AI Slashes Insurance Review Times by 100%

A global specialty insurance leader faced slow underwriting due to manual review of lengthy reports. They implemented an AI solution that automatically summarizes these reports, slashing review times from 10 days to 10 minutes. This 100x faster analysis allows them to serve more clients while maintaining expert-level quality in risk assessment.

Client:

A global leader in specialty insurance, this client boasts a presence in Bermuda, Europe, the UK, and the US. They manage over $3 billion in written premiums annually. Combining extensive experience with a strong financial position, they provide clients with a broad spectrum of insurance, reinsurance, and digital risk management solutions.

Problem Statement:

● Time-consuming manual report review: Subject-matter experts (SMEs) spent days manually reviewing lengthy (100-page) engineering reports for each potential client, creating a bottleneck in the underwriting process and delaying policy issuance decisions.

● Limited scalability: The manual approach to report review limited the company’s ability to serve a larger number of high-value enterprise clients, hindering their growth potential.

● Risk of errors: The manual process of extracting and summarizing information from the reports was prone to human error and inconsistencies, potentially leading to inaccurate risk assessments and underwriting decisions.

Results:

☑️ 10-minute reviews: AI summaries slashed review times from 10 days to 10 minutes.
☑️ 100x faster analysis: They can now analyze 100 times more reports, enabling faster growth.
☑️ Maintained quality: AI summaries match expert quality, ensuring accurate risk assessment.
☑️ Increased efficiency: Streamlined process reduces manual work and allows for faster scaling.

AI Solution:

The insurance company deployed Provectus’ Generative AI solution powered by Cohere’s large language model (LLM) hosted on Amazon Bedrock. This AI system automatically summarizes engineering reports, extracting key information like object details, positive and negative aspects, and KPIs.

How it Works:

Data Acquisition: The client provided 100 sample reports and manually-created summaries for training the AI.

1. Processing Pipeline: The AI pipeline used a combination of tools:
● SpaCy library: Orchestrated key-value extraction, text segmentation, and classification.
● Cohere Embeddings LLM: Converted document sections into vectors for analysis.
● Cohere Command LLM: Performed question answering and information extraction.
● Cohere Classify models: Identified positive and negative points of interest.
2. Summary Generation: The AI generated summaries capturing object information, key points, and KPIs, mimicking the format of human-created summaries.

The solution offered a user-friendly interface for underwriters to access and review AI-generated summaries.

References:

1. Transforming Risk Management in Insurance Underwriting with Generative AI

Industry: Insurance
Vendor: Provectus
Client: A global specialty insurance company