This company is a diversified property and casualty insurance group that serves businesses and individuals in specialty and niche markets. They offer products that include standard commercial insurance, specialty commercial insurance and personal insurance.
Baker Tilly’s internal audit team conducted an annual contract audit for the company’s reinsurance contracts. This contract resided in text narratives, tabular formats and completed business forms stored in PDFs. The contract data contained rules on inclusion criteria, exclusion criteria, coverage thresholds and other parameters used in the premium calculation. Because the contract rules were manually extracted, which required significant effort, only a sample set of 10 contracts were audited.
To reduce risk, the company wanted to audit the remaining 500+ reinsurance contracts and subcontracts.
Baker Tilly worked with the client to create an automated process that use machine learning (ML) and optical character recognition (OCR) conversions of the PDF contracts. There were many subcontracts that has at least 50 contract parameters that needed to be identified using Amazon SageMaker ML models and Amazon Bedrock. The contract parameters were then extracted out by Baker Tilly to compare the actual financial transactions to assure they followed the parameters and terms of the contracts to support testing and auditing all payment transactions and contracts.
The parts of the OCR conversion, using Amazon Textract, with low confidence were reviewed using a human review. Contract data was represented in key-value and tabular formats which were both used to audit the premium payment calculations.
The automated solution reduced manual effort to extract contact rules which allowed all contracts to be audited using all transactions from the audit period. This provided the company with the following benefits:
