Context · Strategy · The Product · Experience · Outcome · Reflection

NAB · Personal Banking · 2026

Growing people, not just building systems.

The AI platform was in place, but most staff did not know how to use it confidently. I designed and built the Banking Domain AI Academy: a persona-based learning programme that turns non-technical staff into confident AI operators across Product, Design, Change, and Risk. Self-paced, guided by an AI teacher agent, no calendar invites, ~2.5 hours to reach baseline capability.

In one minute. Self-paced AI academy for thousands of NAB staff. I designed persona tracks, wrote the curriculum, built the linear learning site, and shipped Teacher Candice, a Cursor agent that guides learners through lessons and scores exercises. PO/BA path live L0–L3.

AI Enablement Digital Learning Design Instructional Design Change Management Cursor Product Design Enterprise AI
My Role
Principal Product Designer · Academy architect & builder · Learning content writer
Stack
Eleventy · GitHub Pages · Cursor agents · .cursorrules
Organisation
NAB · Personal Banking
Scale
Thousands of cross-functional staff · 4 persona tracks
Status
PO/BA track live L0–L3 · expanding personas
Two-layer architecture for the AI Academy: Layer 1 is a browser-based learning site with role quiz and personalised dashboard. Layer 2 is Teacher Candice, a downloadable Cursor agent that delivers lessons and grades exercises.

The academy architecture: a browser-based learning site with a linear curriculum, paired with Teacher Candice, an AI agent students install inside Cursor. Same lessons, two surfaces.

The platform was ready. The people were not.

I had built the Context Lifecycle, a governed AI knowledge platform for the domain, and shipped the Agentic Document Generator, which lets POs and BAs draft requirements with AI against shared context. With executive mandate to scale across thousands of staff in the Banking Domain, the infrastructure worked. Adoption did not follow automatically.

Team members had wildly different levels of AI confidence. Some had never opened Cursor. Others were prompting daily but without guardrails. Technical roles could self-direct through existing engineering materials. Non-technical personas (Product, Design, Change, Risk) had no on-ramp that matched how they actually think and work. Without enablement, the system would stay impressive to executives and invisible to the people who needed it most.

The training is not just about tool adoption. It is about shifting how an entire domain across five disciplines thinks about knowledge sharing, cross-functional handoff, and AI-assisted work. The system is the infrastructure. The academy is the culture change. Both are required for either to work.

Working with the Enablement Manager and an Agile Coach, I designed a persona-based, level-based learning programme: the Banking Domain AI Academy. Not a one-off workshop. Not a mandatory calendar block. A self-paced platform with a linear curriculum, guided by the site and Teacher Candice, where people enter at their level and leave with real capability.

Strategy first. Screenshots second.

Before writing a line of curriculum or building the site, I documented the full enablement strategy: which personas to serve first, which mental models to teach, how levels map to the enterprise AI maturity ladder, and how a downloadable agent would make the learning stick inside the actual work environment.

Why four personas first

The bank defines eight delivery personas. I scoped the academy launch to four non-technical roles where the gap was widest and the multiplier effect highest:

01

Product (PO/BA) · Priority 1

Requirements, user stories, stakeholder presentations, backlog grooming. The Direct Report model: brief AI like a capable colleague.

02

Product Design · Priority 2

Research synthesis, UX copy, design briefs, accessibility reviews. The Art Direction model: style, tone, references, intent.

03

Change Management · Priority 3

Change documentation, stakeholder impact, adoption comms. The Stakeholder Brief model: audience, change, impact, action.

04

Risk · Priority 4

Risk assessment, advisory, change risk documentation. The Assessment Frame model: situation, dimensions, guardrails, output.

Engineering, Architecture, Security, and Solution Design were deferred. They can self-direct through existing engineering training and will join the academy as it matures.

Four persona prompting frameworks: Direct Report for Product, Art Direction for Design, Stakeholder Brief for Change, Assessment Frame for Risk.

Phase 1 persona strategy: four non-technical tracks, each with a prompting framework that matches how people already think. Familiar mental models, not a new language.

The five-level maturity ladder

L0 to L2 are shared foundations for every persona. L3 branches into role-specific tracks with real roadmap scenarios. L4 is multi-persona swarming: multiple agents under human direction. The levelling quiz places you at the right starting point. You do not have to begin at L0 if you are already past it.

Five maturity levels from L0 Unaware through L4 Specialist, with L0-L2 as shared essentials and L3 as persona breakouts.

The maturity ladder: shared essentials through L2 (~2.5 hours total), persona breakouts at L3, swarm workflows at L4. Baseline target: every team member reaches at least L2 Capable.

Teacher Candice: learn by installing an agent

The most important factor in AI adoption is not teaching people how to prompt. It is helping them experience what AI is capable of. Teacher Candice is implemented as a Cursor rules file students download and install. The act of installation is the first lesson: drop the file in, open Cursor, and the AI introduces itself, knows your role, and has a lesson plan ready.

Download → Install → Open Cursor → Learn interactively → Get scored → Record progress

Candice delivers lessons conversationally, grades exercises on a 0–5 rubric with specific feedback, remembers your name and level, and adjusts framing to where you are. One agent file scales to every learner without a human facilitator in the room.

Two layers. One curriculum.

Layer 1 is the interactive learning site: a linear web experience that walks learners through a short quiz, personalised dashboard, level track, scenario knowledge checks, and badge progression. No login. Works in any browser. Progress saved locally.

Layer 2 is Teacher Candice: a downloadable agent package students install in Cursor. Same curriculum, delivered conversationally inside the tool they will use for real work. PO/BA track is fully built from L0 through L3. Design, Change, and Risk tracks are in development.

Academy landing page mockup: role-customised learning, five maturity levels, and a four-step onboarding flow. (Portfolio approximation only.)
4 tracks
Persona-specific prompting models: Product, Design, Change, Risk.
5 levels
L0 Unaware through L4 Specialist, aligned to the enterprise AI maturity ladder.
~2.5 hrs
Essentials curriculum (L0–L2), self-paced across a few sessions. No calendar invites.

What success looks like

01

L2 Capable as the baseline

Every team member can use their role's prompting framework to produce structured, useful outputs. Levels 0 to 2 are Essential.

02

Guided path, no scheduling

A structured lesson sequence led by the site and Teacher Candice. No formal training sessions or mandatory calendar blocks. People learn when they have 15 minutes, not when a room is booked.

03

Role-appropriate confidence

Prompting AI feels like briefing a capable colleague, not writing a command. Each persona gets a framework that matches their existing mental model.

Four steps to start. Lessons that meet you in Cursor.

The onboarding flow takes about two minutes: pick your role, choose your starting level (or let the quiz guide you), land on a personalised dashboard, and begin. Lessons run 5 to 12 minutes each. Some include hands-on exercises with Teacher Candice inside Cursor, scored on the spot.

The learner journey: choose a role → personalised dashboard → install Candice in Cursor → complete lessons and earn badges. These three screens show the moments that matter most.

Step 1: role quiz tailors the learning path to how you actually work. (Portfolio approximation only.)
Installing Teacher Candice: the first agent lesson. Students download, install, and immediately experience what an agent is. (Portfolio approximation only.)
Level completion: badges for L0 and L1, all milestones checked off. Knowledge checks close each level before the next unlocks. (Portfolio approximation only.)

More in the flow

The personalised dashboard and interactive Cursor tour sit between onboarding and the Candice install lesson. Same curriculum, more context for how learners navigate the real tool.

Personalised dashboard with level track, milestones, and lesson progress. (Portfolio approximation only.)
Interactive Cursor tour: numbered pins walk learners through Agent mode, model selection, and the prompt box. (Portfolio approximation only.)

Learning outcomes are practical, not theoretical: use AI confidently and safely, produce real work outputs (briefs, stories, research syntheses), and build reusable prompt templates your team can share. Not practice exercises. Your actual work.

Enablement as product, not an event.

The academy turns platform investment into adoption. People who complete L2 can contribute to and draw from the domain's shared AI knowledge base. People who reach L3 produce role-specific outputs against real roadmap scenarios. The Context Lifecycle, ADG, and every downstream tool depend on this layer working.

Thousands
Of cross-functional staff across the Banking Domain the academy is designed to serve. PO/BA track live first; other personas rolling out.
PO/BA L0–L3
First persona track fully built and live: quiz, dashboard, lessons, Candice agent, knowledge checks, badges.
~2.5 hrs
Designed time to L2 baseline (L0–L2 essentials), self-paced. No mandatory sessions or calendar blocks.

What changed

Before After
AI capability siloed in individual experts Structured linear path from L0 to L2 for every persona, guided by the site and Candice
Non-technical roles had no AI on-ramp Persona-specific prompting frameworks matched to existing mental models
Training meant calendar blocks and slide decks Interactive site plus in-Cursor agent. Muscle memory in the real environment
Infrastructure ready before culture was Enablement programme running in parallel with platform scale

The hardest part of enterprise AI is not the model.

It is adoption. And adoption only happens when people experience AI doing useful work in their own context, at their own level, without needing permission or a technical champion in the room. The academy was my answer to the gap I identified in the Context Lifecycle case study: the system scaled faster than the change management. I would invest in enablement in parallel with the technical build, not sequentially.

What I am most proud of is treating enablement as a product problem with the same rigour as the platform: persona research, level design, agent architecture, interactive UX, and measurable outcomes. Teacher Candice is not a chatbot wrapper. She is a curriculum delivery system that happens to teach what an agent is by making you install one.

Four persona tracks. Five maturity levels. Two layers. One goal: every person across a domain of thousands can reach for AI instinctively, safely, and at the quality bar the organisation requires.

← Previous
Voice of Customer Dashboard
Next →
The Context Lifecycle