AWS Only
Miller Insurance Unlocks Data Insights with Retrieval Augmented Generation
Miller Insurance faced the challenge of unlocking insights from the large volume of unstructured data stored in its OpenText document management system. Firemind partnered with Miller to develop a Retrieval Augmented Generation (RAG) solution, leveraging AWS services to enable Miller's employees to query the data using natural language and uncover valuable insights to drive data-driven decision making.
AWS ONLY ·
Miller Insurance Unlocks Data Insights with Retrieval Augmented Generation
Miller Insurance faced the challenge of unlocking insights from the large volume of unstructured data stored in its OpenText document management system. Firemind partnered with Miller to develop a Retrieval Augmented Generation (RAG) solution, leveraging AWS services to enable Miller's employees to query the data using natural language and uncover valuable insights to drive data-driven decision making.
At a glance
Miller Insurance is a specialist insurance broker that acts as an intermediary between clients and insurers, creating and exchanging a large volume of insurance-related documents and data.
Challenge
Miller Insurance's OpenText document management system contained over 16 terabytes of unstructured data, making it challenging for employees to effectively leverage the data to uncover insights and drive decision-making.
Solution
Firemind partnered with Miller to develop a Retrieval Augmented Generation (RAG) solution, enabling Miller's employees to query the data using natural language.
Services Used
Amazon Bedrock
Amazon Kendra
AWS Lambda
Amazon S3
Outcomes
9/10 accuracy on claim file responses
Claude 2.1 superior model for summarisation accuracy
Business challenges
Underutilised Data
Miller Insurance stores a significant amount of data, around 16 terabytes, in its Open Text document management system.
However, much of this data remained underutilised, as the company was unable to effectively analyse and interpret the information contained within.
The data stored in the Open Text system was a mixture of emails, PDFs, Excel files, and other unstructured formats, with limited metadata available.
This made it difficult for Miller Insurance to understand the content and context of the data, hindering their ability to unlock new use cases and enhance decision-making processes.
"Firemind's PULSE tool was able to quickly and accurately summarise vast datasets of customer surveys, ensuring we could take action against any negative customer experience elements."
Dean Macfadyen, Data Platform Engineer — Firemind Generative AI Customer
Solution
Data ingestion, indexing, NLP and LLMs
Firemind explored the use of the OpenText API to access the metadata associated with the documents stored in the 16TB document management system. We developed a custom Lambda function to identify the different file types (e.g., Word, Excel, PDF) in the .dat files, which lacked metadata. The supported file types were then indexed using Amazon Kendra, a managed search service that can ingest and query unstructured data.
Natural Language Processing (NLP) and Question Answering
Firemind utilised large language models (LLMs) from Anthropic, specifically the Claude Instant and Claude 2.1 models, to enable natural language interaction with the indexed data. The team optimised the prompts and the approach of querying Amazon Kendra to improve the accuracy and relevance of the responses.
They found that the Claude 2.1 model performed better in describing and summarising the information compared to the earlier Claude Instant model.
User interface and interaction
The goal was to provide a simple question-answer (prompt-response) interface that would allow Miller Insurance employees, including non-technical staff, to ask questions about the data in plain language.
Firemind aimed to make the interaction with the LLM easier and more accessible, empowering users across different departments to reason against the data and make informed decisions.
Scalability and production readiness
The PoC highlighted the need for further data preprocessing and integration with the OpenText API to create a more production-ready solution for the future.
Model Spotlight
Claude 2.1
We chose Claude 2.1 for its superior summarisation and accuracy capabilities, achieving 9/10 accuracy on claims file responses.
It excelled at handling complex queries, providing detailed and context-rich answers, making it essential for extracting deeper insights from Miller’s large data set.
Claude Instant
We chose Claude Instant for its speed and efficiency, making it ideal for quickly processing Miller Insurance’s 16 terabytes of unstructured data.
Its low-latency performance allowed employees to query data in real-time, making it well-suited for fast, iterative testing during the POC.
Why Firemind
“This was our first look into generative AI with AWS, and we were astonished by the fast results and summarisations using our own prompts."
Firemind was recognised as an excellent partner to explore this opportunity with AI and ML. The early engagement between First Central allowed Firemind to demonstrate the potential with a part of their toolkit; a generative AI demo that uses a Large Language Model (LLM) to reveal trends and opportunities from customer feedback data.
This demonstration allowed First Central to see what possibilities exist for leveraging their own data to drive efficiency and agent performance. Connecting a similar solution to their call centre and customer data proved how Generative AI and ML can achieve their desired business outcomes.
Broadened use cases
Unlocking insights from previously underutilised data opens up new opportunities for Miller Insurance to explore innovative use cases, such as trend analysis, risk assessment, and customer behaviour insights, enhancing their service offerings.
Scalability for future growth
The scalable nature of the solution ensures that as Miller Insurance continues to grow and accumulate more data, the system can handle increasing volumes without degradation in performance, supporting long-term business needs.
Added value
With quicker access to relevant data, Miller Insurance can respond more effectively to client inquiries, improving customer satisfaction and strengthening client relationships.
Get in touch
Want to learn more?
Seen a specific case study or insight and want to learn more? Or thinking about your next project? Drop us a message below!