Leveraging AI to Mitigate Fraud Payouts

By January 23, 2019 February 18th, 2020 Blog
AI and Fraud

The insurance industry is undergoing a transformation by an increasingly digital world. Customers expect faster claim processing and payouts while insurance companies are looking at ways to reduce the cost of processing claims. Using AI to automate various parts of the claims process is a step in the right direction to achieve both these goals. While undertaking this journey, it is important that fraud detection not take a backseat.

Why is fraud detection important?

Insurance fraud is not a victimless crime. According to the FBI, fraud costs the average US family $400 to $700 per year in the form of increased premiums [2]. Insureds end up footing the bill for fraudulent claims because insurance carriers are unable to differentiate between fraudulent claim payouts and adjust their premiums to ensure they have enough reserves to pay for future claims. Insurance industry puts fraud at about 10% of P&C losses each year [1]. This was estimated to be about 30 billion per year between 2013 and 2017 for P&C Insurance companies [1].

Impact of fraud on insurance carriers:

Insurance carrier’s losses are directly proportional to the premiums. This makes early fraud detection and deterrence a competitive advantage. Carriers that can avoid paying fraudulent claims can offer lower premiums. However, there are costs associated to building and maintaining fraud detection models and maintaining a special investigation unit. In 1990 insurers reported getting back $27 for every dollar they invested in antifraud efforts [1]. The challenge is to maximize the amount of fraud saved by increasing the effectiveness of fraud detection systems and SIU resource utilization.

Pain points with fraud detection systems:

According to the insurance information institute [1], majority of insurance carriers use automated red flags or business rules to detect fraud. Some of the most common issues with this approach of detecting fraud are listed below:

False Positives

Fraud detection systems that solely depend on business rules or static models, often end up flagging genuine claims as being fraudulent. For example: Not all claims filed shortly after increasing coverage limits are fraudulent, as the limits could have changed during the renewal process or due to other life events. Processing false positives have a negative impact on SIU efficiency. Customer experience also suffers due to the added delay in processing claims.

Static models or rule based detection systems

Fraudsters are constantly innovating and finding new ways to exploit insurance carriers. Manually analyzing new fraud schemes, building new rules or models to detect these schemes requires a considerable amount of money and effort. Over time these systems become complex, hard to maintain and end up losing their efficiency.

Lack of evidence for fraud prediction

SIU efficiency is impacted by Fraud detection systems that provide a fraud confidence score without providing supporting evidence. Claims flagged by these systems are usually processed through the same preliminary pipeline without having any hints at what to look for.

4 ways AI can revolutionize insurance fraud detection and processing:

1. Provide reasoning behind fraud predictions:
Due to advances in technology, the days where AI algorithms could only provide a “yes” or “no” prediction are behind us. Algorithms are now capable of providing reasoning behind their predictions. This helps SIU resources to process fraudulent claims faster by providing preliminary findings. Classification of flagged claims by dynamic fraud categories provides dependable measures such as accuracy, cost, effectiveness, etc. that help in calculating ROI.

2. Continuous learning:
Establishing a feedback process between SIU and the fraud detection algorithms enable these algorithms to continuously learn and improve. Reinforcement learning can be used continuously to improve the accuracy of the underlying algorithms. This process can also be used to teach the algorithms new fraud categories and patterns without having to manually maintain any red flag or business rules.

3. Entity resolution and link analysis:
Entity resolution is the process of uniquely identifying entities in a dataset. Link analysis algorithms are only effective when entity resolution processes have successfully identified duplicate entities and has accurately linked various entities and information together. AI algorithms can operate on such datasets to provide outliers, common entities across claims, etc. that can be very effective in detecting new fraud rings and schemes.

4. Processing unstructured data:
Algorithms have the capability to process accident images to determine the extent of damage and estimate payout. They can also process images to detect damage files in a previous claim. These algorithms can also be used to detect duplicate claims filed with similar evidence. Video and sensor data can be analyzed to determine a potential for a staged accident. Claims notes can be analyzed to extract information that is used in link analysis and other fraud detection algorithms.


The ultimate goal of an insurance carrier is to maximize the amount saved by detecting fraud while minimizing the amount invested in SIU and fraud detection systems. AI algorithms have the capability to continuously learn, evolve and aid SIU resources in processing fraudulent claims. With these capabilities, AI is uniquely positioned to make a big impact in the way today’s insurance industry prevents fraud and protects their insureds.

1. https://www.iii.org/article/background-on-insurance-fraud
2. https://www.fbi.gov/stats-services/publications/insurance-fraud

Siddhartha Vowles

Siddhartha Vowles

Siddhartha (Sid) has over 8 years of designing and customizing products and solutions in the insurance domain. He graduated from Carnegie Mellon University with a master's degree specializing in artificial intelligence. Read More Posts

Leave a Reply