There are many signs that the use of analytics within claims operations is evolving quickly, according to Edward Vandenberg -- including the fact that he was chosen as Farmers' Insurance Group's director of claim analytics and business intelligence. "I'm not a quant," he notes, adding, "In the recent past Farmers would have hired a Ph.D." But Vandenberg doesn't lack for qualifications -- he has worked for global insurance firms for more than 25 years, and for nearly a decade he has helped insurers build and deploy predictive models.
In broad terms, Vandenberg's appointment over that of a hard-core "quant" evinces greater business involvement in the application of analytical models into insurance processes. More specifically, the shift is an acknowledgement that the model-development aspect of analytics is now somewhat stable in the insurance industry; the current challenge, at least for the more advanced carriers, is to operationalize models and take a holistic approach to analytics-driven claims operations.
"Nobody's selling predictive modeling anymore -- we're past that stage," says Vandenberg. "Now it's about putting the plumbing in and having a repeatable deployment integration and model maintenance process."
Today the goal is not to implement a single predictive model as a point solution, but rather to have a factory of models that operates on the basis of continuous improvement, Vandenberg elaborates. "It's about creating and maintaining a series of decision-support services within a service-oriented architecture," he says. "I need to run upwards of a dozen models and update them every year; I need to integrate them within business processes. I need to emphasize change management and process improvement and monitor key metrics that track the performance of the models themselves."
Among Vandenberg's current projects, he relates, is an analysis of field claims offices in danger of falling below a target score for customer service and file quality. "We're building a model that will allow us to know 45 days in advance where we may need to intervene before those scores fall below target," he explains.
The challenge in building an effective model for the task is to assemble the many sources of data and identify the most meaningful metrics, Vandenberg adds. "The only thing that 'shouts' in the data is past performance [of the field offices], which can tempt you to overestimate its importance as a predictor," he comments. "If you focus on predictors that shout, you miss all the whispering in the room."
Building the Analytical Enterprise
Claims has lagged other parts of the insurance industry, such as underwriting and actuarial, in the introduction of predictive analytics for decision support, according to Mark Gorman, principal of Minneapolis-based consultancy Mark B. Gorman & Associates. One reason for this is that company leadership tends to be interested in a different aspect of claims information than the claims professionals themselves, he contends. "Claims people are focused on their performance with regard to the actual resolution of the claim," Gorman explains. "The other parts of the company are more interested in the loss history information that the claim relates."
One of the biggest obstacles to better use of analytics in claims, then, is the need to redirect attention to the descriptive side of the claims process, Gorman concludes. "What happened? How many claims closed? How long did it take? What was the severity? The frequency? What percentage went to litigation?" he elaborates. "Those are the numbers critical to analysis within the claims operation."
In addition to conceptual hurdles, carriers face serious technical impediments, adds Michael Costonis, executive director of Accenture's (New York) North American insurance practice, who points out that there are many unexploited areas in claims that could yield improvements in organizational performance, including injury treatment and management and customer attrition response. "One of the most unexploited areas for analytics in claims is process compliance and prediction of outcomes -- a view that doesn't merely ask, 'Do I have fraud?' but also, 'Am I following the right intervention sets to drive the best possible outcome, whatever the nature of the claim?' " he says.
Unfortunately, even companies that have recognized this opportunity may be challenged to exploit it, suggests Costonis. "There is a massive pipes-and-plumbing problem in the average insurance company in a couple of different ways," he submits.
Giving Life to Analytics
For a start, according to Costonis, data quality within claims organizations is typically poor, which means that models are processing incomplete or ambiguous data. Second, most organizations face a massive struggle to consistently access and match data from other domains, such as underwriting, policy, vendor files and general customer files.
"For example, matching lifetime value to claims is much easier said than done; it requires consistent historical data, and my being able to match that to customer data I have across all products I might have with that customer," Costonis illustrates. "That's a key requisite to putting these models into play. In practice, we see that because infrastructure is so poor and data requires so much treatment, models may take a lot of time to complete and be very difficult to invoke on a real-time basis in the processing of claims."
Current predictive models in claims are not transaction-intensive, Costonis continues. "Current models tend to operate on the fringes rather than robustly interacting with core systems and major data sources," he says. "To get to the next generation, the models need to coexist with real-time transaction processing."
Carriers struggle both with creating the teams to build analytic capabilities and being able to put their work into practice, according to John Lucker, principal, advanced analytics and modeling national practice leader and global human capital leader, Deloitte (New York). "The people side of analytics is much harder than many companies appreciate -- it's very hard to find or organically train people to think about analytics in such a way that it will come alive in your business," he says. "There's an analytic solution, but it must be holistic."
Lucker suggests that carriers have had some success with the essential building blocks of analytics -- which he describes as identifying business problems to be solved, generating the analytics or data mining and predictive modeling solution, and solving the data management problem of that particular solution -- but they have found it difficult to make analytics integral to claims or other operational areas. "Once you've managed the technical analytic outcome, there's the business implementation -- if you're going to use the technical analytic outcome, it has to be integrated with your technology platform," he notes.