Summary
Host Aparna Sinha sits down with Vinh Tran, Vice President of Data & AI Platforms and RBC Fellow at Royal Bank of Canada, to unpack the $700M to $1B AI value commitment RBC's CEO made to the street at the 2025 investor day, and how Vinh's team is delivering it by the end of 2027. They get into which use cases move first (back-office and call-center insight, not direct-to-customer), how RBC is measuring ROI across nine flagship projects and five lines of business, why Vinh favors build over buy in a regulated industry, the platform pattern behind RBC Assist and Self-Serve Agents, how the Lumina data platform and a data-product mindset feed the agents, the control plane of guardrails and judges that keeps agentic AI reliable, and why 'AI for All', 96,000 employees and not just 8,000 developers, is the real transformation.
Chapters
Chapters
Why Listen
If you're building AI inside a regulated enterprise, this is a rare, specific look at what it actually takes to move from pilots to production at the scale of Canada's largest bank. Vinh walks through the platform, the standards (MCP, OpenTelemetry), the control plane, and the workforce story, and shares his mental model on Building vs. Buying: 'I really feel that we have to have the courage to build. We're not going to write every line of code. We're going to use the tools. We're going to use the cloud. We're going to use the community.'
Key Takeaways
- RBC's CEO committed $700M to $1B of AI-driven value at the 2025 investor day, to be delivered by the end of 2027 across nine flagship projects spanning all five lines of business plus technology and operations
- The real transformation isn't the 8,000 developers and data scientists who already used AI, it's getting AI into the hands of all 96,000 staff: mortgage brokers, branch managers, account managers, engineers
- RBC is starting with back-office and call-center insight use cases (not direct-to-customer) while the technology and the organization build the muscle; fully autonomous customer-facing AI is 'possibly, one day'
- In a regulated industry, 'build' beats 'buy' not because it's cheaper but because it gives you control of your destiny: your policies, your patches, your compliance posture, plus a stronger engineering culture
- Three pillars scale AI across the enterprise: a foundation platform (gateway-fronted approved models), universal access and literacy, and reusable scaffolding so builders don't start from an empty folder
- Reliability for agentic workflows comes from a control plane: LLM-as-judge guardrails, PII and content filters, a gated MCP server registry, human-in-the-loop on material changes, and continuous evaluation from production telemetry, not once-a-week batch evals
- Sometimes you don't need an agent at all, deterministic workflows should stay as deterministic code; agents are for complex, changing, human-interaction-heavy work like intent classification and summarization
- Engineers become builders: coding tools are giving 10x productivity, but 'just because Vin can vibe code a system, we should not put it in production and run our banking systems against that'
- Don't fall in love with today's tools or paradigms, 'we're in chapter one or two of a six-chapter book.' Build AI literacy, get started, iterate
Insights
In the second episode of Enterprise Aligned AI, host Aparna Sinha sits down with Vinh Tran, Vice President of Data & AI Platforms and RBC Fellow at Royal Bank of Canada, ranked #1 in Canada and #3 globally for AI maturity on the 2025 Evident AI Index, to get into what an enterprise-wide AI transformation actually looks like from the inside.
The Scope of RBC’s $1B AI Transformation
At RBC’s 2025 investor day, the CEO committed $700M to $1B of AI-driven value, to be delivered by the end of 2027. Vinh joined the Borealis team shortly after to lead the platforms that make that possible. Nine flagship projects, with hard KPIs tracked by finance, span all five lines of business plus technology and operations. But the number Vinh keeps coming back to isn’t the dollar figure, it’s 96,000. Historically AI at the bank served 6 to 8 thousand developers and data scientists. The transformation lifts everyone with a computer: mortgage brokers, branch managers, account managers, wealth advisors.
