Predictive Analytics in Claims: Clearing the 6 Hurdles

Predictive analytics has been proven to improve claims-fraud management, and over the past several years has played a starring role in changing how insurers look at pricing. But forget claims fraud and pricing for a moment. Industry experts now are identifying new opportunities to apply predictive analytics in the claims process, and those advancements could mean big changes in the industry.

"We're starting to see some indications in the market that disruption may be starting to build from a claims standpoint," says Mark Gorman, CEO and founder of The Gorman Group, an insurance business-analytics and integration consultancy. "We're also seeing that the claims themselves are becoming both more complex and better adapted to increased automation. By more complex, I mean, what are insurers going to do with a claim when it's for an automobile that's been designed not to have an accident; how do you adjust for that?"

Gorman also says applying predictive analytics in claims triage, especially with adjuster assignments, can help insurers control staffing and costs. "We have to make sure we have the right people focused on the right decisions," Gorman says. "This is where predictive analytics is being used to support decision-making, not to replace it. Providing automated information and decision support to get them up and running and productive faster is increasingly becoming a driver in the market."

Another area of interest for insurers is after the first-notice-of-loss, Gorman says. "What I'm hearing most from senior management is the need for early identification of the surprise claim; the claim that looks like it's going to be $6,000 or $7,000 and becomes a $200,000 or $300,000 claim, and nobody seems to understand why," he says. "Predictive models applied not just at first loss or first report of injury but during the claim process, should enable them to much more rapidly identify and deal with them, because those are the claims that give everybody heartache."

For some recent research, Gorman and Stephen Swenson, insurance development executive at analytics and business intelligence software provider SAS, spoke with predictive analytics executives serving their claims department counterparts, and claims executives responsible for departmental results, about the biggest challenges they face in expanding the use of predictive analytics in claims. "I see a pretty big disparity around claims and predictive analytics in claims. Some of the larger insurers say they have a number of predictive models that are interacting with claims information, but they are separate and distinct from the business processes running reports," Swenson says. "They're trying to get the models to interact with the actual business process on a more real-time basis, looking at the impact of those model scores and then trying to bring them into the claim."

For their report, "Building Believers: How to Expand the Use of Predictive Analytics in Claims," Swenson and Gorman highlighted six hurdles and action items to clear them.

Hurdle No. 1: Uncertainty About Predictive Analytic Results

Action items:

* Clarify the business objectives.

* Focus on gleaning clear and well-defined metrics from the predictive analytic initiative.

* Ensure analytic resources are in place.

Clarity, resources, the talent to relay the business objectives, and the more granular metrics derived, are important. Insurers need to have in place the people who understand what the claims department is looking for, so that they don't build analytics that match what they already know, Gorman says. "Historically the primary focus [for predictive analytics professionals] has been on pricing precision and it was driven that way by one of the market disrupters. When the focus goes to claim management, those resources need to take a different tact and making sure that they have the flexibility to do so, which has been a challenge in some organizations."

Small insurers tend to have a tougher time. "Many of the smaller insurers just don't have predictive modeling expertise," Swenson says. "Most of that is going to be housed in their underwriting actuarial arena," which can pose challenges. "The organizational priorities are different when you're an actuary vs. when you're trying to run a claims organization. So they may not have access to the type of skill sets and talent that can help people in the claims organization understand the thought process around what the predictive analytics can do for them."

Hurdle No. 2:Threat of Litigation Influencing Decisions

Action items:

* Avoid the impulse to react to the potential legal risk.

* Balance the need for change with appropriate risk evaluation.

* Make sure your model lifecycle process and approach are well documented.

Data access and management capabilities need to be in place in order for claims departments to operate on the data, Swenson says. "Whether it's some type of a claims data warehouse, individualized reporting marts or something similar, insurers need to be able to get at that data and they need the predictive analytic and mining tools to be able to create predictive models to do different types of modeling on claims data," he says. "For example, to return a predictive score for a personalized auto model at FNOL and have that score represented in the actual claim adjuster's screens. They can utilize that information as additional insight. As the data becomes richer, you need a model-management type of capability that continuously calibrates those models, so that they're always being tuned and the information is not being dismissed by the claim adjuster because the model is not right, or the model has become aged in such a way that the scoring doesn't reflect a real/true understanding of today's environment."

Gorman points out that predictive analytics in claims are a decision-support tool, not a decision automation tool. "If we are consistent about how the rules are applied and how the information is utilized and know that it's an enhancement to the way decisions are made to bring greater clarity and improve the speed with which a claim can be processed, adjusted and adjudicated, we avoid most of the litigation issues. Where insurers get into litigation is when there's a perception that the predictive analytic models are being applied without the application of business acumen by the claim adjusters. So making sure that there's good documentation across that life cycle is important."

Hurdle No. 3: Competing IT and Claims Processing Projects

Action items:

* Incorporate predictive analytic data standards into all claims IT projects.

* Move from an "either/or" to a "both/and" mindset when developing business requirements for core system replacement.

* Focus on well-defined and clear performance outcomes of the predictive analytic initiative.

Gorman and Swenson break insurers into three groups: those that have already done a claims system change, those in the process of doing a claims system change and those that haven't yet started.

"For those who have already done so, then the whole process becomes an evaluation and standardization of how they're capturing data out of that system, how they're storing it and how they're analyzing it, so that's standard," Gorman says.

And, Swenson says, there's great opportunity in this group. "Those organizations that have gone down a claims transformation project, that have put a new system in place, should aggressively be deploying predictive analytics throughout those claims. Oftentimes they have pretty good data that surrounds these systems, so they've gotten their data in a place where they can gain access to it. Predictive analytics really serves to optimize those investments."

Those insurers falling into the other two groups have a tendency to view this as an either/or issue, Gorman says. "It does not have to be an either/or situation. Insurers need to apply the business acumen from a requirement standpoint to understand how that data is going to be captured, stored, deployed and utilized when that new system is up and running. It becomes a focus on data as a corporate asset, not an operational and transactional requirement. Insurers that are doing this - even though they stop and say, 'we're not going to do anything about this until the system is fully deployed; but we're going to do all of our requirements, so we know how that system is designed and is going to be usable the minute it gets deployed,' - are jump-starting this process."

Hurdle No. 4: Failure to Pursue Both Indemnity and LAE Impacts

Action items:

* Build awareness of and commitment to a holistic approach, not a singular focus.

* Build and implement both frequency and severity models.

* Design flexible rule and workflow environments.

"This is the fastest-moving piece [of predictive analytics in claims], especially anywhere there's a medical-related claim. But that doesn't mean it doesn't apply to others as well," Gorman says. "From the LAE (loss-adjustment expense) perspective, two issues emerged [in the research]: the vendor management around litigation and then, just beginning to emerge, the issue around subrogation."

Taking a holistic view of claims becomes crucial, Swenson says. "Claims transformation projects are trying to drive out, not necessarily the indemnity, but the overall cost; the infrastructure in trying to streamline those things. We found that insurers are a little bit more concerned about trying to go back to the pure loss of claims. Ultimately, that's a very doable situation with predictive analytics, if they've got good history of how claims have performed in the past."

Hurdle No. 5: Relying on Adjuster Intuition - Not Data - to Make Decisions

Action items:

* Start where you know the data will support an effective model-building effort.

* Build 'believership' among key people from senior management through select field adjusters.

* Don't require adjusters to base claims decisions solely on the model; let model scoring be "an additional piece of information."

After the data is in place, having the right people, those with the vision of what to do with the data, is a must, Gorman says. "Interestingly enough, we've been able to find out that there's oftentimes more vision than people within the organization know because they just haven't asked the question and removed the constraints."

Business and functioning units also need an information roadmap that conveys clarity. "[Insurers] have a tendency to look at transactional roadmaps and functional roadmaps and transaction-processing roadmaps," Gorman says. "We don't have as much of a tendency to look at information roadmaps. Insurers that are [paying attention to information roadmaps] are making great leaps, especially as they involve a multitude of people within the organization."

Hurdle No. 6: Limited Change Management Focus

Action items:

* Put key individuals in place to manage the development and deployment aspects of predictive model claims scoring.

* Align people, processes and technology in a comprehensive, project-driven approach.

Change management often is a challenge for any project, and indeed, is one of the biggest hurdles for claims analytics, Gorman says. "Information without business acumen doesn't really go anywhere," he says. "The critical skill set is in balancing the four components of change: people, process, technology and organizations."

In his work with insurers, Swenson sees people applying tools and technologies to processes they don't know or understand, and the credibility of technology begins to suffer. "There has to be a clear understanding of the process and how the technology helps that process."

It all comes back to people, Gorman says. "In addition to the technology, insurers need to make sure they have the resources available to use it. Don't ignore the fact that in insurance, while a directive may come from the top or an opportunity may be identified from the bottom up, the true/lasting change happens 'middle out.' If 'believership' can be built with the people who are responsible for the operations and processing of the organization, true change starts to take place."