Automating Insurance Underwriting with Customized AI Solutions

Summary

This client aids insurance underwriters in enhancing accuracy, maximizing capacity, and ensuring traceability. 

They faced challenges with manual underwriting because it was slow and prone to errors. Insurers struggled to handle diverse document formats efficiently, especially complex formats like checkboxes and handwritten notes, which led to inefficiencies.

To address this, we developed a customized AI-driven solution. Using our unique Structured Data Extractor, which leverages advanced Computer Vision tech, we automated data extraction from various insurance documents. 

Our AI-driven solution benefits the underwriting industry, enabling insurers to make smarter, faster, and more accurate underwriting decisions.

‍

Insurance Companies Face Document Processing Challenges

Insurance underwriting involves extensive risk assessment of individuals and assets and determines the terms and conditions of insurance policies based on these risks.

Previously, our client handled underwriting manually by reading supporting documents in PDF, including policies, claims, and other relevant paperwork to extract key information. This manual data processing for our client was labor-intensive and error-prone.

Besides that, the client encountered issues with processing documents due to:

• The high volume of insurance applications
• Varying structures of the data within insurance applications, such as checkboxes, tables, and long answers. Some documents are also submitted in a handwritten format. 
  

Analysts spent significant time extracting data from these various formats, leading to delays and inaccuracies. The extracted data included essential details such as company name, annual revenue, and headcount.

The client further struggled with processing checklists, which often required handling difficult-to-extract checkboxes. This variation and the complexity of insurance documents made the extraction difficult, especially in accurately matching questions with answers and getting relevant information.

Our client also faced challenges with automating their workflow while adhering to high regulatory standards. 

The client partnered with us to address these challenges. 

We developed an AI-driven solution capable of automating the extraction and processing of data from diverse insurance documents, improving efficiency, accuracy, and compliance in the underwriting process. 

Customized AI Solutions Enhance Insurance Document Processing

Implementing our purpose-built Generative AI solution transformed the client’s workflow. Here is a brief overview of what it looks like post-implementation: 

1. Automated document submission:

Customers submit their insurance documents directly into the system.

2. Efficient information extraction:

Upon submission, our system automatically processes the document, and extracts and organizes relevant information.

3. Structured response: 

The final output is a well-organized JSON format with all necessary information for immediate use or further analysis.

Here’s a detailed look at how we achieved this and the technologies we used:

We created separate solutions for every document type our client has to handle. 

We used “Structured Data Extraction,“ a proprietary solution we developed specifically to identify the structure of each document and determine what data (a key-value pair) is needed from that document.

1. Computer vision: to identify the document structure and link to the right text checkbox data
   

Linking to the correct text presented a challenge, so we calibrated a computer vision model to identify the nearest text.

Computer vision enables computers to interpret and understand visual information, such as images and videos, using machine learning and neural networks. It recognizes and analyzes the layout and specific elements of documents. 

We specifically leveraged computer vision to recognize and analyze elements like large checkboxes, which we used as anchors for detecting the nearest text. This capability is important for linking data correctly, especially when dealing with varied formats within insurance documents.

From these anchor points, we directed the system to search for related text (e.g., up or left from the anchor point), ensuring precise data extraction.

2. Optical Character Recognition (OCR): to extract data and handle handwritten text effectively
   

We used state-of-the-art OCR technology to extract text from relevant text boxes and handwritten data across documents. This way, even the most challenging handwritten data can be converted into accurate digital text.

3. JSON response
   

Once we extract the information, including critical data points such as different financial metrics (total gross revenues for current, last, or two years ago), city, state, zip, phone, etc., it is then formatted into a JSON response. This structured data is then seamlessly integrated into our client's systems, improving data management and accessibility.

We specifically used JSON, an open standard format that uses human-readable text, to store and transmit data. It enabled us to efficiently organize the data extracted from insurance documents making it easy to integrate into downstream applications.

Insurance Underwriting with AI Offers Superb Precision and Speed

Our configured Document AI has transformed the client’s underwriting process, drastically reducing manual labor and error rates while improving processing speed and data accuracy. 

Variation in document structures, a known issue in the insurance industry, made it very difficult to process checklists and extract data, particularly from checkboxes. Many fintech companies encounter similar issues and struggle to process such complex documents, highlighting the need for a solution capable of handling diverse document types effectively.

We achieved over 95% accuracy by automating data extraction, ensuring that important information is captured and processed correctly. This minimizes errors and streamlines compliance and decision-making.

We have also cut document processing time per document to under 15 seconds. This improvement allows our client to handle a higher volume of documents efficiently, accelerating the overall underwriting process.

To meet the specific needs of different insurers our client handles, we developed and implemented three custom templates for three different insurance companies. Each template adapts to each company's formats and data fields, further optimizing the document handling process.

By addressing their challenges with our cutting-edge technology and purpose-built AI solutions, we have empowered our client to stay ahead in a competitive market. 

As the client reviews the next steps, this project shows the potential to transform document processing across the insurance industry. It provenly offers a robust automated model for others facing similar challenges. 

Document AI can also serve as the foundation for deeper, end-to-end workflow automation in the future. Other AI Agents, like Decision AI, can use its outputs to automate other tasks—such as complex, decision-making work.