Snapshot
I led service design across three connected services on a province-wide registry platform, taking each from ambiguous policy to a build-ready backlog. AI ran through every phase of the work; a human checkpoint followed every AI output.
ServiceWho it servesThe problemWhat I delivered
Condo Registry Service (CRS)Condo corporations (board volunteers, managers); Consumer Services agentsA brand-new provincial fee had to be billed and collected from ~10,000 condo corporations, with no reliable unit count dataInterviews, 2 personas, service blueprints, end-to-end user flows, payment, correction and cheque workflows, cross-team integration
Registry Agent Network (RAN)200+ private registry offices; the public submitting documents; Land Titles examinersDocuments arrived by mail and were scanned only after registration, so examiners worked on paperScan-on-intake service model, readiness survey, current and future state blueprints, status maps, business requirements, draft epics
Land Titles (LTS) AdminLand Titles internal staff; Albertans submitting paper documentsInternal staff needed digital tools for paper submissions that had no home in the new platformRequirements translation, AI-assisted blueprint, 8 prioritized epics, dependency and readiness mapping, UX handoff
Headline outcomes
01Shipped the Tribunal Service Fee payment experience, from policy to production in 3 months.
02Reframed RAN intake so documents are scanned at the counter, letting examiners review digitally instead of waiting on mail.
03Turned LTS Admin's open-ended requirements into 8 sized epics sequenced against two release code freezes.
04Aligned 10+ cross-cluster teams (identity, finance, data, design system, print services) on shared workflows rather than one-off builds.
The challenge
A multi-product platform, legacy underneath, paper at the edges.
Alberta was moving decades-old land registry systems onto ARLO, a single online platform. Strip away the government context and the problem looks familiar to any scaling business: many teams who each own one piece of the customer journey.
Three things made it hard:
- ✦Policy moved faster than data. New legislation created a fee before anyone knew how many units each condo corporation had. Unit counts lived in scanned plan images, not in a database.
- ✦Physical and digital had to work together. Paper documents, cheques, scanners and mail rooms were not going away. The service had to treat them as first-class channels, not edge cases.
- ✦No team owned the whole journey. Identity, payments, reporting, printing, the design system and an outsourced vendor each held a piece. A good screen was worthless if the hand-off behind it broke.
My job was to make the whole service visible and make a complex process easy to use.
How I work
AI-augmented, human-validated.
I use AI to compress the slow parts of service design, so more of my time goes to judgment, facilitation and decisions. The rule I hold myself to: AI drafts, I verify against the evidence, and stakeholders confirm before anything moves downstream.
PhaseWhere AI accelerated the workWhere I stayed in charge
Plan researchDrafted interview guides and survey questions from project goals and policy documentsCut leading questions, matched language to each audience, tested with SMEs
SynthesizeClustered interview notes, surfaced patterns, drafted persona attributes and quotesTraced every insight back to a participant; turned interviews into insight themes designers can build for
MapTurned workshop stickies and process notes into first-draft blueprints and swimlanes; flagged likely inefficienciesWalked each blueprint with the people who do the work; corrected steps the model inferred
DefineTranslated meeting notes and emails into structured requirements and acceptance criteriaResolved conflicts between teams; decided what was suitable for MVP
PrioritizeDrafted epics and backlog items; compared effort, clarity and risk across workflowsSized with the team; sequenced against release dates and dependencies
CommunicateDrafted demo scripts, presenter notes, help content and executive summariesEdited for tone and accuracy; presented in person
Guardrails I kept
No personal or sensitive data went into tools outside approved government environments. Interview data was de-identified before synthesis.
Why it matters to your team
A repeatable operating model, not a one-off experiment. It shortens the time from raw research to a decision-ready artifact, while keeping the evidence trail a skeptical stakeholder will ask for.
01
Condo Registry Service
OutcomeA new self-serve service where condo corporations verify their unit count, request a correction and pay the Tribunal Service Fee online or by cheque — backed by an internal workflow for the agents who review them.
The brief. New legislation introduced a Tribunal Service Fee of $9 per unit per year to fund Alberta's new condominium tribunal. Roughly 10,000 condo corporations would owe it. There was no portal, no billing system and no trustworthy unit count.
1. Discover: who actually pays this fee?
I led interviews with condo board volunteers and internal Consumer Services agents. Most corporations are run by unpaid volunteers, often retirees, who carry the financial duties of a small business.
AI stepDrafted interview guides with Claude from the legislation and project goals, then tuned them with policy SMEs.
AI stepClustered de-identified notes into goals, behaviours, pain points, workarounds, mental models and patterns, tagging each sticky to a participant.
HumanKept a pattern only when it held across participants, and validated the analysis with the product owner and the Consumer Services lead.
interview analysis board — one cluster
What we learned
- ✦Board volunteers are experienced professionals. They want control and data, not hand-holding.
- ✦Trust in property managers is low. Any flow that routes payment through the manager meets resistance.
