## Overview Canada Life, one of Canada's leading financial services companies, undertook a comprehensive contact center transformation journey that represents a significant LLMOps implementation at scale. Serving 14 million customers (approximately one in three Canadians) with insurance, wealth management, and health benefit solutions, Canada Life faced critical operational challenges in their contact center operations and embarked on a multi-year modernization initiative centered around Amazon Connect and AI capabilities. The presentation was delivered by Jenny Dabout (VP Contact Center) and Arpita Bhattacharya (AVP Contact Center), highlighting the close collaboration between business and technology leadership. The transformation addresses a fundamental tension in modern contact centers: while industry predictions suggest dramatic automation potential (Gartner predicting 80% autonomous resolution by 2029), customer behavior data reveals that even Gen Z customers (71%) and baby boomers (94%) still prefer phone calls for complex issues. Canada Life's approach balances automation with human touch, focusing on channel diversification rather than pure call deflection. ## Initial Problem Statement and Context Before modernization, Canada Life's contact center faced severe operational friction points that necessitated transformation rather than incremental improvement. The average speed to answer reached 5 minutes in specialty queues during busy seasons, with wait times extending to 30-40 minutes. The routing infrastructure was complex and inefficient, driving up transfer rates and lacking personalization capabilities. Perhaps most critically, the organization had little to no self-service options, forcing customers to use phone calls for even simple transactions like address changes. The pre-modernization ecosystem was characterized as "siloed, manual, and linear" with hundreds of 1-800 numbers. The IVR system provided only basic triage for phone and email. Every customer required manual authentication. Data fragmentation across systems drove up costs, created inefficiencies, and resulted in inconsistent customer experiences. When customers escalated from web-based self-service to agent assistance, no contextual information accompanied the escalation, forcing customers to restart their journey entirely—a classic symptom of disjointed systems that many contact centers struggle with. ## Technical Architecture and LLMOps Foundation The transformation established a modular, reusable architecture designed to accelerate innovation for both AI and non-AI capabilities. At its core, the technology blueprint consists of several layers: **Backbone Layer**: Amazon Connect serves as the foundational contact center platform, supported by Salesforce for CRM capabilities and Calibrio for workforce optimization. This provides the unified infrastructure for scalability, reliability, and workforce management. **Intelligence Layer**: The core AI capabilities leverage AWS services including LLM models accessed through Amazon Bedrock, Amazon Lex for conversational interfaces, and AWS Lambda for serverless compute orchestration. This layer delivers the intelligent, personalized information processing needed for customer experiences. **Integration Layer**: Knowledge bases, API integrations to backend systems, SharePoint integrations, and comprehensive monitoring and metrics infrastructure enable the 360-degree customer view. Data integration across these systems is critical to maintaining and sustaining the architecture. This architecture embodies Amazon Connect's agentic AI framework, which consists of four core components: understanding customer intent across all channels, reasoning to determine next best actions based on business rules and history, taking action such as processing refunds or updating accounts, and memory to ensure every interaction builds on previous conversations. These components are powered by knowledge layers connecting to enterprise systems, omnichannel communication capabilities (voice, chat, email, SMS, social), and orchestration layers combining structured workflows with conversational interfaces. ## Migration Execution and Methodology The migration was executed with remarkable speed and discipline. Canada Life migrated 21 business units in 7 months with 100% uptime (zero downtime), converted 150 toll numbers, completed 14 major tech deployments, collaborated with 38 teams, and impacted over 2000 agents with a unified agent desktop. This pace was achieved through what the presenters call their "secret sauce" or "Mac sauce"—a strategic decision to create a technology blueprint with out-of-the-box capabilities that every business unit would receive, avoiding customization unless it directly impacted day-to-day work or caused non-productive time. This blueprint approach enabled rapid execution by requiring only one sign-off with risk partners rather than separate business cases for each of the 21 business units. The methodology emphasized several critical success factors: **Business-Technology Partnership**: The VP of Contact Center (business) and AVP of Contact Center (technology) operated as "twins in a box" rather than "two in a box," focusing on one unified mission with faster decision-making (literally in minutes) and clear prioritization frameworks for handling competing demands across business units. **Process and Planning**: Detailed planning extended beyond the migration itself to include production validation processes and post-production SWAT team support structures. The iterative approach was likened to "building a house brick by brick," ensuring each component was solid before proceeding. **Testing and Training Integration**: Testing and training teams participated from the requirements phase through production releases and post-production support, creating critical feedback loops. This continuous involvement enabled rapid issue identification and resolution. **Measurement Framework**: The "measure it to manage it" philosophy drove continuous improvement, with comprehensive metrics and monitoring informing decision-making throughout the journey. ## AI Capabilities in Production Canada Life's AI implementation