Skip to content

« all case studies

Chapter 6 · 2025-2026

Expert Insights: AI Platform Leadership + Responsible Rollout

Leading Vanguard's AI-driven portfolio analysis product from architecture through pilot to launch, reaching 110,000 advisors on day one.

AI · GenAI · platform · governance · responsible AI

Context

Vanguard needed to scale portfolio specialist-quality expertise beyond 1:1 engagements through a real-time AI-driven product that gave advisors reliable insights for complex portfolio construction. The product had to improve advisor productivity and client outcomes at scale.

The constraint: this couldn’t be a GenAI demo. It had to be credible as a scalable platform with a real rollout path, pilot → broader distribution, and with risk and governance treated as first-class requirements from day one. Regulated finance does not forgive ambiguity here.

My role

I operated as delivery lead for Expert Insights, the primary voice shaping both the product narrative and the delivery plan. Introduced the team and product at the June Board Gallery Walk. Authored the core storytelling and architecture artifacts. Identified as Delivery Lead in BCAAS, April 2026.

What I did

Standardized the insight contract, so the AI output could scale

Defined a repeatable structure for what an “expert insight” is, making the model’s output understandable, testable, and consistent:

  • What, identify the observation.
  • So what, explain why it matters.
  • Now what, suggest actions.

This turns probabilistic LLM output into a stable product pattern. Teams can evaluate it, design around it, and deploy it consistently. It also became the lens used in eval, every insight has the same three jobs, so quality has a definition.

Set platform-level architecture direction

In ITSS FAS AI Expert Insights, the architecture for the pilot is framed at three layers:

Not one-off prompting. A strategic enabler with a roadmap spanning production pilot through broader scaling phases.

Drove governance and risk readiness proactively

Didn’t treat governance as an afterthought. Elevated it as a dependency and made a specific ask:

We need active ITSS support on the AI risk/governance to support our rollout plan.

The same governance posture got reinforced in direct internal messaging with leadership stakeholders, including specific content asks to make the governance dependency visible at the right level.

Led the responsible-AI storytelling for the board

In the June Board Gallery Walk planning, the team aligned on a board-ready narrative explicitly focused on responsible AI, not just the use case. I contributed the framing that the board-level “hook” is that the team is using AI to judge AI output (judges + confidence bands) and that this is not “running a calculation through ChatGPT”, it’s working inside Vanguard’s guardrails.

The narrative had to land for an audience that might “know nothing about AI.” Before/after quality framing, plain-language confidence framing, careful depth control. The work was as much about translation as it was about technology.

Enabled broad alignment by sharing a single source of truth

Actively distributed the architecture pages, the ITSS deck, and the supporting video into the right channels so decisions could move at the speed of the program rather than the speed of the documentation hunt.

Constraints I navigated

Governance and risk approval as a gating dependency. Elevated the need explicitly and pulled it forward into the plan so it never became the blocker.

Executive communication complexity. Board may know nothing about AI. The narrative had to do real work, simplified visuals, before/after, careful depth, without losing the technical credibility that made it trustworthy in the first place.

Quality proof for probabilistic systems. “How good is good enough” isn’t implicit; it had to be shown. Baseline vs. experiment quality improvement, judge accuracy framing, confidence bands.

Outcome

  • Live in production with pilot. Oct 2025.
  • Pilot with Sales. Kickoff Jan 2026.
  • Direct to advisors and beyond. Q1 2026.
  • Launch day reach: 110,000 financial advisors.
  • Product framed as enhancing, not replacing, advisor judgment, which is what made the rollout durable.

The launch was covered in Barron’s, The Daily Upside, InvestmentNews, Citywire, ThinkAdvisor, Financial Advisor Magazine, PLANADVISER, and WealthManagement.com. See the press section for the full list.

How earlier chapters showed up here

The threads from previous chapters are the load-bearing structure of this one:

  • Pattern standardization → scalable experience. The “what / so what / now what” structure does the same job for AI output that AXS component schemas do for UI, it makes the surface predictable.
  • Platform-first design. Foundational LLM + scalable API + prompt-embedded methodology. Same instinct as framework-agnostic AXS: design for the substrate, not the moment.
  • Governance as product requirement. Same posture as AXS governance (Chapter 5): boundaries are not friction; they are what make the system durable.
  • Executive-ready storytelling. The same translation work that turned AXS into a business platform turned Expert Insights into something a board could endorse.

Why it matters

This is the chapter where the platform discipline of the earlier years compounds. AXS was the substrate. Expert Insights is the proof that the substrate scales into AI.