- ✦“Titled unit” versus “residential unit” confuses nearly everyone, and the fee is charged per titled unit. That is a dispute waiting to happen.
- ✦Online payment is acceptable, but cheques must stay. One participant tracked approvals with a physical stamp on every cheque.
- ✦Participants saw CRS as a future data hub on Alberta condos, not just a fee tool.
“The only concern is if a condo doesn't understand unit means titled unit, and then they try to dispute the amount.”
— Participant 4
From this I built two behaviour-based personas, plus an internal Consumer Services agent persona. AI drafted the persona cards from the synthesis; I rewrote them against the raw notes.
persona card
The Gatekeeper
Condo board volunteer
persona card
The Data-Driven Doer
Condo board volunteer
persona card
Consumer Services agent
Internal Land Titles staff
2. Map: the data problem behind the screen
Mapping the service blueprint exposed the real risk: the fee depends on unit counts the system could not reliably produce.
~14,000
active condo plans
71%
simple — one plan, one corporation
29%
redivisions or phased plans needing manual calculation
AI stepUsed Claude and Copilot to turn workshop notes into a three-stage blueprint (data collection, notice generation, notice mailout) across condo corporations, Land Titles, Print Services and Consumer Services.
AI stepExplored AI document extraction to read unit counts from scanned condo plan sheets, as an alternative to manual lookups.
HumanConfirmed with Land Titles staff that the legacy report could not count redivision units, which changed the design direction.
CRS service blueprint — data collection → notice generation → mailout
3. Design decisions
DecisionWhy
Verify before you payIf the data may be wrong, let the condo corp check it. Users see their unit count and can request a correction before paying.
One correction request per year, with a signed declarationKeeps agent workload predictable and gives an audit trail.
Keep the cheque channel, digitize behind itConsumer Services expected about 80% of corporations to pay by cheque. Finance uploads cheque data weekly so paid status shows in the portal.
Model multi-year balances up frontI mapped 8 status scenarios across 2026 and 2027 (owing, paid, in review). Rule: a payment covers the oldest year first.
Plain-language help on “titled units”Directly answers the top confusion from research.
AI stepGenerated edge-case scenarios for the multi-year logic, which I tested with finance and developers.
HumanRan decision sessions where finance, policy and product signed off each rule.
4. Deliver: end-to-end flows across two perspectives
- ✦Condo corporation: sign in, search corporation number, verify unit count, pay, receive receipt — or branch into a correction request.
- ✦Consumer Services agent: receive the request in a dashboard, review documents, approve or reject, trigger the recalculated fee and notice.
- ✦Revenue Management: record cheques, reconcile, issue refunds.
sign in → search → verify → pay flow
correction request flow
The Tribunal Service Fee shipped to production.
02
Registry Agent Network
OutcomeA scan-on-intake service model that lets 200+ local registry offices accept land title documents from the public and submit them straight into the digital examination queue.
The brief. Registry agents are private businesses licensed by the province; most Albertans know them as the place to renew a driver's licence. Lawyers file about 70% of land titles work electronically. Everyone else had to mail or drop off paper. Success means registry agents consistently submitting documents into the pending registration queue for Land Titles examiners.
1. Discover: the bottleneck was the scanner
I mapped the current state with Land Titles staff. Paper packages arrived by mail, were examined on paper, and were scanned only after registration, for retention. Examiners never saw a digital copy until the work was done.
AI stepFed the current-state blueprint to Claude and asked it to flag hand-offs, waits and duplicate steps. It highlighted the post-registration scan, double data entry and mail transit as the biggest delays.
HumanTested each flagged inefficiency with examiners and document reception staff. Two were real constraints (Legal requires Land Titles to keep the physical copy); the rest became design opportunities.
The reframe
Move scanning from the end of the process to the front counter.
Registry agents scan at intake, upload to ARLO, and mail the originals afterwards for retention. Examiners start on the digital copy while the paper is in transit, and scanned submissions get priority over mailed ones.
BEFORE — mail → examine on paper → scan for retention
AFTER — scan at counter → examine digitally → mail originals
The paper still travels, but it is no longer on the critical path.
2. Research: can 200+ offices actually scan?
The reframe only works if offices have the equipment and skills. Communication with agents was limited, so I designed a readiness survey instead of relying on assumptions.
- ✦Section 1, all staff: office volume, scanner hardware, software, file formats, scanning habits, comfort with digital files, pain points.
- ✦Section 2, supervisors: compliance with the provincial equipment and security standards they already operate under.
AI stepDrafted question banks and answer options with Claude, then checked them for leading or double-barrelled questions.
HumanReviewed with registry operations and the agent associations, and cut the survey to about 10 to 15 minutes.
3. Define: blueprint, statuses and requirements
- ✦Future-state service blueprint across five stages: intake and triage, create document, collect payment, scan and create the request, mail originals. Frontstage, backstage and support systems on one page.
- ✦Status map of a submission from Draft to Not submitted, Completed, Received and Archived, so agents, examiners and devs used the same words.
