A leading global insurance provider focused on foreign investment property insurance receives more than 100,000 claims per year, and as fraudulent activity becomes more sophisticated, it’s critical that the company stays several steps ahead of those attempting to game the system.
Estimates put fraud at approximately 10% of the property/casualty insurance industry’s incurred losses each year, resulting in $34 billion in losses annually, according to data from the Insurance Information Institute.
An important early step in detecting fraud is identifying the factors that lead to fraud. What specific events typically occur before, during, and after fraudulent activity takes place? What other characteristics are generally seen with fraudulent claims? The insurance provider traditionally had attempted to detect fraud by using only a rule-based system and human experts. As the number of insurance claims continues to increase, the company’s current fraud-detection solution simply wasn’t scalable and prevented them from examining other key features within the data to identify potentially fraudulent cases.
The insurer was facing three major challenges when attempting to detect fraud at scale. First, the company’s rule-based, manual approach was not able to incorporate the ever-increasing volume of unstructured data, which prevented them from discovering valuable business insights that could be incorporated into the modeling.
Second, the company was not able to scale the process due to its reliance on legacy technology and manual work, and therefore was not able to keep up with the sophisticated fraud patterns that continue to rapidly evolve.
Finally the company faced a pain point with AI adoption that is felt across industries: the adoption of AI required hiring a best-in-class data science team, and the team would then have to build, test, train and deploy the AI models into production — requiring months of work and the potential loss of millions of dollars in payouts. Enter OneClick.ai.
The insurer adopted OneClick.ai’s platform to automate the process of designing, training, and deploying a custom fraud detection AI model via API to scale as the business continues to grow.
Once the historical record data was uploaded to the OneClick.ai platform, a custom model was designed for the company to deploy into production in the time it would normally take a Data Scientist weeks if not months. By fully automating this process, the insurance company only needed to dedicate one employee who had a background in programming, and was able to perform basic data processing and query the database using SQL.
Using the OneClick.ai platform, a custom deep learning model was generated, at 96% accuracy and 80% recall, with no feature engineering needed by the insurance company. Additionally, the model was deployed in a matter of days from start to finish to ensure fraud cases are now detected as early as possible, dramatically reducing the amount of false claims and payouts that were previously being approved.
- High accuracy and recall: The custom deep learning model generated by OneClick.ai had 96% accuracy and 80% recall.
- Save time and resources: No feature engineering required by the insurance provider.
- Reduce fraudulent payouts: The insurer deployed the new model in days, not months, significantly reducing false claim submissions and payouts, ultimately saving the company money.
OneClick.ai aims to help both tech and non-tech users to adopt advancements in AI technologies. OneClick.ai has automated the process and transformed how AI projects are planned and implemented in businesses. Working with all data types (numerical, categorical, text, image and time series or dates – and any combination together), without any AI technical background required, users can benefit from the efficiency of automation in building various AI applications like sales forecasting, customer retention, click prediction, image classification, object recognition, recommendation systems, etc.