Manulife AI Research

AI research and applied research are embedded within our business and technology functions. We focus on practical applications that improve outcomes for our customers and operations.

Latest Updates

2026 A Momentum-Based Normalization Framework for Generating Profitable Analyst Sentiment Signals publication
2026 Understanding Multi-Structured Documents via LLMs publication
2025 Implementing Retrieval Augmented Generation Technique on Unstructured and Structured Data Sources in a Call Center of a Large Financial Institution publication
2024 Reactive to Preventive: Managing Fraud with Databricks OpenCV and GenAI talk
2024 Fin-ALICE: Artificial Linguistic Intelligence Causal Econometrics publication
2023 The Emotion Magnitude Effect: Navigating Market Dynamics Amidst Supply Chain Events publication
2023 Enhancing Financial Market Analysis and Prediction with Emotion Corpora and News Co-occurrence Network publication
2023 Using Social Media to Help Understand Patient-Reported Health Outcomes of Post-COVID-19 Condition: Natural Language Processing Approach publication

View all publications →

Embedded Domains

AI capabilities are developed within dedicated teams across the organization:

Advice

Customer and advisor self-service and advice solutions

Distribution

AI solutions for sales assistance and virtual coaching

AI for Operational Tasks

Eliminating or significantly reducing manual processing

Virtual Assistants

Conversational AI solutions that answer questions and execute tasks for employees, customers, and advisors

View all domains →

Responsible AI

Our approach to AI is grounded in responsible development and deployment practices:

- **Governance:** All production models undergo review by model risk management - **Three Lines of Defense:** Business units, risk management, and internal audit maintain independent oversight - **Transparency:** We prioritize interpretable models and explainable outputs where feasible - **Fairness:** Bias testing and monitoring are integrated into our model development lifecycle - **Privacy:** Data minimization and privacy-preserving techniques are core design principles

Open Source

We contribute to the broader ML community through open source projects:

models-eval-playbook

active

CLI-based playbook for evaluating LLMs with MLflow logging. Supports multiple providers including Azure OpenAI, Ollama, OpenRouter, and Alibaba Cloud. Enables testing a model under evaluation against a configurable judge model.

Applied AI Research ·

financialqa

active

Multi-structured financial document question answering using RAG. Extracts text, tables, and figures from financial report PDFs via Azure Document Intelligence, with LangChain-based chunking and Azure OpenAI embeddings. Associated with the ECIR 2026 paper "Understanding Multi-Structured Documents via LLMs."

University of Waterloo Partnership ·

View all projects →