- ✦Business requirements translated from workshops on point-of-sale integration, refunds, rejections, scan quality checks and document tracking.
- ✦Draft backlog: 15 epics across 6 work buckets, from scanning infrastructure to reporting.
Draft→Not submitted→Completed→Received→Archived
AI stepConverted workshop stickies into structured requirements and first-draft epics in Copilot.
HumanSurfaced the open questions the model could not answer — who checks scan quality, what happens if the paper never arrives. These became the agenda for business decision sessions.
RAN future-state service blueprint — 5 stages
MVP scope: the top 8 document types by volume.
03
Land Titles Admin
OutcomeA sequenced, build-ready backlog for the internal staff who handle mailed-in paper submissions, with every dependency named and an MVP that fit two release windows.
The brief. The new platform was built for digital submitters. But paper still arrives every day from the public and from registry agents. Internal Land Titles staff needed to create submissions on a customer's behalf, attach scans, fix bad scans, print notices and archive originals. Requirements were scattered across teams and partly unknown.
1. Translate: from scattered asks to one picture
AI stepGathered stakeholder notes, emails and workshop boards, then used Claude and Copilot to structure them into requirements grouped by workflow, and to draft a six-stage blueprint: intake, search and create, upload scans, submit to examiner, replace scans, printing.
HumanWalked the blueprint with document reception, examiners and the scanning vendor, and corrected where the model had assumed today's process.
“How does the system know a submission started on paper?”
— The question that unlocked the design
Instead of adding a new field, the team agreed to infer it from who created the request — a registry agent account, a law firm account, an internal staff account, or a member of the public.
2. Prioritize: what can design safely work on next?
I ran a dependency and design-readiness session. Each workflow was rated on how well we understood it, how complex it was, how many questions were open, and which teams it depended on. The guiding question: if we only had capacity for two workflows, which would create the most learning and reuse?
EpicSizePlan
Create a submission for a paper customerMMVP, first
Replace scans attached to a submissionMMVP, second
Archiving physical originalsMMVP
Mark registry agent paper as receivedM to LMVP
Deficiency workflowLBacklog
Resubmitted paper submissionsLBacklog
Post-registration queueLBacklog
Notice printingM to LBacklog
Two items moved out of scope as business processes rather than product features. Three queue workflows (deficiency, post-registration, notice printing) were grouped so they could reuse the examiners' existing queue patterns instead of new builds.
AI stepDrafted epics, user stories and acceptance criteria, and a cross-team matrix of which question belonged to which of 8 teams.
HumanSized with the team and sequenced against two code freezes.
dependency + design-readiness grid
3. Hand off: make the UX designer fast
I translated each requirement into a short design brief with the user, the trigger, the rules and the open questions, and linked blueprints into the stories. The design estimate came out at 1 to 2 sprints of immersion plus 2 to 3 sprints of design.
Orchestrating across the platform
The hardest work wasn't on any screen.
It was getting identity, finance, data, print, design system and vendor teams to build to the same service, on the same timeline. The blueprint was the contract between teams: each box on it named an owner, a hand-off and a release.
cross-team orchestration — 8 platform teams around one blueprint
User access
Defined new roles for Consumer Services agents and Revenue Management, and an interim workaround until the finance role shipped.
Finance
Designed how online payments flow from the platform's payment service into CRS, and how cheques get recorded so paid status stays accurate.
Data & reporting
Set MVP reports (who has paid, outstanding balances, payment method) and post-MVP ones (corrections, condo statistics).
Print Services
Mapped the print-and-mail process for notices, from data file to envelope, so the notice template fit their production line.
Design system
Assessed whether CRS needed custom components; kept MVP to standard patterns to save rework.
Platform
Helped scope the CRS dev environment, data extract and integration pattern.
AI stepDrafted agendas, summarized cross-team meetings into decisions and owners, and kept dependency lists current.
HumanFacilitated the sessions and chased the decisions with stakeholders and leadership.
Impact
What changed.
Launched a new revenue-collecting service for about 10,000 business customers, with self-serve correction built in.
Redesigned an intake model across 200+ partner locations so the back office works digitally from day one.
Gave an internal operations team a sequenced MVP instead of an open-ended wish list.
What I learned
01Fix the data before the UI.The best CRS decision, verify before you pay, came from mapping the data problem, not the screens.
02Keep the analog channel, digitize behind it.Cheques and paper were not legacy to kill. They were real customer preferences to support well.
03AI is a first draft, never a finding.It saved me hours on synthesis and mapping. It also confidently invented process steps, which is why every output went back to the people who do the work.
What this means for your team
The same patterns show up in fintech, insurance, health, logistics and any platform company: new pricing rolled out to existing customers, partner or franchise networks, physical-to-digital hand-offs, and many internal teams owning one journey. I bring a working AI-augmented practice that gets teams from ambiguity to a prioritized backlog faster — with the evidence trail leaders need to trust the result.
