While risk management information systems are distinguished from claims management systems for their benchmarking and business intelligence functions, some experts see predictive modeling as the next phase in the RMIS evolution.
Predictive modeling feeds historical data through algorithms to devise a statistical model and gauge the probability of future events.
“We've always done retrospective looks at data and benchmarking,” said Russell Lindberg, vice president of sales and marketing at Kaysville, Utah-based RMIS provider MountainView Software Corp., a RMIS unit of Gallagher Bassett Services Inc. “I think the next big thing will be predictive modeling.”
Risk managers accustomed to running complex queries on sets of data should be able to move to predictive modeling, he said.
“It's about looking at data in a more automated fashion and running queries against the data to find anomalies and trends and then pointing them out to a risk manager that they might now have the time, desire or experience to know where to look,” Mr. Lindberg said. “It's already here in some forms, but it will continue to get better.”
While many insurers use predictive modeling in areas such as underwriting and detecting claims fraud, it has yet to find a foothold among risk managers, said Mark Stergio, CEO and senior vice president of Atlanta-based Risk Sciences Group Inc.
“You think about how long we have been talking as an industry about predictive analytics and not much has been done,” Mr. Stergio said. “However, I think predictive analytics will soon have a significant impact on the RMIS world and how claims are adjudicated.”
Mr. Stergio said the Crawford & Co. unit soon will implement a capability on its bundled RMIS to score claims on several characteristics.
“We are planning a full implementation of models that will predict the probability of litigation, fraud, subrogation and return-to-work estimates,” he said. “We will score every historical claim; and anytime there is a change on a claim, we will screen it again and see how that probability changes and how we can react to that.”
Bob Adams, director of industry software and solutions at Falls Church, Va.-based Computer Sciences Corp., said predictive models could be helpful in areas such as workers compensation, where a model could help spot claims most likely to benefit from early intervention.
“Risk managers are being required to provide and focus at the granular level within claims, to be able to predict with good accuracy a high-cost claim the second it comes in the door,” Mr. Adams said. “Resources are often limited; and by having this further analysis, it allows risk managers to focus on the smaller percentage of the claims that cost the highest and are most likely to grow out of control.”
Aaron Shapiro, executive vice president at Glencoe, Ill.-based RMIS provider Origami Risk L.L.C., said a RMIS that truly incorporates predictive analytics must enable users to push and pull data from several predictive or benchmarking sources.
“What we are hearing is that risk managers are looking for a way to hold their private data up against multiple predictive data sources,” Mr. Shapiro said. “Any one individual source is biased to a particular environment, so the claim handling procedures and processes of one particular TPA is not necessarily entirely indicative of the outcomes you ought to have on a particular set of claims.”
David P. Duden, Hartford, Conn.-based director of Deloitte Consulting L.L.P., said even with disparate data streams incorporated, use of predictive modeling via RMIS will require risk managers to craft “synthetic variables” from existing data.
For example, a risk manager could take internal data about a risk and combine that with data from public records location information culled from a geo-coding device and enter the data points into a model (see chart).
However, Mr. Duden warned that the level of statistical acumen needed to devise reliable predictive models likely is outside the wherewithal of even the most advanced risk management departments and RMIS providers.
“Building models and making them actionable is tough,” he said, noting that the “quants” adept at crafting advanced statistical modeling are in high demand in many industries and especially on Wall Street. “The people part is hardest,” he said.
So the broader use of predictive models in RMIS will take greater collaboration between RMIS providers and predictive modeling heavyweights such as International Business Machines Corp. and SAS Group. “These advanced analytic models require a huge infrastructure investment,” Mr. Duden said. “Even the largest broker-based RMIS providers likely don't have the budget to do this alone.”