Insuretech - Improving Claims Management

COVID-19 has created previously unthinkable consequences for our society. Organised crime has been quick to respond, mounting large-scale orchestrated campaigns to defraud customers, preying on fear and anxiety related to COVID-19. A recent KPMG Australia report also showed that more than 80% of senior executives believe their organisation is increasingly vulnerable to fraud.

To protect against fraud, one would normally need to perform numerous checks for any inconsistencies. This can be very time-consuming when done manually. Luckily, Artificial Intelligence (AI) and automation are able to reduce the time burden; which is just one of the many advantages of using technology in fraud detection.

Automation is also very useful in reducing the repetitive manual labour involved in cross-referencing different datasets, which usually takes several hours. Eliminating this would give employees the opportunity to focus on more value-added tasks. Ultimately, the results obtained from such digital solutions are guaranteed to be unbiased and objective in nature and essentially can help minimise human error. 

From a business point of view, Machine Learning allows companies to create test cases and train the software to detect specific fraud. This means solutions are carefully tailored to the particular needs of each company.

KPMG’s Insuretech proof-of-concept was built in response to the increase in travel claims experienced by the insurance industry in the first half of 2020. The key functions being those of detecting fraudulent documents and verifying the accuracy of information submitted with claims. In order to achieve this, our solution makes use of specially built AI models as well as third-party services, brought together through the Microsoft Power Platform. 

In conjunction with automation,  Optical Character Recognition (OCR) was used in order to extract and convert unstructured text from images and pdfs into a structured and standardised format ready for processing. Another element of this solution involves the use of the Object Detection model. This model has been trained on several different airline logos to be able to detect these logos on tickets. Data extracted from supporting documentation through OCR and Object Detection can then be verified against information provided with the claim.

The application of fraudulent document detection is not limited to the insurance industry. It is of relevance to a multitude of use cases, in multiple industries. One of the components of our solution makes use of the Azure Machine Learning Studio in which a Principal Component Analysis (PCA) model, a Machine Learning technique, was selected and trained on a representative dataset of real flight tickets. The dataset consisted of different text fields which are present on standard airline tickets and their corresponding position relative to each other. Therefore, if a ticket is not in the correct format, it will be flagged as being possibly fraudulent.