About the Customer
Known for its strong commitment to financial strength, integrity, and customer service, our client is a diversified financial services company that provides a wide range of products and services, including life insurance, retirement plans, investment services, and wealth management solutions.Â
Focusing on helping individuals and businesses achieve financial security, the client serves millions of customers across the United States through a network of financial professionals, advisors, and strategic partnerships.Â
The Customer's Challenge
The client was experiencing bottlenecks for customer contract questions, especially as the company grew, and needed to implement automation to process these inquiries more efficiently. Before introducing an AI solution, customers of the organization experienced an average response time of 3 days consisting of an associate taking the contract question, exchanging multiple clarifying emails, manually identifying and reviewing the appropriate contract for relevant language, and responding to the customer.Â
With the volume of inquiries increasing annually by 40 percent, the business needed help scaling its intake process and reducing response time, which directly impacted the customer experience.Â
Additionally, the responses coming from the labor-intensive manual lookup of relevant contract language and information often overlooked key details, lacked completeness in answering customer questions, and created additional communication churn when finishing a customer intake.
phData's Solution
As part of a leadership-driven, company-wide initiative to leverage AI in all parts of their business, the financial firm needed data science to begin managing the uptick in customer inquiries with their contracts. phData was contracted to execute a solution and proposed building a custom chatbot powered by a vector database of the firm’s customer contracts on AWS.Â
With the goal of reducing customer response time and the back-and-forth of clarifying emails exchanged, the retrieval augmented generation (RAG) solution would aim to pull customer contract information quickly and efficiently when prompted by an associate at the company.Â
The three components that needed implementation by phData to create the solution were:
- A contract RAG ingestion pipeline
- A vector database
- The chatbot user interface, with each of these components needing to be hosted securely within an AWS services architecture
The RAG ingestion pipeline is powered by the invocation of an AWS Lambda function. The function retrieves the Contract PDF & metadata and converts the PDF to pure text using Amazon Textract. The text is broken up into chunks with LangChain, and vector embeddings are created from the text chunks using Amazon Titan LLM. The vector embeddings are stored, along with contract metadata, in a Lance DB vector database using an AWS S3 bucket as its object store.
The RAG Chatbot is interfaced using a Stremlit UI application in a Docker Container running inside Kubernetes on AWS EKS. An associate sets input parameters in drop-down menus that reflect the metadata stored with the vectors. After using the inputs to retrieve relevant vectors from the vector database, both the original question and the vectors are submitted to the Claude Sonnet 3.5 LLM from Amazon Bedrock, and the LLM’s response is presented to the associate in the Streamlit UI.
Results
As a result of the newly introduced AI application by phData, contract questions that previously took days to retrieve manually are researched immediately with a retrieval time of less than five seconds. Previously, a team of five full-time employees could handle the research of ~250 contract questions monthly, and due to a consistent backlog, response time averaged three days. With the help of the chatbot solution, the five full-time roles have been reduced to the part-time responsibility of prompting the AI chatbot to generate same-day responses – a 70%+ reduction in response time.Â
Even with an expected 40% increase in annual customer contract questions, the firm will not require an increase in labor investment. By reducing the manual work of the full-time employee, phData was able to save the company $400K in labor costs annually.Â
The solution was a cost-effective way for the company to start utilizing AI within its organization and with phData showing the incremental value of AI, the organization plans to accelerate future AI projects.Â
Expected yearly increase in annual customer questions will not require an increase in labor investment.
Automatic retrieval of contract information.
Reduction in client query response time.
Yearly cost savings in reducing manual work of FTEs.
Why phData?
phData is the leading AI and data services company. We specialize in AI and data applications, from conception to production. Our global delivery team partners with the world’s top brands to execute data initiatives in artificial intelligence, data engineering, applications, analytics, and managed services for cloud platforms.
Take the next step
with phData.
Learn how phData can help solve your most challenging data analytics and machine learning problems.
