AWS Partner Network (APN) Blog
Achieve Hyperautomation for Insurance Document Processing with Hyperscience and AWS
By Priya Chakravarthi, Director of Product, Hyperscience
by Sai Kotagiri, Partner Solution Architect, AWS
Hyperscience |
Rising inflation is driving insurance companies to explore ways to streamline their operations and reduce expenses. They are looking to consolidate their infrastructure and software investments, focusing on tools that offer more comprehensive, end-to-end solutions. The emergence of Large Language Models (LLMs) has enabled insurance companies to re-imagine their existing business processes and develop solutions that increase automation in complex manual tasks such as insurance submission intake, underwriting, and claims management in commercial, property, and liability insurance.
In a previous post , we explored how Hyperscience addresses the limitations of legacy automation solutions to help insurance agencies optimize document processing. In this post, we will explore how the newly unveiled Hyperscience Hypercell, the all-in-one enterprise AI infrastructure software platform, allows insurance companies to leverage Hyperscience’s deep integration with the AWS services. This integration will allow them to get more value from their existing investments by developing turnkey automation solutions.
Hyperscience, an AWS Partner and AWS Marketplace seller, offers proprietary AI models, applications, and workflows through the Hyperscience Hypercell. The solution delivers unmatched [99.5%] accuracy and [98.0%] automation across the entire spectrum of documents that insurance companies rely on to run their business processes. The Hyperscience Hypercell also supports integrations with foundational models as well as commercial Large Language Models (LLMs) such as Claude 3 available through Amazon Bedrock using custom code blocks. This helps insurance companies automate their previously manual business processes by leveraging their investments in both Hyperscience and AWS.
Challenge: Insurance Submission and Underwriting is a Complex and labor-intensive process
The insurance industry relies on standardized forms for commercial insurance submission intake and underwriting process. Insurance companies must process a variety of documents, including Statement of Verification (SOV), ACORD forms, supplemental forms, and loss run reports, submitted by their customers through brokers.
The first step in the risk assessment and underwriting process involves identifying any missing or invalid information in the submissions and performing data validation across all the submitted documents. This is a complex and labor-intensive task, often involving multiple manual and repetitive steps to sanitize data, validate business rules, and request missing information from the customer
Figure 1: Sample ACORD 125 form commonly known as Commercial Insurance Application. It is used to record business location, description, contact details, prior insurance and loss history information.
Hyperscience’s approach
Hyperscience’s Hypercell introduces capabilities that aid insurance customers in planning, assembling, and deploying workflows that can automate complex and labor intensive document processing.
Planning a workflow
When planning a workflow, customers must evaluate the various capabilities offered by the Hyperscience Hypercell and determine which ones add value to their operations. The solution offers the following capabilities to automate the insurance intake process:
- Ingestion of documents from AWS S3 buckets and simultaneously classifying them as ACORD 125, loss run report, supplemental documents etc. This eliminates manual ingest and sorting processes during insurance document intake.
- Collation and extraction of relevant information accurately across multiple documents.
- Elimination of downstream impact by validating extracted coverage information with liability coverage limits using a Large Language Model (LLM) such as Claude 3 available through Amazon Bedrock.
- Industry-leading supervision capabilities that flags any discrepancy between the coverage limits extracted from the ACORD form and the applicant’s preference extracted from supplemental application.
- Incorporation of Subject Matter Expert (SME) feedback into the workflow by generating a task for a human to accept or reject the discrepancy in the application.
- Support for data extraction with LLM, to retrieve information such as the agency name, applicant’s name, location, and other relevant details. The response from the LLM can then be used to either accept or reject the application.
Assembling a workflow
Hyperscience Hypercell enables insurance companies to privately train AI models using their own enterprise data and provide a user-friendly interface for business users to supervise results based on their domain expertise. Customers can assemble their workflows in 3 stages.
