Artificial Intelligence, Machine Learning & The Future of Claims Litigation.

The cost of litigating claims has in the past, always been a study in historical, not behavioral factors. I think most adjusters would agree, that the biggest goal in claims litigation is to avoid it entirely, but unfortunately, that’s not reality. The saddest factor for many carriers is that in some cases, the uncertainty and risk of litigation can end in a claim that costs more to litigate then the claim was worth.

In 2017, the Insurance Information Institute put the cost of defense or containment on commercial multi-peril at 39 cents on the dollar. For medical professional liability, the cost increases to and for product liability it is as high as 77 cents. For workers’ compensation where the employee gives up the right to sue the employer for injuries that happen in the workplace, that number amounts to 13 cents.

No alt text provided for this image

The metrics of predictive cost involving claims litigation are arguably the amount of time spent on settlement discussions and potential trial and trial preparations, expert witnesses when needed, discovery, depositions, plaintiff’s attorney, and that’s just the beginning.

Additional difficulties come when trying to predict the actual possibilities and results of dismissal, defending, or worse paying out a settlement. These have always been historical factors that are as accurate predicted as pulling the lever on a slot machine and predicting what, if anything will come out of it. These analysis are purely reactive, not proactive. While historical data can provide statistical outcomes and a certain “probability” of outcome, it cannot ultimately predict cost or outcome. It becomes even more difficult when a claim, from the onset, looks like it will potentially be litigated when as an alternative, a talented adjuster could ultimately set the expectation of the consumer and alleviate or steer the claim in a more positive direction over time that may determine a favorable outcome for the company and the claimant.

Most consumers don’t have an “appetite” to litigate, it is usually an emotional decision based on a perception of fair or unfair treatment or an unwillingness on the behalf of the carrier to give the consumer what they feel they were promised in good faith when they accepted the policy. Customers don’t tend to remember anything about their carrier other than the way they were treated on their first claim.

The unfortunate thing about historical data is it cannot predict a future outcome, it can only tell you what has happened in the past, not the ever changing variables in the present. but like a slot machine, what if you could accurately predict the statistical outcome or probability of how the machine will react if you pull the lever a certain amount of times? Then the outcome becomes more predictive, which means the odds of accurately predicting the outcome become much, much better.

Probability must meet certainty, but how does that happen? In the real world, it doesn’t. We simply cannot achieve that level of perfection, but with newer technology and predictive approaches, we can get closer and closer to determining a predictable outcome. It may not be perfect, but it will be much closer to perfect than we have been in the past. The past may dictate the present, but it does not have to dictate future outcomes.

So again, how do we get there? How do we provide claims management with the tools necessary not only to alert adjusters to the probability of litigation so they can attempt to steer their claimants away from it, but in the event of the inevitable litigation, how do we predict the outcome and minimize the damage? I’m glad you asked!

According to the ABA (American Bar Association) roughly 36% of law firms are either already using some form of Artificial Intelligence or plan to implement it in the near future. Industry experts cited in both Property and Casualty 360 & Insurance Thought Leadership articles have predicted that AI could evolve to the point where it can determine if a Claim will ultimately end up in litigation with a 90% accuracy rate based on current data and statistics, and that data is what is currently considered structured data, leaving an even higher percentage to be determined if carriers can get a handle on unstructured data.

While some analytics can be achieved by tracking your own litigation trends, it will take quite some time and money in order to acquire enough data to actually achieve desirable and actionable metrics. The other issue is measuring the human interaction between your claims adjusters and the client in order to determine how the litigation was handled or if it could have been potentially avoided. Cutting edge companies provide an platform for measuring very important factors involving the corporate experience with claims litigation through their analytical platform. An example is CaseGlide. There are others but their focuses are usually mainly on billing and litigation costs, not actual cases and the probability of settlement.

West Law was a platform that was well ahead of its time as it pertained to broad legal case historical precedent and analytics but its focus was not primarily claims litigation and it is not an inexpensive solution. It is clear that the larger your data set is, the better the likelihood is in predicting the likelihood that a claim will end up in litigation. But unfortunately, most corporations are looking for answers within, and that vastly limits the extent of predictability. Predictability is best achieved when you have a large amount of relevant data to sample from, and in order to do that, you have to have access to large amounts of case law as it applies to claims litigation. So the question then becomes, how will artificial intelligence help in gathering such data sets as to make this endeavor wholly meaningful?