in production spans multiple capabilities, each addressing specific operational challenges: **Call Summarization**: The first AI capability deployed, call summarization proved transformative by taking call transcripts and generating 3-4 sentence summaries attached to customer cases. This reduced after-call work significantly, sometimes to zero, and enabled seamless context transfer when customers called back. The capability contributed to a 47-second reduction in average handle time per call and, when measured across just two months of steady-state AI operation, resulted in a 25-second reduction in after-call work specifically. **Automated Authentication**: Implemented to secure interactions while maintaining seamlessness, automated authentication handled 37% of authentication processes, removing friction from both customer and agent experiences while maintaining security and compliance requirements critical for financial services. **AI-Powered Chatbot ("Cali")**: The most recent major deployment, launched just prior to the presentation, is an AI assistant chatbot named Cali operating within authenticated spaces. Cali achieved remarkable results immediately: 83% consistent containment rate (meaning 83% of interactions were fully resolved by the chatbot without escalation to human agents), deflection of 1900 calls per week, and 80% positive customer feedback. This represents "intelligent engagement" rather than simple automation, freeing agent capacity for complex interactions while maintaining customer satisfaction. **Voice-to-Text and IVR Enhancements**: Comprehensive IVR review and optimization shortened prompts and improved routing accuracy, contributing to 2% transfer rate reduction. The implementation of 101 automated contact flows across all business units replaced manual, business-unit-specific flows, delivering consistency and speed while reducing complexity by 88%. **Proficiency-Based Routing**: This capability enabled an 83% reduction in routing profiles, dramatically decreasing operational costs and maintenance effort while improving first-contact resolution through better skills-based matching. ## LLMOps Operational Practices Several aspects of Canada Life's implementation reveal mature LLMOps practices: **CI/CD for Contact Flows**: The organization built continuous integration and continuous deployment pipelines specifically for contact center operations. IVR changes can be deployed in minutes, while major deployments complete in hours—critical capabilities given the seasonal nature of contact center volumes and the need to respond quickly to business events. **Unified AI Pricing Model**: By adopting Amazon Connect's AI bundle pricing, Canada Life eliminated complex vendor management and unpredictable costs. The bundle includes unlimited consumption of self-service capabilities (web-based chat, voice interactions), agent assistance capabilities (real-time suggested responses), post-contact summarization, and automated performance evaluations—all tied to channel usage rather than AI consumption, simplifying financial planning and removing barriers to AI adoption. **Performance Monitoring and Evaluation**: Automated performance evaluation enables 100% interaction analysis rather than the traditional 3-5% sampling. This comprehensive visibility revealed significant gaps between perceived and actual performance. The presenters cited an example of another organization that believed agents delivered or promised travel quotations in 30% of calls based on sampling, only to discover through 100% automated evaluation that the actual rate was closer to 10%, enabling targeted training interventions. **Omnichannel Enablement**: In 2025, Canada Life launched chat capabilities across all contact centers, with plans to expand AI assist across every team and product group in 2026. The focus is on diversifying customer interaction models from purely linear phone-based service to enabling customers to reach out on their own terms through self-service IVR, self-service chat, click-to-call, and enhanced email capabilities. **Knowledge Management and Agent Assist**: Current knowledge fragmentation across procedures, SharePoint sites, and duplicated resources represents a pain point the organization is addressing through agent assist capabilities that will provide knowledge at agents' fingertips in a unified interface, leveraging LLM capabilities to surface relevant information contextually. ## Business Impact and Outcomes The outcomes from Canada Life's LLMOps implementation substantially exceeded initial business case projections, representing what the presenters characterized as "leapfrogging" their transformation expectations: **Customer Experience Metrics**: - 94% reduction in wait time - 92% reduction in average speed to answer (now consistently 18 seconds across the entire year) - 2-4% reduction in abandonment rate depending on business unit - 12% increase in voice of customer survey scores for larger business units with ~500 headcount - Every attribute on voice of customer surveys improved in 2025 **Operational Efficiency**: - 10% reduction in average handle time overall - 47-second reduction in AHT per call through compounding capabilities - 60% reduction in complexity and cost through unification from three applications to one - 88% reduction in cycle time for deployments - 83% reduction in routing profiles - Over 90% operational cost savings through out-of-box features and automated exchange management **Financial Impact**: - $7.5 million in savings in first half of 2025 alone - $0.5 million cost savings from just two months of steady-state AI operation across limited business units - Projected $43 million in cost savings over five years - Projected reduction of 200,000 call volume - Projected 1.1-minute reduction in average handle time per call (future state) **Workforce Optimization**: - Reduced onboarding requirements by over 100 headcount in 2025 despite hiring 400+ agents annually - 4% improvement in shrinkage by converting non-productive time to productive activities like coaching and training - 8% attrition rate (significantly below industry averages) - Best-in-class employee engagement scores, nearly 