Contact us
Search

  • LTIMindtree Euclid – Agentic AI for Insurance

    LTIMindtree Euclid – Agentic AI for Insurance

Overview

The insurance industry is in the process of embracing Artificial Intelligence (AI) in every aspect, which has in turn transformed the entire value chain – translating to an accelerated pace of technological innovation. Insurers are integrating AI into their core operations, from underwriting and claims to distribution and regulatory mandates. Towards this, platform-driven Agentic and Headless AI services have enabled the seamless infusion of intelligent capabilities into insurance enterprises, helping them realize intelligent value chains. The ability of insurers to harness AI capabilities will be pivotal to understand customer needs, optimize operations, and drive sustainable growth. LTIMindtree has defined a platform-centric approach to establish an enterprise-grade Agentic AI ecosystem delivering actionable intelligence to the insurance business.

Insurance Tech Office

The Insurance Technology office is a dynamic hub of innovation, dedicated to transforming the insurance industry through advanced technological solutions. We are at the forefront of the latest technologies, consistently developing modern tools and platforms that cater specifically to the needs of insurance companies.

By harnessing the power of Generative AI (Gen AI), we create business solutions that not only streamline operations but also enhance decision-making processes. Our platforms are designed to improve efficiency, reduce costs, and provide superior customer experience.

Our comprehensive efforts have positioned us as pioneers and leaders in the field, driving the adoption of new technologies and setting new standards for industry. On the other hand, our innovative culture ensures that insurance companies can stay competitive in an ever-evolving market, offering unique solutions that meet the demands of the modern world.

In the year 2024 alone, we filed five patents to create innovative solutions in the Gen AI and Agentic AI domains.

LTIMindtree Euclid for Insurance

Gen AI has generated significant excitement for its ability to create intelligent, multimodal responses to natural language requests. However, its limitations become evident when tackling complex business challenges requiring autonomous intelligence tailored to specific contexts. These challenges include the need for advanced data integration, deeper contextual understanding, and the mitigation of hallucination and bias-related issues. To fully harness the potential of next-generation AI capabilities, a well-designed AI ecosystem is essential one that integrates agentic AI capabilities with robust data, models, knowledge, responsible AI principles, security measures, and governance frameworks.

LTIMindtree’s Euclid platform is designed to deliver a comprehensive enterprise-grade Agentic AI ecosystem to establish a value extraction framework for business enterprises.

Euclid delivers actionable Intelligence to the insurance business by orchestrating autonomous AI Agents that perform several business functions across the value chain. Its primary components include –

EUCLID Platform

EUCLID Agents

EUCLID Agents

Composes new Agentic AI solutions by orchestrating an existing pool of autonomous AI agents. It also enables business users to create new agents from scratch.

EUCLID Knowledge

EUCLID Knowledge

Helps automate the discovery of domain-specific knowledge from unstructured documents in the form of ontologies, taxonomies, and knowledge graphs. It manages, matches, and compares ontologies and knowledge artifacts, and ensures security and PII attributes identification, etc.

EUCLID Foundation

EUCLID Foundation

Delivers comprehensive enterprise-ready Gen AI and Agentic AI ecosystems. Its key features include a Gen AI Façade, RAG Pipelines, Mindful AI, Security Guardrails, Semantic Caching, FinOps, Platform Administration and Governance, etc. It can also integrate with any third-party AI governance tool.

EUCLID Data

EUCLID Data

Offers automated data pipelines to extract contextual data sets from structured and unstructured data sources. Its key features include Intelligent Document Processing, Multi-modal Content Extraction, Embeddings generation and Vector Store Creation, domain-specific data APIs for EUCLID Agents, etc.

EUCLID Models

EUCLID Models

This is a platform for automated Models/ LLMs’ lifecycle management. Key features include – Automated collection of user’s feedback, Data exploration and readiness checks, Model fine-tuning and customization, Reinforcement Learning, Model serving, Model setup, hosting, governance and lifecycle management.

