Rousmaniere: How Artificial Intelligence Might Work for Us

Conversations at the National Workers’ Compensation and Disability Conference led me to see two possible paths for artificial intelligence in workers’ comp claims. We tend to focus on the more difficult and perhaps not, in the near term, the more rewarding path.

AI has been burdened over the years with the expectation that its job is to help workers’ comp organizations to predict. I have always felt that intelligence is equated the foresight.  But as we close out 2017, the reality is that prediction based on data is endorsed by at most half of organizations in our industry.

Valen Analytics, which develops systems to help property/casualty insurers make better underwriting decisions, writes about “an enduring divide between analytics adopters and naysayers.”  In its 2018 Outlook, issued in early December, it reports from its survey that “48% of underwriters dismiss data for [their own] judgment.”

Rising Medical Solutions’ Workers’ Compensation Benchmarking Study has been probing claims executive attitudes about predictive analytics since its first annual survey five years ago. In its 2017 report, discussed before a large audience in Las Vegas, principal researcher Denise Algire referred to the limited role predictive analytics still plays in claims. 

Her report says that among claims organizations, 32% were engaged in predictive modeling in 2017, an eight percent improvement over the 2014 study.

Both Valen’s and the Rising Medical Solutions’ studies argue that organizations that predict using data perform better. Here is what Algire wrote: “With the growing cost and complexity of claims, utilizing tools to identify high risk cases as early as possible is a clear competitive advantage… Organizations that utilize systems to drive best practices report notably better claim outcomes.” 

Perhaps we are overlooking opportunities that don’t fit into the category of prediction. If AI is to deliver better results, it’s because it fills an important gap in how we conduct business.  AI can help to predict accurately. But it can also help to remove errors in operations. 

To predict accurately – the loss for a policy or the cost of a claim – a prodigious amount of historical data is needed, the data needs to be carefully analyzed, and lessons learned are then applied to new situations.   This data pruning process is arduous, and our psychological dependence on human judgment is profound. Hence the divide between predictive models and actual practice.

The other use of AI is more mundane, easier to implement and easier to accept because we all are well acquainted with how our current practices fail — repeatedly.   This second path for AI came into focus during a conversation with a senior executive at a major claims organization at the Las Vegas conference.  He said, very simply, that humans are very apt to overlook things. This applies to doctors making diagnostic decisions (about 30% are at least initially wrong), and to claims managers riding herd on a hundred open claims. Nobody to my knowledge has studied the error rate among adjusters, but claims audits often uncover serious gaps. Almost always something has not been executed, either due to perfunctory review or failure to complete.  Manual checklists can’t do the job.

Here are three scenarios for reducing errors of omission by workers’ comp claims payers.  Each of them involves the application of various tools to create, in effect, an artificial claims persona.

One example is a virtual medical analyst. According to the Workers’ Compensation Research Institute, the average lost time claim involves 22 medical visits with about three services per visit. Nine of these visits are with a physician. The AI-driven analyst would aggregate the information submitted about these visits, analyze them, and point out where there are variances in diagnosis and treatment that a claims or case manager needs to address.

Another example is a virtual intake specialist. This virtual team member develops a complete FROI, collects other data from the claimant and employer, and makes sure that state and company standards are met regarding completeness and timing. The artificial specialist could engage in synthetic voice interaction with claimant and employer to collect information, with the always present capacity to switch to a human voice.

A third example is a virtual customer service assistant, who will step up the quality of engagement of claimant, worksite, provider, case manager and adjuster. “She” will interface with emails, text messaging, artificial voice phone calls, conveying information and/or collecting information.  She will work largely on an automated prompt basis under the eye of the adjuster.  Everyone will stay informed at pretty much no expense of time or attention by the adjuster.

These three examples show how AI can get people to share information and to be kept in the loop.  Nothing of consequence is overlooked. The claims and case managers lose none of their discretion. And, the predictive potential of AI can rise out of this foundation to really impact outcomes.  

ABOUT THE AUTHOR

Peter Rousmaniere is widely known throughout the workers’ compensation industry, both for his writing and consulting experience. Based in the picture perfect New England town of Woodstock, VT, he is a regular on the conference circuit, and is deeply in tune with trends and developments within the industry. His passion is writing and presenting on issues largely related to immigration, and he maintains a blog on the subject at www.workingimmigrants.com.

Share This Article: