Newsletter Sign-up

By THOMAS MCCARTHY, executive vice president of San Diego-based Mitchell International and founder of Mitchell Medical, a strategic business unit of Mitchell

When insurance technology developers first began automating the much-needed medical bill review process for injuries incurred in the casualty setting, many repeatedly fell into the same trap. It was not for lack of trying, but their end product too often resulted in an automated process that was still limiting, awkward and/or inefficient.

When technology developers were tasked with automating calendars that would enable adjusters to see plotted procedures and visits more efficiently without all of the manual headaches, the result was the same five-row and seven-column format--only this time it was on a computer screen instead of on paper.

Developers missed the mark in this case and needed to take automation to the next level by providing a much more in depth tool that incorporated charts by provider or treatment category, such as Gantt charts that show the whole episode of care by provider from date of loss to current date.

That is the trap out of which developers needed to pull themselves--realizing that simply transferring an outdated mechanism into a new format does not significantly improve the actual process of claims handling.

There is a way to escape this trap. Building a tool to automate only a single facet of the claims process, such as the assignment of claims to adjusters, stops short of what is needed and produces nothing more than a tool to assign claims to adjusters.

The skill set issues, overloaded adjusters, turnover, training, and other issues still have not been resolved, resulting in the same inefficient review process. It is the equivalent to dropping a 525 horsepower engine into a Honda Civic without any consideration to changes in suspension, drive train or braking, which ultimately results in less than peak performance. Analytic modeling, however, can streamline the review process.

The following scenario sheds some light on how analytic modeling can help predict claim severity early in the review process based only on attributes found on a medical provider bill. Mitchell analyzed 50,000 Michigan PIP claims to develop a predictive algorithm.

One of the first findings Mitchell's analysis revealed is that incorporating more data into the model substantially increases the statistical confidence level of predictive accuracy--nearly to a point of certainty that would indicate a large indemnity claim. As more attributes and data are collected, the predictability of outcome becomes even more certain.

Hence, accessing a larger volume of data sets to create the model increases the accuracy of the predictions at an even earlier point in the review process. In theory, a sophisticated set of models based on extensive data sets and large amounts of data can confidently predict the severity outcome promptly after first notice of loss.

The trap arises when a model is implemented but no other changes are made to the process. As a result of falling into this shortsighted trap, difficult claims are still assigned to experienced adjusters and basic claims to newer, less experienced adjusters in the hopes of ending up with the best mix to adequately manage their development by raising the stakes as they learn.

Absolutely nothing has changed, except that claims can be assigned faster and with more knowledge of the expected difficulty of handling the claim issues. If management does not harness the underlying power of the model, it is risking a prime opportunity benefit.

There's no doubt that the process needs to change so that the focus shifts towards the attributes of the claim in order for adjusters to take advantage of a more sophisticated tool. Once skill is removed from the front-end of the equation and adjusters focus on process, skill will fall into place naturally.

Let's assume we can divide soft tissue claims into four categories that we can describe as 1) Easy, short term care expected, not a high medical expense 2) Relatively easy but some medical issues (fraud suspicion, causality) need to be examined and decisions made by an adjuster 3) Some degree of difficulty likely requiring professional medical review and 4) Complex, gnarly claims with high exposure and advanced medical treatments including long term care

It only makes sense to engineer a different process for each category since predictive tools will determine which of the above categories is in play with a high degree of certainty. We will use Categories 1 and 4 for illustration. Since claims in the first category are not as complex, a less experienced adjuster can start the process by checking policy to determine coverage up front and then send the claim to auto adjudicate.

Automatic adjudication for all claims classified under Category 1 should include automated bill and code review, processing through a robust business rules engine with payment through electronic funds transfer with electronic remittance advice.

Category 4 claims require a lot of attention and scream for specialization and segmentation since these are more significant injuries that are predicted to be large dollars and complicated claims. While a licensed adjuster should own the claim, the claim could be segmented among a team of specialists who would follow a set process to gain immediate control of the issues.

For example, one team member could gather medical records and extract their salient points, while another manages the case from a medical point of view with the facility and treating physicians. Additionally, a specialist versed in legal issues may review liability and exposure issues and so forth.

This process sounds complex, but all of these functions can be initiated early in the claim using an effective predictive model. In the long term, claims can be further segmented if special services and modification to vehicles and homes is required.

When it comes to dealing with the claims categorized as 2s and 3s, these may follow the generalist model and become a vehicle that management uses to impart knowledge and train staff. As adjusters master 2s, they can advance and begin to handle 3s until they are seasoned enough to quarterback fours.

The message here is clear--we can do so much more than just create an assignment tool with relatively small technology steps if we think broadly about the entire claim continuum. A small step forward in technology can result in a giant leap in process if traps are avoided. If not, a small step in technology results only in a small step in technology.

Avoid the trap by examining and rethinking the entire process surrounding the area where you implement analytic algorithms. When companies first deployed predictive modeling in underwriting and pricing, they had to rethink and redesign how to access, distribute and control the pricing process that had now gone from one price fits all to stratified pricing at many levels. This approach took creative thinking and allowed companies to exploit the Internet to gain market share.

Another way to look at these opportunities is as models that can help give management more control over a process to predict claim severity. Management should maintain that control and continue to integrate forward to assure that the company gains the benefits at hand by redesigning the process of handling the claims now that the outcomes are more certain.

September 1, 2009

Copyright 2009© LRP Publications