Statistically, only about 10% of an insurers claims end up in litigation. That 10% however, has the biggest effect on the company’s bottom line. Interestingly, a great deal of analytical data that can be used in determining the liability of a claim entering litigation is located in information that can be gleaned in the FNOL (First Notice of Loss) A well acquainted lawyer with a history of claims litigation can accurately predict if a claim will be litigated by using a practiced eye in the FNOL process to review the Claims Representative Notes(What they told the claimant that can be used against them), interviews (Inconsistencies in Interviews can either help or hurt the Claimant), Statements and Claim Narratives and Diaries (What was said by both the insured and the claimant could indicate inconsistencies that point to litigation or fraud in some cases).

No alt text provided for this image

What’s the Data?

First of all, we have to understand the roles that Artificial Intelligence and Machine Learning will perform in the litigation experience, and understand the distinction between the two. The artificial intelligence piece, while valuable in terms of cost saving and efficiency, plays a secondary role in the process. Machine Learning or ML is the hammer that drives the process, as you want systems to acquire data that makes them smarter and smarter, while AI automates some of the processes that would otherwise require human intervention. As a symbiotic relationship, more ML can drive more intelligent AI, but the prize is always in the data that is acquired as the more you acquire, the lesser or greater the variables may become, but in most cases, patterns become much more clear in litigation when the same data patterns can be reviewed over and over again, and measured against predictable outcomes.

Simply defined, the AI in litigation is derived from the development and use of computer programs that perform tasks that normally require human intelligence. For the foreseeable future, AI capabilities only permit computers to approach human cognitive functions.

Machine Learning is the brains behind the advance of AI. As more data is given to it, it learns more about the nature of the litigation, thereby enabling the AI to achieve more and more cognitive ability.

As for data, the data that we would use is similar to the data that is currently used in looking at the severity of a claim, where litigation is most certainly a factor. This would include third party data from external sources, such as historical litigation data from other companies that have similar lines, similar financial structure, similar geographical risk pools, and similar risk. A larger pool of litigation will lend itself to a larger likelihood of algorithms identifying trends and outcomes that would result in a similar likelihood of litigation and outcome. Working with development and data teams a litigation expert could invariably provide the right elements to create accurate algorithms that can be used to determine the possibilities, and again as the data set grows and more factors are introduced, the ML algorithms can be tweaked to provide a much more definitive and accurate set of outputs.

 “Text mining can be used to delve into such unstructured free form data and help identify co-morbidities that significantly drive up claim severity. Additionally, third party data commenting on the individual’s lifestyle and habits add a layer of information about the claimant that further helps to segment the litigation propensity of the claim. “

It is important to note what exactly it is you are trying to predict, and to ensure that your systems can consume and present the data in such a way as to make it palatable and understandable to it’s audience. You cannot simply throw out statistics and assume that a claims system that is running on some archaic server in your storage closet is going to provide the type of data you would want to acquire and present.

If this is an endeavor your team wants to take on in house, you will need to consider first building out a data structure that is capable of accumulating data from various sources as well as a great deal of unstructured data from external and third party sources via a robust service layer and some automated form of data gathering intelligence, whether via purchasing the data from a data mining company or mining it yourself.

No alt text provided for this image

Algorithms must then be designed alongside predictive algorithms by both data engineers and litigation experts to create models that will accurately predict the cost and potential outcome of a claims litigation with the model building on increasing levels of certainty over time as more and more data becomes available and is acquired and presented.

Models again, can be built with variables derived from a model that incorporates data from FNOL + External Historical + Internal Data = (0) Litigated or (1) Settled as a simply defined metric, but more scores can be added as more variable algorithms are created that can potentially score many facets of a claim and the probability of litigation can be determined with even greater certainty.

The probability of litigation can continue to be measured as more data is added to the claim Post-FNOL. Once the result of the claim is settled, the additional data will be added to the Internal Data and Historical model thereby growing the data pool.

Artificial Intelligence can be used in conjunction with system events and triggers during the life of a claim, by analyzing & acting on the result of an algorithm during the cycle of the claim, certain processes can be triggered by these results and in turn handled by artificial intelligence such as triggered correspondence, notification, and event diaries and claims file updates. This limits human interaction to complex decisions that may not yet be able to be performed by AI, such as financial decisions, reserving, remediation and followup calls.

When a Claims manager can recognize that the resulting data will help determine the deposition of a claim and whether the likelihood is that it will move to litigation, they can now act proactively to put together a strategy to minimize cost and if possible avoid litigation altogether.

With the uncertainties in today’s markets, it is prudent for carriers to look at minimizing and really understanding the full cost of processing, and let alone litigating a claim. That is what will keep the bottom from falling out of your bottom line.

Credits: Insurancethoughtleadership.com, Property and Casualty 360

Feel free to contact us at nieuwendam.com to learn more, or email me at andre@nieuwendam.com

%d bloggers like this: