Client background

The client is a prominent organization within the insurance industry and has offered comprehensive property and casualty (P&C) insurance solutions since its founding. Renowned for its innovative risk management strategies, this insurance organization ensures dependable coverage and peace of mind for its policyholders. Its global presence and unwavering commitment to quality service have solidified its reputation as a trusted leader in the insurance sector.

The business challenge

The insurer faced a significant challenge in processing over 80,000 claimant questionnaires, each containing 30 pages and over 260 questions. These questionnaires, representing individual claims in a class action settlement, were designed and administered by a third party, who provided the individual responses in PDF format. Manual extraction was initially considered but presented several key challenges such as:

• Data extraction complexity: The questionnaires were unstructured and complex, featuring multiple variant versions, skip logic, data tables and varying-length free-form responses.
• Speed to insights: The insurer needed faster visibility into new liability exposure obligations.
• Accuracy and consistency: Acquiring data from unstructured data needed to be accurate and consistent.
• Cost minimization: Manually processing each questionnaire presented a significant cost.
• PII abstraction: Personally identifiable information (PII) needed to be identified and hidden.
• System integration: The individual responses needed to be integrated into several IT applications for claims management.

Strategy and solution

Baker Tilly's advanced business technology services team collaborated with Baker Tilly’s forensic, litigation and valuation services team to employ a three-phase process to deliver AI-powered solutions at scale, incorporating vulnerability scanning, monitoring and elastic cloud capacity. The phases consisted of:

1. Discovery phase: Identified key decisions and relevant metrics.
2. Pilot phase: Demonstrated the feasibility and value of the solution on a small scale.
3. Scale phase: Deployed the workflow for large-scale implementation in the cloud, adjusting capacity based on demand.

The core features of the solution included:

• Automated OCR: Convert PDFs into structured, searchable text using Amazon Textract
• Context-aware AI: Extract key data points from free form text using Amazon Bedrock and Amazon OpenSearch Service.
• Automated quality assurance: Measure the effectiveness and accuracy of data extraction using AWS Batch and Amazon Elastic Container Service.
• Manual intervention: Human-in-the-loop (HIL) experts are able to review and correct data as needed using a Streamlit user interface.
• Business intelligence: Visualize insights and monitor processes through intuitive dashboards using Amazon QuickSight.
• System integration: Provided structured data back into core systems for business operations and review.

The implementation of Baker Tilly’s AI-powered solution delivered significant improvements and benefits for the insurer. Key results of the project included:

• Reduced processing time per questionnaire by 79%.
• Saved $1,750,000 in manual processing costs.
• Improved data consistency by 2% for high priority data attributes.
• Reduced liability exposure visibility time from two weeks to three days.
• Enabled determination of liability exposure based on individual responses shortly after arrival.

These measurable outcomes have led to increased company and stakeholder satisfaction with faster response times and greater transparency. Likewise, settlement payments are now expedited, which has improved overall process efficiency.