a magnifying glass sitting on top of a piece of paper

Business Problem:

A prominent insurance company was facing challenges in managing and streamlining its customer support services. Customer queries, ranging from policy inquiries to claims information, often led to inefficiencies and delays. The company needed a solution to enhance its customer support process and provide faster, more accurate responses to customer inquiries.


Step Involved:


Claims Settlement & Policy Enquiry automation for US Motor Insurer


1.User Query Processing
  • User Query Submission: Users initiate queries in JSON format, which are communicated through AWS Connect.
  • Intent Detection: AWS Lex, a natural language understanding service, plays a crucial role in identifying the intent behind each user query. This intent categorization determines whether the query pertains to structured or unstructured data.
2.Structured Data Processing
  • User Query Handling: Customers interacted with the system by submitting queries in JSON format. AWS Lex, a natural language understanding service, played a pivotal role in processing and understanding these queries.
  • Data Preprocessing: Structured data, including information from insurance policies and claims, is stored in a combination of Amazon RDS databases and data warehouses. Regular preprocessing is carried out to ensure data cleanliness.
  • Clean Data Storage: After cleaning, the structured data is saved in an Amazon S3 bucket within a designated “clean” folder.
  • Data Retrieval: Predefined prompts are generated based on the user’s intent and schema information derived from the structured data. These prompts are sent to a data retrieval service that fetches the necessary information from the structured data sources.
  • Response Generation: The data retrieval service processes the queries and generates structured responses, making them easily understandable for the users.
3.Unstructured Data Processing
  • User Query Handling: For user queries categorized as related to unstructured data, a separate path was followed.
  • Data Preprocessing: Unstructured documents, such as insurance policies and claims documents, are stored in an Amazon S3 bucket. These documents undergo preprocessing to ensure data cleanliness and consistency.
  • Embedding Vectors: Amazon Kendra, a search service, plays a crucial role in indexing and creating embedding vectors for the unstructured documents in the S3 bucket.
  • Similarity Search: When a user query was identified as related to unstructured data, AWS Lex generated an embedding vector for the query text. This vector was subsequently used to compare against the embedding vectors of unstructured documents stored in Amazon Kendra. This similarity search process identified relevant documents.
  • Response Generation: Relevant unstructured documents are retrieved based on similarity. These documents, along with the user’s query and a predefined prompt, form the input for a foundation model, referred to as Bedrock Claude.
4.Interaction with Bedrock Claude
  • Foundation Model Processing: The query, along with the identified relevant documents and prompts, is passed to Bedrock Claude. Bedrock Claude is responsible for processing this input.
  • Response Generation: Bedrock Claude generates responses based on the input, and these responses can be consumed by the users in English or any other specified language.
5.Path for Structured Data
  • Prompt Creation: For structured data, predefined prompts are generated based on the schema of the RDS database or data warehouse.
  • Dynamic Prompt: If the schema changes, the prompt is dynamically updated to reflect these changes. Otherwise, it remains static.
  • Interaction with Bedrock Claude: The structured query, such as a request for total account balance information, is passed to Bedrock Claude. Bedrock Claude interfaces with APIs to retrieve the result.
6.System Components

The entire use case leverages several AWS services and components, including Amazon Kendra, Amazon S3, AWS Glue, Amazon EC2, AWS Lex, prompts, embeddings, APIs, and Bedrock Claude.

This architecture allows for seamless handling of user queries, offering efficient responses for both structured and unstructured data requests. It provides a powerful system for managing, retrieving, and generating data-driven responses while ensuring data quality and consistency.

  • Amazon Kendra: Amazon Kendra is an intelligent search and question-answering service provided by Amazon Web Services (AWS). It is designed to make it easier for organizations to search and discover information across large amounts of data. Kendra leverages various natural language processing (NLP) and machine learning techniques to provide highly accurate and context-aware search results.

  • AWS Glue: AWS Glue is a fully managed extract, transform, and load (ETL) service offered by AWS. It is designed to simplify the process of preparing and loading data from various sources into data lakes, data warehouses, and other data stores for analysis, reporting, and other data processing tasks. AWS Glue is a serverless service, which means that you do not need to provision or manage servers; AWS handles all the underlying infrastructure for you.
  • Amazon s3: Amazon S3, or Amazon Simple Storage Service, is a scalable and durable object storage service provided by Amazon Web Services. It is designed to store and retrieve any amount of data, from anywhere on the web, making it an essential building block for various cloud-based applications and services.
  • AWS Lex: Amazon Lex is a service provided by Amazon Web Services that allows developers to build conversational interfaces, or chatbots, using natural language understanding and processing capabilities. It enables you to create chatbots and voice-based applications that can understand and respond to user queries in a conversational manner.


  • Enhanced Customer Support: The innovative approach led to a remarkable improvement in customer support, providing efficient and accurate responses to customer queries, irrespective of complexity.
  • Reduced Response Time: The combination of automation and intelligent data retrieval significantly reduced the time taken to respond to customer inquiries, leading to enhanced customer satisfaction.
  • Improved Accuracy: By relying on structured data and unstructured documents, the system provided precise information, mitigating errors in customer support responses.


  • Data Preprocessing: Ensuring data quality for unstructured documents posed an ongoing challenge. The company continuously invested in refining and streamlining the preprocessing pipeline.
  • Schema Changes: Any alterations in the structured data schema necessitated updates to predefined prompts. The system’s flexibility allowed for dynamic prompt adjustments to maintain alignment.
  • Authorization and Security: Data security and authorization management were of paramount importance to safeguard sensitive customer data. Robust security measures were put in place to ensure compliance and protection.

Future Directions

The company remains dedicated to the continuous enhancement of data retrieval and response generation algorithms. This commitment ensures the system remains at the forefront of advancements in AI and data processing. Plans are underway to expand the system’s capabilities to support multiple languages, catering to a diverse customer base. This will involve language processing and translation modules. To further enhance user experience, the company is exploring the addition of a voice interface for customer interactions. This innovation will make customer support even more accessible and user-friendly.


The implementation of this advanced AI-driven system delivered significant improvements in customer support operations. The company experienced a substantial reduction in response time and reported higher customer satisfaction due to the system’s ability to provide accurate and efficient responses to customer inquiries.

  • Enhanced Customer Support: Customer support operations saw significant improvements, with an efficient, responsive, and error-reducing system in place.
  • Improved Efficiency: The reduction in response time directly impacted on customer satisfaction, as quicker query resolution led to happier customers.
  • High Accuracy: The accuracy of responses was significantly enhanced, reducing errors, and ensuring that customers received the most precise and up-to-date information.

This case study serves as a model for organizations seeking to transform their customer support processes by harnessing the power of AI and data retrieval techniques. It highlights the tangible benefits of adopting innovative solutions to improve customer interactions and support services.