Euclid @Scale:

EUCLID @Scale

Leveraging the Nvidia enterprise AI ecosystem, Euclid @Scale supports AWS, Azure, and Nvidia-native deployments on K8 clusters. It helps ensure Euclid’s deployment at scale.

LTIMindtree Euclid Microlab

LTIMindtree’s Euclid Microlab is designed to accelerate your AI journey with an exclusive Gen AI sandbox available on Azure and AWS Cloud – the platform can explore demonstrable solutions with real datasets.

Key Features:

  • Ready-to-use Gen AI capabilities in a box
  • Exclusive Gen AI sandbox on Azure/AWS Cloud
  • Demonstrable Gen AI solutions using real datasets
  • Assistance in identifying the right LLM for specific use cases
  • Implementation of industry-popular Gen AI solution patterns
  • Secure and private with access controls and audit logs
  • Powered by Canvas AI and integrated with multiple LLMs
  • Consulting and handholding for tech and functional needs
  • GenAI evaluation capabilities, including RAG and model evaluation

Success Stories

Financial Research Co-Pilot

Increased productivity and accuracy with 60% savings

Business Problem

  • Fund research experts spend 40-80 hours preparing each report, which are critical for enterprises aiming to optimize their investments.
  • The process is highly manual, leading to long turnaround times, with customers often unwilling to wait up to two weeks for these reports.

Business Solution

  • Enhanced the productivity of financial researchers by automating report drafting using Gen AI’s Natural Language Generation (NLG) capabilities.
  • Pre-trained Gen AI models on historical financial research reports to replicate human writing styles and ensure quality comparable to manual reports.
  • Leveraged Gen AI to summarize large source PDFs and generate specific, relevant sections of research reports efficiently.

Potential Benefits

  • Significantly improved productivity and report accuracy.
  • Achieved a 60% reduction in time and costs associated with the process.

AI Ecosystem Stack

  • Azure OpenAI, GPT 4 Omni, RAG

AI-powered Claim Summarization for Claim Adjudicators

Enhance efficiency by intelligently consolidating claim information, enabling adjudicators to quickly grasp the status and critical details of a claim

Business Problem

  • Claim adjudicators currently spend significant time reviewing multiple sections and pages of a claim to gather key information and assess its status.
  • The process is tedious and time-consuming, requiring careful analysis of each section and cross-referencing information to ensure accuracy and completeness.
  • This challenge becomes even more pronounced with complex or lengthy claims, adding to the workload and potential for errors.

Business Solution

  • The Claim Summarizer Tool automates the process of compiling key information from the Claim Notebook, enabling adjudicators to quickly understand the status and critical aspects of a claim.
  • The tool retrieves claim data from enterprise systems and processes it to provide a consolidated view of the claim in a single click, eliminating the need for manual cross-referencing.

Potential Benefits

  • Increased efficiency by reducing the time and effort required for claim analysis.
  • Improved decision-making accuracy through automation, minimizing human error.
  • Enhanced overall productivity, allowing adjudicators to focus on higher-value tasks.

AI Ecosystem Stack

  • Llama Index with Azure OpenAI GPT-4o.

Streamlining Underwriting Workflows with AI-Powered Solutions

Boosted efficiency by automating claim analysis, minimizing manual work and improving decision-making accuracy

Business Problem

  • Agents currently email Underwriters (UWs), who must manually extract information from emails, attachments, and forms to create accounts and submissions.
  • This process, involving around 270 emails weekly for 315 users, takes anywhere from 2 hours to 2 days per request.
  • The manual nature of this workflow is time-consuming, diverts underwriters from core tasks, and can result in missed opportunities due to bandwidth limitations.

Business Solution

  • Emails are routed directly to a dedicated mailbox, then stored in an S3 bucket.
  • Gen AI models extract relevant information from emails and attachments, automating account and submission creation.
  • The processing time per email was reduced from 4 hours to 15 minutes, with exception cases handled within this timeframe if additional details are required.
  • Standard emails are fully automated and queued for underwriter review, ensuring minimal delays.

Potential Benefits

  • Significant productivity gains by automating the processing of unstructured data.
  • A notable reduction in task completion time, freeing underwriters to focus on high-value activities.
  • Enhanced efficiency in handling large volumes of requests without compromising accuracy.

AI Ecosystem Stack

  • Python, Text-Embedding-Ada-002, Azure OpenAI Models, Azure Document Intelligence Service, Outlook Plugin, Azure AI Vision Service, MongoDB, Qdrant Vector DB, LlamaIndex

Boosting Migration to Achieve Efficiency for IT Team

Achieved a 25% faster time-to-market and improved code conversion efficiency from 20% to
70%

Business Problem

  • Transitioning from legacy systems to modern tech stacks, such as migrating Stored Procedures to REST
    APIs and Java to Spring Boot, to reduce capital (CAPEX) and operational (OPEX) expenses.
  • Converting COBOL documentation from legacy mainframes to CloudBase to further cut costs.
  • Automating the generation of test cases (e.g., pyUnit/jUnit) to improve engineering efficiency.

Business Solution

  • Applied prompt engineering techniques to optimize the code conversion process.
  • Used AWS Claude for enhanced engineering efficiency.
  • Split large code files (over 200K tokens) into manageable chunks for seamless processing.
  • Achieved significant improvements in code conversion, ranging from 20% to 70% efficiency depending on the use case, compared to manual methods.

Potential Benefits

  • Enhanced code conversion efficiency by up to 70%, reducing manual effort.
  • Accelerated time-to-market by 25%, enabling faster delivery of solutions.

AI Ecosystem Stack

  • AWS Bedrock, Claude Sonnet

‘Ask Anything Chatbot’ for Customers

Boosted productivity by 30-40%, doubled first-call resolution rates, enhanced CSAT scores, and cut training time by 50%

Business Problem

  • Customers often struggle to access key insurance information, such as coverage details, claim submission processes, and policy expiration dates.

Business Solution

  • Implemented an “Ask Anything” Q&A capability for intuitive information retrieval.
  • Enabled chat-style search across enterprise documents for seamless access.
  • Integrated the solution with customer and policy data, making it extensible to end-customers for self-service options.

Potential Benefits

  • Increased productivity by 30-40%, significantly boosting operational efficiency.
  • Doubled first-call resolution rates, ensuring faster issue resolution.
  • Enhanced customer satisfaction (CSAT) scores through improved service quality.
  • Reduced training time for employees by 50%.

AI Ecosystem Stack

  • Lang Chain, Adv RAG, GPT 4

Digital Transformation: Case Summarizer

20-30% gain in policy issuance time, 20-25% productivity increase, streamlined case resolution, faster issue resolutions

Business Problem

  • Tracking case details, such as record receipt times and reminder history, has been challenging.
  • Additionally, long policy issuance times and slow service issue resolutions have created inefficiencies.

Business Solution

  • Summarized case information and provided Q&A functionality for easier tracking.
  • Summarized new business and in-force requests.
  • Customized summary creation for documents, forms, and licenses.
  • Offered real-time assistance with next-best actions for resolving issues.

Potential Benefits

  • Reduced policy issuance times by 20-30%, significantly improving efficiency.
  • Increased overall productivity by 20-25%.
  • Streamlined case resolution processes, enabling quicker issue resolutions.

AI Ecosystem Stack

  • Lang Chain, Adv RAG, GPT 4

AI-driven Captioning and Semantic Matching for Enhanced Asset Retrieval

Streamlined captioning and image matching increased newsletter efficiency by 60-70%

Business Problem

  • The client was manually tagging images, videos, and documents based on result statements, which was both inefficient and prone to errors. Each time new result statements were added, the tags had to be updated manually.
  • The system also faced challenges in effectively matching relevant assets (like images and videos) to these statements, which led to poor content retrieval and irrelevant results.

Business Solution

    • The solution involved using AI to automate the tagging process. Instead of manual tagging, an AI-driven model was introduced to generate captions for images, videos, and documents automatically.

