Exploring the Potential of Generative AI in Insurance Underwriting
Imagine an AI assistant that could revolutionize insurance underwriting. What if it could instantly analyze complex medical charts, provide real-time risk assessments, and free up underwriters to focus on what they do best? In this post, we'll explore a future where AI doesn't replace underwriters, but supercharges them. Discover how cutting-edge AWS technologies could transform the daily life of an underwriter, making decisions faster, smarter, and more consistent than ever before.
Tony Howell
Amazon Employee
Published Jul 19, 2024
Insurance underwriting is a complex process that involves assessing risk and making informed decisions based on vast amounts of information. As this information grows in volume and complexity, the industry faces increasing challenges. Could generative AI and agentic workflows offer solutions to these challenges? In this blog series, I will explore this possibility by examining a hypothetical AI underwriting assistant.
There's a saying gaining traction in the AI community: "AI won't replace humans; but humans using AI will replace humans not using AI." This adage perfectly encapsulates the potential of AI in insurance underwriting. The goal isn't to replace human underwriters, but to augment their capabilities, allowing them to work more efficiently and focus on tasks that truly require human judgment and expertise.
Imagine an AI assistant that could transform the way underwriters work. What if it could instantly retrieve relevant information, perform complex calculations, and provide insights on risk factors? This AI wouldn't make final decisions or replace the nuanced understanding that experienced underwriters bring to the table. Instead, it would serve as a powerful tool, enhancing the underwriter's ability to process information, spot patterns, and make informed decisions.
In this post, I'll explore the potential features and benefits of such a system. I'll look at how this AI assistant could:
- Streamline routine tasks, freeing up time for complex analysis
- Provide quick access to up-to-date information, ensuring consistency in applying guidelines
- Offer insights and visualizations that enhance understanding of risk factors
- Support learning and development, especially for new underwriters
By augmenting human capabilities in these ways, an AI assistant could potentially revolutionize the underwriting process. It could lead to faster turnaround times, more consistent assessments, and ultimately, better outcomes for both insurance companies and their customers.
Let's dive into the potential features of our hypothetical AI underwriting assistant and how they could benefit the underwriting process. These types of features are all possible today with generative AI and patterns such as retrieval-augmented-generation (RAG) and agentic workflows.
- Intelligent Information Retrieval
- Function: Uses natural language processing to understand queries and fetch relevant information from company manuals and guidelines.
- Benefit: Reduces time spent searching for information, ensuring access to up-to-date guidelines.
- Example: An underwriter asks, "What are our guidelines for applicants with controlled hypertension?" and receives a concise, relevant answer within seconds.
- Calculation Assistant
- Function: Provides a suite of commonly used calculators in one interface.
- Benefit: Minimizes errors in routine calculations and standardizes the process.
- Example: "Calculate the BMI and percentile for a 10-year-old male with a weight of 120 and a height of 55 inches"
- Rating Table Navigation
- Function: Offers digitized, searchable rating tables with multi-factor lookup.
- Benefit: Speeds up the rating process and enables exploration of how changes in factors affect overall ratings.
- Example: "Show me how changing an applicant's cholesterol level from 210 to 180 would impact the rating for a 45-year-old male smoker"
- Underwriting Guidelines Summarizer
- Function: Generates quick summaries of guidelines for specific conditions, highlighting recent changes.
- Benefit: Ensures underwriters work with current guidelines, improving consistency in decisions.
- Example: "Provide a summary of updated guidelines for diabetes, highlighting changes in A1C thresholds."
- Terminology Explainer
- Function: Acts as a context-aware glossary for industry jargon and medical terms.
- Benefit: Supports ongoing learning and improves communication accuracy.
- Example: Explains complex medical terms like "myocardial infarction" in layman's terms during application review.
- Requirement Checker
- Function: Generates checklists of required documents and medical tests based on applicant information.
- Benefit: Streamlines the application process by ensuring all necessary information is collected upfront.
- Example: "What are the requirements for 45-year-old applicant for a $1 million policy"
- The assistant analyzes the guidelines and applicant details and generates a checklist:
- "Policy Requirements Checklist
- EKG
- Full Blood Work / Panel (cholesterol, glucose, liver enzymes)"
- Risk Factor Education Tool
- Function: Provides in-depth explanations of how different factors impact risk assessment.