Measuring ROI of AI Across Business Units
When the CEO goes to the street and commits a number, that number needs to be measurable. RBC tracks ROI across those nine projects with clear KPIs, split between revenue uplift and cost reduction, with finance and accounting teams doing the quantification. A concrete example: call-center productivity. AI drafts the post-call notes and summaries that agents used to write by hand, shaving meaningful time off every call and materially reducing the cognitive load on staff who take dozens or hundreds of calls a day.
Build vs. Buy: Why Build in Regulated Enterprises
Vinh is unapologetic about building. Three things changed the math: open-source ecosystems are now production-grade, coding tools give 10x to 100x engineering leverage, and the cloud democratizes services that used to be the domain of large vendors. But cost savings aren’t the main reason. In a regulated industry, building gives you control of your destiny, your policies, your patches, your compliance posture, and it builds a culture of engineering excellence that you can’t buy.
Self-Service Agents and Platform Standards
RBC’s platform scales AI through three pillars: a foundation with pre-approved models behind an LLM gateway (with model risk, security, and bias reviews done up-front), universal access and literacy, and reusable scaffolding with LangGraph references, MCP tool registration, and knowledge connectors, so builders aren’t starting from an empty folder. Standards matter too: RBC is standardizing on Model Context Protocol (MCP), agent-to-agent, and OpenTelemetry to stay pluggable in a market that mostly isn’t standardizing.
Preparing Data for AI: The Lumina Platform
Data is the fuel. Vinh credits his predecessor for starting the Lumina data platform five to six years ago, data lake, warehouse, semantic layers, metadata, and for moving the bank to a data-product mindset with curated data hubs (retail credit, mortgage) instead of everyone copying data everywhere. For other enterprises: the blocker isn’t the tool. It’s the endless architecture meetings. Start, classify, iterate, and use AI itself to generate metadata and tags.
RBC’s Control Plane for Reliable Agentic AI
Reliability in banking workflows comes from a control plane: LLM-as-judge guardrails where one model reviews another’s intent classification, PII and sensitive-content filters, an MCP gateway that enforces human-in-the-loop for material actions, and an AI architecture review that asks the most important question first: do you actually need an agent for this? Deterministic workflows should stay deterministic code. Evaluation is continuous, pulled from production telemetry, not a weekly batch of 1,000 prompts.
RBC Assist and ‘AI for All’
What started in May 2024 as an internal chatbot has evolved into a no-code agentic productivity platform now used across the bank every day, product managers synthesize Jira boards, staff analyze uploaded files alongside MCP-served enterprise data, and anyone can build an agent without writing code. The workforce story is the one Vinh is proudest of: RBC serves 17 million clients with 96,000 staff today and wants to serve 25 million with the same team, which means making every employee more effective, not cutting.
Hiring the Next Generation of Builders
Engineers become builders: fewer ceremonies, faster delivery, but also a much harder peer-review problem when AI is writing thousands of lines of code. “Just because Vin can vibe code a system, we should not put that in production and run our banking systems against that.” The next generation of RBC engineers needs AI fluency, orchestration instinct, and the discipline to make sure the code the AI writes is the right code, with the right tests, for the right use case.
Advice for Enterprise AI Leaders
“We’re in chapter one or two of a six-chapter book.” Don’t fall in love with today’s tools or paradigms. Get started, build the AI literacy, pivot with the industry.
Speaker
Vinh Tran
Vice President, Data & AI Platforms @ Royal Bank of Canada
Vinh Tran is the Vice President of Data & AI Platforms at Royal Bank of Canada and an RBC Fellow at RBC Borealis. He leads the transformation of Canada's largest bank into an AI-powered enterprise. With nearly 10 years architecting cloud, data, and AI platforms across major Canadian banks, Vinh has built enterprise-scale infrastructure on Azure, AWS, and GCP serving thousands of developers and millions of customers. His work spans from pioneering multi-cloud Kubernetes environments to now spearheading generative AI, autonomous agentic systems, and enterprise data ecosystems, all while maintaining the rigorous security and compliance standards required in financial services.
