The Value of Predicting Claims Litigation with Artificial Intelligence
The time-tested axiom in risk management: “If it’s predictable, it’s preventable,”¹ holds true today in the business of property and casualty insurance claims. What’s impressive is that we can take that axiom and energize it with artificial intelligence, machine learning, and natural language processing. Early applications of this technology have shown promise with a single carrier able to reduce claims from a major storm by at least $10 million by predicting potential litigation.
The story of how to analyze a claim lies in the unstructured data. There is a great deal of information contained within the first notice of loss (FNOL), claims representative notes, interviews, and statements by the insured and claimant. A domain expert can glean a lot from this unstructured data and make effective predictions such as the likelihood of the claim going into litigation and indications of organized activity.
Replicating the above process in software is a challenging task. Traditional machine learning techniques typically require an expert to label each claim in a sample set with its predicted outcome, which is tedious and time-consuming. Even with that, the results are only as good as the sample set of data is a good representation of future claims (think biased data). Additionally, most machine learning models perform poorly in the presence of skewed data. Less than five percent of a carrier’s claims typically end up in litigation. However, that five percent is the claims which have the biggest impact to the company’s bottom line. Accurately identifying those claims early in the process can save the carrier millions of dollars. This requires novel approaches using a combination of various artificial intelligence (AI), natural language processing (NLP) and machine learning (ML) techniques.
The solution to the weakness of the existing system is to combine AI with expert systems. An insurer can get into the expert AI systems with the goal of replicating the expert’s knowledge and skills.
This combined system answers questions like: which claim will be litigated? Which claims are being facilitated by professionals masking their identities? Which claims are fraudulent, and what was the specific scheme used by the insured or claimants? Asking the right questions and knowing which words and phrases to look for are the first steps in predicting if the claim will fall into claims litigation.
An example of a company that has been able to apply this process is Infinilytics. They combine their expert system with AI to create a platform that learns new words and phrases from the set of claims using unsupervised leading techniques. This system can then overlay rules on unstructured data to determine behavioral patterns and analyze sentiments. Integrating this data creates a workable system to predict claims.
Applying this technology to the claims process inside of carriers will enable legal departments to cut costs a prioritize claim with the highest risk of large settlements and predict up to 90% of claims that could result in litigation.
Infinilytics
SVIA Startup Member
See Sri Ramaswamy, CEO of Infinilytics speak at InsurTech FUSION 2019 Rise of a Digital Insurance Industry: LIFE, HEALTH, WEALTH and P&C 2.0 on June 18-19 in San Francisco, CA.