Key features of the solution include:

    • Image Captioning: AI-generated descriptions for images (up to 60 words).
    • Video Captioning: AI extracts descriptions from video frames, summarizes background audio, converts it to text, translates it to English, and combines it with visual descriptions.
    • Document Processing: Summarizes text from word documents to create concise descriptions.
    • NLP-based Matching: Uses a sentence similarity model to match assets with result statements based on their semantic meaning.

Potential Benefits

  • By automating the captioning and image matching processes, we significantly improved the efficiency of our newsletter production.
  • This improvement led to a 60-70% increase in overall efficiency.

AI Ecosystem Stack

  • GPT Vision-4 Preview, Multi-Modal RAG

Requirement Bot: Automated Product Filing for SMEs

Reduced product filing time across various coverages and attributes, achieving an accuracy rate of over 90%.

Business Problem

  • Business Analysts (BAs) and Subject Matter Experts (SMEs) were spending significant manual effort reviewing Product Filings, Rate files, and JIRA/ADO to create a “Product Definition” template. This process was time-consuming and inefficient.

Business Solution

  • We developed an Application Interface that takes a Product Filing as input and generates a Requirement Matrix as output. The client’s insurance policy documents were stored in AWS S3, with embeddings stored in a Vector DB and AWS Aurora.
  • We implemented Retrieval-Augmented Generation (RAG) to extract relevant information from these documents. By using advanced Prompt Engineering techniques (including prompt tuning and few-shot learning), as well as parallel/multi-preprocessing, we customized Large Language Models (LLMs) for optimal performance across different coverages.
  • Various LLMs were evaluated using a rouge score to select the best-performing model for this use case. The solution fully automates the process of generating the final Excel file, with an integrated UI.

Potential Benefits

  • This solution has significantly reduced the time required to produce product filings across multiple coverages and attributes.
  • Additionally, the accuracy rate has exceeded 90%, ensuring both greater efficiency and precision.

Partner Ecosystem

  • AWS Bedrock, Claude Sonnet, Claude V2.1, Jurassic 2 Ultra, Titan Embeddings G1 – Text

Accelerating Software Development using GitHub Copilot for Unit Testing

Significantly improved the development process by reducing Time-to-Market by 30% and lowering the defect leakage rate

Business Problem

  • User story linkages are managed in Rally, but when a feature changes or an enhancement request comes in, it’s essential to ensure the entire test suite is updated with all relevant linkages and scenarios that can be covered by unit tests.

Business Solution

  • Implemented unit testing coverage to shorten testing cycles and improve time-to-market.
  • Leveraged GitHub Copilot for automated unit testing.

Potential Benefits

  • This approach has reduced time-to-market by 30%, enabling faster product delivery.
  • It has also lowered defect leakage, ensuring higher quality and reliability in product releases.

AI Tools

  • GitHub Copilot

Centralizing Knowledge Database for Enhanced Self-Service

Centralized knowledge base and AI bot integration enhanced efficiency and empowered user autonomy

Business Problem

  • Disparate organizational data sources need to be integrated to create a central knowledge database for self-service.
    Client expressed concerns about IT being slow, less responsive, and lacking modern systems.

Business Solution

  • Developed a conversational AI bot and integrated it with MS Teams to provide a self-service knowledge database across the organization.
  • By implementing a self-service solution for password reset requests, we can achieve a 20-25% efficiency gain.
  • Currently, we handle 500-600 requests per month, with about 25% (125-150) being password resets. Self-service is expected to reduce this by approximately 40 tickets.
  • This improvement will also enhance the end-user experience, increasing our Customer Satisfaction (CSAT) score from 3/5 to 4/5.

Potential Benefits

  • By centralizing access to information through the knowledge base, we’ve significantly boosted efficiency and empowered users to operate more autonomously.
  • The integration of the chatbot has also improved the user experience, making it easier to find information quickly and accurately.

AI Ecosystem Stack

  • Python | Mongo DB | Chroma DB (Vector DB) | GPT 4.0 | Apache Airflow (Orchestration) | Kafka (Middleware)

Resources

Contact Us