- Benefit: Enhances underwriters' understanding of risk factors, leading to more informed decision-making.
- Example: Illustrates how family history, lifestyle, and medical conditions collectively impact cardiovascular risk.
- Medical Chart Analysis
- Function: Analyzes medical charts such as ECGs to identify potential health conditions.
- Benefit: Assists underwriters in interpreting complex medical data, potentially flagging issues that require further investigation.
- Example: Analyzes an applicant's ECG, identifying patterns that suggest atrial fibrillation, prompting the underwriter to request additional cardiac evaluation.
You are Ms. Susan Underwriter. How might this AI assistant transform your daily work? Let's walk through a typical day with this hypothetical tool:
9:00 AM: You start your day with a complex case: a 50-year-old applicant with a history of hypertension and a family history of heart disease, applying for a $2 million policy. Instead of spending your morning poring over manuals and calculating risk factors, you ask the AI assistant for help.
Within minutes, the assistant provides:
- A summary of relevant guidelines for hypertension and family history of heart disease
- Recent policy changes affecting this type of application
- A preliminary risk assessment based on the provided information
- A list of additional tests or information needed
9:15 AM: With this comprehensive overview, you're able to quickly identify the key areas that require your expertise and judgment. You decide additional information is needed and ask the assistant to generate a list of necessary medical tests using the Requirements Checker.
10:30 AM: You receive the applicant's ECG results. Instead of spending time analyzing the complex chart yourself or waiting for a medical expert's review, you use ask the assistant for help. The AI quickly processes the ECG and flags potential irregularities that suggest atrial fibrillation.
11:00 AM: Armed with this insight, you consult with the assistant to understand the risk factors of atrial fibrillation on life insurance risk. This helps you determine that you need more information about the applicant's cardiac health before making a decision.
2:00 PM: A colleague asks for help understanding a new guideline for diabetes risk assessment. You use the you ask the AI assistant to quickly pull up a quick, concise summarization of the Underwriting Guidelines, complete with recent changes highlighted.
3:30 PM: You're working on a particularly complex case involving multiple health conditions. The assistant helps you visualize how these conditions interact, providing insights that help you make a more informed decision.
4:30 PM: As you're wrapping up for the day, you realize you've processed significantly more applications than usual, including several complex cases that would have typically taken much longer. The AI assistant has streamlined many of your routine tasks and provided valuable insights on medical data, allowing you to focus on the aspects of underwriting that truly require human expertise and judgment.
Throughout the day, the AI assistant has helped you:
- Save time on information retrieval and routine calculations
- Ensure consistency in applying the latest guidelines
- Enhance your understanding of complex risk factors and medical data
- Improve accuracy in identifying potential health issues
- Focus more on critical thinking and decision-making
While the AI assistant has been invaluable, your expertise, judgment, and ability to handle nuanced cases remain crucial. The AI doesn't replace your role – it enhances it, allowing you to work more efficiently and focus on the aspects of underwriting where human insight is irreplaceable.
As we look to the future, the potential applications of AI in insurance underwriting are vast and exciting. The evolution of generative AI and large language models opens up new possibilities for creating more sophisticated and capable underwriting assistants.
In our next post, I'll take a deeper dive into how these concepts might be realized in practice using AWS services. Then, I'll explore a proof-of-concept for an AI underwriting assistant leveraging cutting-edge AWS technologies, including:
- Amazon Bedrock: A fully managed service that provides access to high-performing foundation models through an API.
- Amazon Bedrock Knowledge Bases: A capability that allows easy connection of enterprise data sources to foundation models.
- Claude 3: Anthropic's latest and most capable AI model, accessible through Amazon Bedrock.
- Agentic patterns and Tool-calling: We'll explore how tool-calling can be implemented to allow our AI assistant to interact with various underwriting tools and databases, by using tool-calling capabilities of the latest LLMs.
Stay tuned as we continue our exploration of AI in insurance underwriting, demonstrating how AWS technologies can bring these innovative ideas to life and potentially revolutionize the underwriting process.
Any opinions in this post are those of the individual author and may not reflect the opinions of AWS.