Modernizing Data and AI at New York Life

Introduction to New York Life

• New York Life is the nation's largest Mutual Life Insurance Company, beholden to its policy owners rather than Wall Street.
• The company has a 179-year history of resilience, thriving through economic upturns and downturns.
• New York Life is embracing technology, data, and AI to deliver exceptional experiences for its agents and policy owners.

Data and AI Strategy at New York Life

• The strategy is to activate the power of New York Life's data through enterprise AI and data products to deliver meaningful business impact.
• This strategy is built upon four pillars:
  1. Scale the data foundational and technology to build enterprise data products.
  2. Elevate data intelligence through AI and analytics.
  3. Transform the culture to be guided by data-driven decisions.
  4. Empower proactive and personalized experiences for agents and policy owners.

Transforming the Data Platform

• New York Life's previous data platform was a Hadoop on-prem data lake environment with various challenges, such as reaching end-of-life, limited vendor support, and the need for improved performance and new user personas.
• The new approach is to build a lakehouse architecture using S3 as the data lake and Redshift as the data warehouse.
• This journey involved a mixture of rehosting and re-engineering components, with the help of different teams within the organization:
  • Enterprise Cloud Platform Team: Responsible for setting up the AWS environment, approving services, and establishing guardrails.
  • SREs (Site Reliability Engineers): Responsible for the platform-level infrastructure, including Terraform coding and IAM principal development.
  • Organizational Engineering Teams: Responsible for building data pipelines, ingestion, ETL, and domain models.
  • Consumers and Personas: Analytical engineers, data scientists, MLOps, and others consuming the data.

Architectural Best Practices and Principles

• All data pipeline activities land data in a HIPAA-compliant account to handle PII and PHI securely.
• Different roles have different privileges, with clear segregation of read and write access.
• AWS Glue is used for data ingestion, with a framework that can handle different data sources and patterns.
• The data lake maintains full type-2 history, and Iceberg tables are used for performance improvements.
• The data governance account is responsible for tagging documents for use by appropriate models.
• The consumption side includes a Redshift data warehouse, with separate clusters for ETL/ELT and exploratory data analysis.
• The Producer 2.0 operational data store is integrated with Redshift for unified analytics.

Enabling Gen with Data

• The data lake structure is used to centralize high-quality and reliable data for data scientists and Gen applications.
• A generic framework is built to ingest unstructured data from various sources, improving data access and scalability.
• Data governance and compliance are ensured through tagging and tracking data lineage for ML and Gen projects.

MLOps Architecture and Principles

• Key principles guiding the MLOps architecture include scalability, automation, reproducibility, modularity, security, and democratization.
• The development architecture leverages SageMaker Studio for exploratory data analysis, with MLOps engineers building templates and dockerizing experiments.
• The operational architecture is modular, with separate teams responsible for the UI, AI/Gen solutions, and infrastructure monitoring.
• Additional components, such as OpenSearch for vector databases and DynamoDB for storing model metrics and data, were introduced to support Gen use cases.

Improving the Claims Manager Experience

• The business partners at New York Life identified a pain point in the claims management process, where custom letters were being manually generated based on complex medical records.
• The goal was to standardize the letter generation, improve the customer experience, and reduce legal risk.
• By leveraging large language models, the time to generate these letters was reduced from hours to seconds, while ensuring consistency, readability, and appropriate tone.
• The experience-based acceleration approach was crucial in breaking down the problem, testing different solutions, and delivering an excellent outcome.

Lessons Learned

• Design for scale, but build iteratively, prioritizing architectural reviews and following AWS best practices.
• Focus on building a solid data foundation, and bring the right models to your data.
• Embrace a fail-fast mindset, and be flexible in adapting the approach as needed.
• Break down silos within the organization, fostering collaboration between business and technology stakeholders.
• Maintain a clear vision, but be open to changing the course as required.

Conclusion

The modernization journey at New York Life demonstrates the power of embracing technology, data, and AI to drive innovation and deliver exceptional experiences for both internal and external customers. The key takeaways include the importance of architectural best practices, the value of experience-based acceleration, and the ongoing need to adapt and learn in the face of ever-changing technological landscapes.