1. Setup Phase: In this phase customers can configure templates, S3 buckets, security credentials, and accuracy settings. Customers can:
- Provide native S3 integration for secure document ingestion and output storage. End-to-end security and access control is ensured by using the customer’s secure AWS Access Key ID (for SaaS deployments) or their EC2 instance’s IAM role credentials (for on-prem deployments).
- Integrate insurance company’s private, enterprise data with Hyperscience’s proven AI models by leveraging Hypercell’s model lifecycle management and training data management systems to maintain auditability, traceability, and governance of these models. The training and deployment of Hyperscience’s proprietary models can be done without the need for a data scientist or coder. Business users can easily train, QA, and supervise the process using their intuition and expertise.
- Easily configure the expected accuracy of the workflow that would determine the automation of the entire workflow. Hyperscience’s proprietary AI models provide unmatched accuracy and automation across the entire spectrum of documents and information assets that insurance companies rely on to run their business processes.
Figure 2: Document processing flow setup in Hypercell – defining S3 buckets as input listener and output notifier .
2. Build Phase In this phase, customers can build upon Hyperscience’s out-of-the-box Intelligent Document Processing (IDP) workflow to develop a comprehensive workflow tailored for commercial insurance intake. Customers can:
- Utilize the pre-deployed building blocks with Hyperscience’s proprietary models, trainable using customer data.
- Provide real-time data validation by integrating with Anthropic’s Claude 3 (hosted on Amazon Bedrock), using configurable prompts from pre-built blocks or custom code.
- Additionally, make human supervision part of the workflow to mitigate any discrepancies.
Figure 3: Coverage information across the ACORD form and supplemental documentation is validated by making API call to Anthropic Claude 3 model hosted on Amazon Bedrock, using a custom code block.
Hyperscience Hypercell empowers insurance companies to unlock the full potential of AI while maintaining control over their data and processes, ensuring compliance and mitigating risks through human supervision and expertise.
3. Processing Phase The screenshot below demonstrates the execution of the workflow for the commercial insurance document intake process in the Hyperscience Hypercell. The workflow demonstrates how Hyperscience can process, sanitize, and validate commercial insurance documents. This is accomplished through a turnkey no-code, low-code interface that allows business users to supervise results based on their domain expertise. The business user can analyze every step of the automated workflow and examine the inputs, outputs, and logged messages for each processing block.
Figure 4: End-to-end workflow of commercial insurance documents intake process from within Hyperscience’s Hypercell. Anthropic Claude 3 model, hosted on Amazon Bedrock, is used for validating commercial insurance documents.
Deploying a workflow
The application is containerized and can be deployed in your organization’s public cloud using Amazon Elastic Kubernetes Service (Amazon EKS). Its modular architecture allows the application to be scaled horizontally by simply adding new nodes to the Kubernetes cluster. This allows Hyperscience to be scaled up for high-volume use cases to process millions of documents daily. Hyperscience Hypercell can be deployed to use AWS RDS as a database, Amazon SQS for asynchronous processing of data streams, and can be deployed across multiple AWS Availability Zones for high availability.
Conclusion
With the launch of the Hyperscience Hypercell and out-of-the-box S3 integration, Hyperscience increases the number of direct integrations it supports with AWS products, allowing common customers in the insurance vertical to automate complex insurance use cases. Hyperscience’s turnkey enterprise AI infrastructure allows insurance companies to leverage LLMs hosted on Amazon Bedrock in a safe, intelligent, and curated environment. Using the power of Hyperscience Hypercell deployed on scalable AWS infrastructure, Insurance companies can hyperautomate repetitive manual tasks mitigating business risks associated with hallucinations and inaccuracies, and reduce operational costs.
Hyperscience – AWS Partner Spotlight
Hyperscience is an AWS Advanced Technology Partner and AWS Competency Partner that offers an Intelligent Document Processing (IDP) solution which delivers customers increased value by providing the highest levels of automation and accuracy to make human work more impactful and data more actionable.