5% higher than national average - 2% transfer rate reduction through improved routing and IVR optimization ## Change Management and Adoption Strategy Canada Life's approach to change management reveals sophisticated understanding of human factors in technology transformation. Rather than treating change as purely operational, the leadership viewed it as fundamentally emotional. Their framework centered on three principles: **Clarity**: Building a bold narrative explaining why the change was happening, how it would empower agents, and what customer experience improvements would result. Transparency reduced uncertainty and built trust throughout the organization. **Capability**: Moving beyond generic training to create role-based, scenario-driven training that reflected agents' daily lives. An adoption and readiness team supported agents from day zero through post-production, ensuring not just operational readiness but sustained competence. **Confidence**: Incorporating agents as early adopters and champions from the requirements phase onward, rather than merely during business testing. This participatory approach made agents feel heard and created genuine stakeholders in the transformation's success. This empathy-driven approach, treating agents as partners in the process rather than subjects of mandated change, contributed significantly to the high employee engagement scores and low attrition rates achieved. ## Future Roadmap and Agentic AI Evolution Canada Life's 2025 focus centered on enriching employee and customer experiences through omnichannel capabilities across four themes: seamless customer experience, agent empowerment, operational excellence, and omnichannel enablement. The 2025 goal was articulated as enabling "AI powered vision for any client, any channel, any conversation"—fundamentally diversifying the linear phone-based model to give customers flexibility to engage on their own terms. The 2026 roadmap includes several advanced capabilities: **Conversational AI**: Evolution of the Cali chatbot from its current capabilities to fully conversational interactions, representing a shift from transactional automation to natural dialogue. **Agent Assist at Scale**: Expanding knowledge management capabilities across all business units to address current fragmentation and provide contextual, AI-powered information retrieval at agent fingertips. **Next Best Action**: Implementing predictive capabilities to guide both automated systems and human agents toward optimal customer outcomes. **Custom Summarization**: Moving beyond generic call summarization to business-specific summary formats tailored to different product lines and use cases. **Fraud Detection**: Implementing AI-powered fraud detection capabilities within the contact center workflow to enhance security while maintaining customer experience. **Predictive Analytics**: Leveraging Amazon Connect's predictive analytics capabilities to anticipate customer needs and proactively optimize operations. The presenters expressed particular excitement about the convergence of conversational AI (represented by Cali becoming conversational), predictive analytics, and data analytics—viewing this combination as having transformative potential to fundamentally reshape customer service. ## Critical Assessment and LLMOps Considerations While the results Canada Life presents are impressive, several considerations merit attention when evaluating this case study: **Vendor Lock-in and Platform Dependence**: The deep integration with Amazon Connect and AWS services (Bedrock, Lex, Lambda) creates significant platform dependence. While this enabled rapid deployment and unified capabilities, organizations should consider multi-cloud strategies and vendor negotiating leverage over time. **Claimed vs. Realized Benefits**: The presentation occurred at AWS re:Invent, a vendor conference, and was delivered in partnership with AWS representatives. The $43 million five-year projection is forward-looking and should be viewed as aspirational rather than realized. The $7.5 million first-half 2025 savings represents actual results but lacks detailed attribution methodology. **Generalizability**: Canada Life's success depended heavily on strong business-technology partnership, significant organizational commitment, and substantial resources to migrate 21 business units with dedicated teams. Smaller organizations or those with less mature operational capabilities may find this model challenging to replicate. **AI Limitations**: While 83% chatbot containment is impressive, it also means 17% of automated interactions still require human escalation. The presentation doesn't detail failure modes, customer frustration with automation, or scenarios where AI performs poorly. The emphasis on maintaining human agents acknowledges that full automation remains distant despite industry hype. **Change Management Complexity**: The successful adoption relied on exceptional change management with early adopter programs, comprehensive training, post-production support teams, and empathetic leadership. This represents significant investment beyond pure technology costs that organizations must budget for. **Privacy and Compliance**: As a financial services company handling sensitive customer information, Canada Life operates under strict regulatory requirements. The presentation doesn't detail how AI implementations maintain compliance, handle data privacy, or manage audit trails—critical considerations for regulated industries. **Model Observability**: While the architecture includes monitoring and metrics, the presentation lacks detail on LLM-specific observability: how are model predictions monitored, how is drift detected, how are hallucinations prevented or caught, and what feedback loops ensure model quality over time? Despite these considerations, Canada Life's implementation represents a sophisticated, production-scale LLMOps deployment that achieved meaningful business results through disciplined execution, strong partnerships, and focus on both customer and employee experiences. The emphasis on measurement, iterative improvement, and balanced automation reflects mature operational AI practices.