AI WORKFLOW & EXPERIENCE
From Copilot to agentic AI. Transforming the business.
Turning a government-wide AI mandate into a validated, sequenced portfolio of use cases — owned by the branches who'd run them, not imposed on them from above.
PROBLEM
A top-down AI mandate is easy to declare and hard to ground.
FCAC, like most public-sector regulators, is being asked to do more with less; rising regulatory complexity, legacy systems, fragmented workflows, and growing public expectations for fast, fair, accessible financial information. AI was named directly in FCAC's 2026–2029 Strategic Plan as a way to close that gap.
But by 2025, AI maturity inside the agency was still individual-level productivity: Microsoft Copilot rolled out agency-wide, well used, but not yet touching how entire workflows ran.
Moving from personal productivity to workflow-level, agentic AI is a different kind of problem. It isn't a technology rollout. It's translating a mandate into a portfolio of use cases grounded in what employees actually do, validated by the branches who'd own them, and feasible enough to build without creating new risk.
ROLE
Lead, Entreprise AI Strategy & Delivery
I designed and led the discovery-to-delivery process for FCAC's agentic AI program. I ran the workshops that surfaced opportunities from staff, built the prioritization and feasibility methodology used to rank them, and sequenced a roadmap now in active execution.
TEAM & SCALE
10
Core team — me, 1 UX, 2 developers, 1 BA, 5 external consultants
35+
Employees across PAB, RPE, SEB & HRB engaged in initial discovery
5
Branches taken through structured, bilingual validation processes.
3 tier
Delivery model balancing speed against governance and risk
A program run solo at the design and strategy level, working directly with branch executives, business leads, and a cross-functional operating-model team to take the work from workshop to roadmap to active build.
_edited.jpg)
The portfolio as a system:
~200 raw opportunities, surfaced from the people doing the work — narrowed by a defensible, repeatable method into a sequenced set the business helped build.
Context & approach
01
Starting from the work, not the technology
Working with external consultants for discovery only, we grounded ran workshops across four branches, mapping the pain points in their workflows to potential AI solutions.
To make AI tangible for non-technical staff without steering them toward a specific tool, we used a curated deck of plain-language AI solution patterns — intelligent assistants, automated reporting, risk scoring — as a bridge between the pain points they'd just mapped and concrete opportunities they could name themselves. Each opportunity got a first-pass score on feasibility and impact, producing a shortlist instead of a wish list.
This work had to be run in a day. It got people talking and excited about AI, which was the first step. But the challenge with bringing external partners in is always the limitation in their contextual understand. When the consultants left, there was a slight disconnect between the use cases they had identified and the root causes of the underlying problems. It was up to me to dig deeper and hone in on what our employees were really struggling with.
Service Journey workshop — ~40 employees across four branches mapping where the friction already lives, before any talk of tools.
DIAGRAM · REFINEMENT
~200
Raw Ideas
DISCOVERY
Candidates
VALIDATION
Validated Use Cases
EXECUTION
18
5
Assessed for feasibility, mapped for priority, sequenced by development logic.
02
Validating ideas with the people who'd own them
Workshop output is a starting hypothesis and a number of the initial ideas weren't yet grounded in operational reality or clearly enough articulated by the business to act on. So I designed and led a second round of structured validation sessions, branch by branch, run bilingually in English and French. Along with the 4 branches we'd already engaged, I also led sessions with our Legal branch to identify overlooked possibilities.
Each session followed a consistent facilitation structure. I built lightweight UX research into the pre-work itself; a short validation survey and a "day-in-the-work" suggested diary study so that participants could identify repetitive, duplicative, or time-consuming work ahead of time, rather than relying on recall when in the room.
In workshops we dug deep into the problem, being ruthless about what issues were pervasive and codified enough to solve.
Every session followed the same six beats:
01
Problem statement
02
Root cause
03
Frequency & volume
04
AI Potential
05
Expected Impact
06
Visioning Exercise
5 branches taken through validation process:
Insights, Policy, Financial Literacy Branch
Supervision & Enforcement Branch
Legal Branch
Human Resources Branch
Communications Branch
03
A two-axis methodology that could hold up to scrutiny
With a regulator's risk tolerance in mind, I built prioritization on two independent axes — priority and feasibility — so that an exciting idea couldn't advance just because it was exciting.
