(Image credit: Dollar Photo Club.)

To the average observer, the insurance industry appears to have fared well over the last few years. But in reality, as of year-end 2014, the combination of capital accumulation, competitive pricing, weak investment returns and rising loss expense has been driving down returns on equity (Source: 2015 EY US Property/Casualty Insurance Outlook). Claims loss represents the largest expense item for an insurance company—regardless of whether it covers health, property casualty or life insurance—ranging from 70 to 80 percent of the operating cost.

Insurance claims leakage—or money lost through failures within existing processes, including inefficient claim processing, human error or fraud—can make up nearly 18 percent of insurance companies’ costs (Source: Accenture–Claims at a Crossroads). To combat this problem, companies use claims audit tools to discover areas of weakness within processes and apply training, business rules and automated tasks to minimize leakage. However, audits have limitations: they only identify incidences that resulted in claims leakage in post mortem and only analyze a sample of the claims, as it is not feasible from an efficiency standpoint to check every claim for leakage.

The ability to monitor and mitigate leakage during the early stages of claims processing is key to an insurance company’s profitability. By identifying points in its process flow that pose the greatest risk for claims leakage and then streamlining and systematizing procedures to incorporate and harness new capabilities, such as machine learning, insurers can make a significant impact on their bottom line.

Machine learning enables insurers to take observations and findings from claims audits, pull those insights upstream and insert them into critical stages of the claims process, including investigation, evaluation and settlement. This allows insurance firms to take action to reduce claims leakage and overpayments before money leaves the firm.

Existing Insurance Claims Process:

Machine learning can automated prediction of which claim has a high probability of resulting in leakage, based on historical data. These higher risk claims can then be treated with a greater level of care or be handled by a higher skilled claims adjuster in order to minimize the aggregate leakage for the portfolio. Machine learning provides deeper insights into the mechanisms responsible for leakage, with these insights captured and leveraged systematically with future claims.

Claims leakage forensics can be automated with machine learning and applied much earlier in the claims processing lifecycle to evaluate the claims and optimize payouts, minimizing the occurrence of leakage. By pinpointing claims with characteristics similar to those of past claims that have experienced any type of leakage, insurance firms can rapidly identify cases that might warrant re-opening and reassessment.

If carriers can achieve even moderate amounts of predictability from their claims data, the overall cost savings can be substantial, both in the reduction of claims leakage and in material improvements in workforce utilization.

Machine Learning Optimized Insurance Claims Process:

In order to apply this machine learning-based approach to claims leakage, an insurance company will need the following capabilities:

• Sufficient claims and audit data: Machine learning is only as good as the data it is fed. Most insurance companies will have a sufficient volume of data to yield the predictability needed to achieve substantial cost savings.
• Big data infrastructure: Of the “three Vs” of big data (volume, velocity and variety), the most difficult in insurance claims is “variety.” The fundamentally unstructured and heterogeneous nature of claims data requires a flexible, unstructured data store.
• Advanced machine learning technology: The complex nature of insurance data makes data preparation, fusion and machine learning modeling non-standard and non-trivial. A tool with advanced machine learning techniques is needed to deal with the special data modalities represented by free text and temporal event data.
• Expertise: The complexity of big data infrastructure and machine learning tools can be daunting. In order to generate valuable insights from their data such as high-risk claims, insurers need skilled individuals to help them navigate that infrastructure and automate key elements of data science

Today, only a subset of all claims can be audited after claims have already been paid out, leading to significant losses. Machine learning magnifies the ability of a small number of auditors to compress their knowledge into a predictive model that can be applied much earlier in the claims process. This allows the breakthrough leap from a claims leakage process that is reactive to one that is proactive—leading to enormous potential savings.

Disclaimer: The views and opinions expressed in this article are those of the authors and do not necessarily reflect the official policy or position their employers.