized models and scoring engine capabilities. Although a considerable investment of time, internal resources, and capital is needed, the ultimate financial benefits of claim predictive modeling have been impressive. Early adopters of claim predictive analytics are achieving four to eight percent annual reductions in their loss ratios. These savings have been achieved in part by the redeployment of supervisory resources and claim management improvements by nurses and SIU referrals.
For commercial auto and personal auto bodily injury models that Deloitte has developed for its clients at FNOL, the average claim severity for the best ten percent of claims is usually in the range of 50 percent better than average. For the worst ten percent of claims, the average claim severity is frequently over 200 percent worse than average at FNOL.
Although it takes longer to retrieve information from third-party claimants who may not have much incentive to divulge additional information in a timely fashion, the segmentation power of the models is still significant when external data is leveraged to supplement the traditional internal data available at FNOL. So again, it is no longer a matter of if the models work, but how an organization can leverage the model's segmentation power to develop a competitive advantage in the marketplace.
Deloitte worked with Experian to develop a set of claim predictive analytics offerings that include a hosted solution for insurance companies and self-insureds. The firm says this solution is designed to bring the best of two worlds together—a tested, production-based suite of claim predictive models hosted in a secure, technical environment. This private cloud setting can offer many of the same benefits of a custom-built, stand-alone solution, with five distinct potential advantages:
- Deployment to results cycle time is reduced from years to months. This can in turn translate to a faster ROI for claim organizations.
- Costs can be transformed from a heavy, front-end investment to a more predictable, per-transaction basis or a 'pay as you go' model.
- Technical resource and integration requirements are virtually eliminated.
- Implementation, data, and deployment risks are reduced and shared without compromising security.
- The available data universe may be much larger and more diverse than that of a stand-alone, custom-built implementation.
The basic mechanics of how claim predictive models function in this hosted environment begin with collecting a limited ,but highly prescriptive set of initial data at FNOL (and/or designated points thereafter). This information typically consists of data captured at FNOL, and from other basic sources such as policy/insured/claimant data and external third-party sources.
The model output received from the hosted environment can either be referenced independently, or integrated into the organization's core claim business processes and work flow. Re-submission and re-scoring of claims based on pre-defined claim events, scoring triggers, and as the claim lifecycle evolves can also be designed into the iterative process.
Clearly, there are both functional and technical options available for insurance companies and self-insureds to enable claim predictive modeling capabilities without burdening or relying solely on their IT departments. This is excellent news for claim executives who have been struggling for years to find the right balance between resource allocation, straight-through processing opportunities, and identifying the claims with the largest loss potential before they actually become large losses.