Priority weighed frequency, employees affected, time per process, potential time savings, data-quality improvements, and strategic alignment. Feasibility used a two-step approach: a hard gating checklist first, then a weighted readiness score that scored in multiple areas across 4 conceptual pillars. This data allowed us to plot every potential use case analytically, turning a long backlog into a defensible, sequenced roadmap leadership could interrogate rather than take on faith.
STEP 1 - HARD GATE
Collapsible sections
Any single failure stops the use case here — full stop.
✓ Is the process digital end-to-end?
✓ Does usable data already exist?
✓ Can a human-oversight step be inserted?
✓ Does the risk fit FCAC's tolerance?
STEP 2 - WEIGHTED READINESS SCORE
Only survivors of the gate get scored across four foundations, each broken into defined sub-criteria.
Technical & Data Readiness
Is the data and tech actually there to build on?
Process readiness & stability
Is the workflow stable enough to automate?
Security & privacy risk
What's exposed, and how is it controlled?
Operational capacity
Can the branch absorb and run it?
04
Sustaining momentum, through open communication
Branches don't trust a prioritization process that swallows their ideas silently, so I built transparency into governance from the start. Every branch received a roadmap update showing what advanced, what stayed in the backlog, and why, with an explicit commitment that any use case proving feasible would be brought forward as capacity allowed.
That discipline produced a sequenced, three-batch roadmap. Strong ideas that failed the gate, like a writing-style agent, were documented and held in backlog rather than quietly dropped.
THE ROADMAP: FROM SIMPLE TO COMPLEX
Scaling Value
Higher-complexity, higher-value triage work where the impact justified the build.
-
Supervisory document review
-
AI-assisted complaint handling
-
Mandatory-report screening
Pattern Setting
The strongest feasibility scores — chosen to establish reusable delivery patterns.
-
CIC inquiry tagging
-
CIC insight generation
-
CIC response drafting
-
ATIP for employees
-
Legal agreements reviewer
Navigating Complexity
Carefully scoped explorations of ambiguous problem spaces.
-
Compliance profiling
-
Stakeholder segmentation agent
1
2
3
Sequenced by feasibility first, then value. The easiest wins go first and set patterns to be reused for harder, higher-value cases further down the line.
the design decision that mattered most
UX as the operating model, not a phase
The feasibility gate requires a clear human-approval step before any agent can act. The principles I helped establish push teams to ask a UX-research question before an engineering one: can the underlying data or process simply be improved upstream, rather than reaching for AI to patch around it downstream?
The most important decision wasn't any single use case; it was making human-centered practice the default for how every future use case gets built.
05
A delivery model that balances speed against governance
The three-tier model lets the team move fast in low-risk environments while reserving full deployment for use cases that have proven governance, security, and adoption readiness. I built reusable intake artifacts — a use-case scorecard, a quick risk assessment, a release-evaluation framework — so every future use case gets assessed against the same explicit, repeatable bar.
1
Sandbox
Move fast in a low-risk environment. Prove the concept before anything touches a real workflow.
2
Controlled Pilot
Limited, monitored use with real data and a human in the loop. Testing governance, not just function.
3
Entreprise Deployment
Reserved for use cases with proven governance, security, and adoption readiness. Full rollout, earned.
SPEED
GOVERNANCE & RISK
METHODS & TOOLS
Service journey mapping
Workshop facilitation (EN/FR)
Roadmap design
Solution-pattern card sets
Risk-tiered rollout
Weighted scoring & feasibility-gating
Transparent backlog communication
Human-in-the-loop design principles
Two-axis prioritization (priority × feasibility)
Diary-study research (“day-in-the-work”)
RESULT
A validated AI portfolio the business helped build — and now trusts.
A repeatable, human-centered model for identifying use cases with impact, with the business as a partner and risk front of mind. More than just solutions; a holistic approach to delivery.
01
~200 raw ideas, surfaced directly from the people doing the work, narrowed through a defensible two-axis methodology to 18 candidates and 5 funded first-execution starters.
02
Branch-validated coverage across the agency, each with priorities confirmed by its own business owners, not assigned from above.
03
Active execution on five use cases, including three CIC-focused agents (inquiry tagging, insight generation, response drafting) and a carefully scoped Legal Provisions Agent.
04
A documented operating model formalizing the three-tier delivery approach and the framework for choosing between low-code (Copilot) and custom-built (Foundry) solutions.
05
A program written into FCAC's formal performance accountability structure — delivering high-value agentic AI use cases aligned to enterprise architecture, on an ongoing basis.
Active execution is underway on five use cases; and the methodology is being used to triage emergent cases, outliving any single use case and putting people at the heart of AI execution.