Widlight
An AI-led sustainability web app built for Accenture's Season of Impact showcase
Timeline
January 8 – April 9, 2026 (3 months)
Teams
1 PM, 2 product designers, 1 brand designer, 2 front-end developers, 1 back-end developer
Client
Accenture — Gowtham (Business Architect Lead), Julie (Principal Director, Customer Transformation), Julian (Director, Corporate Citizenship & Social Innovation)
Role
Led design direction and owned the plan page end-to-end. Conducted user research and SME sessions. Authored the JTBD scoping list, AI prompt structure, and developer handoff documentation.
Research
In-person usability testing (25 participants across 2 rounds), SME consultations, user interviews, field research
Outcome
Full product delivered to Accenture for soft launch. Consistently positive client feedback throughout the engagement
THE BRIEF
A gardening app on paper. A corporate impact story in practice
Accenture's existing prototype was a questionnaire that collected yard information and spat out a garden plan. An AI chatbot sat loosely on the side.
Early conversations with Gowtham and Julie surfaced what the project actually was: Wildlight wasn't just a gardening app — it was Accenture's corporate impact story made tangible, built for internal use and planned for their Season of Impact event, with downstream goals in municipal adoption and insurance partnerships. That changed what mattered. The plan page wasn't an output screen anymore. It was the most important surface in the product.
DISCOVERY
The first few weeks were spent listening — to our SME, users, and the gardening centres
SME SESSIONS
Elliott, native pollinator ecologist
Valuable insights into the problem space that we are not familar with. Provided insights into the mental models of gardenners.
FIELD RESEARCH
6 interviews with intermediate gardeners
Site assessment is where people give up — researching plants, conditions, and timing across multiple sessions.
WHAT THE NUMBERS SAID
January nursery visit
No native pollinators in bloom — a product insight about seasonality that reshaped the plan page phase timing.
Three things came back across all three tracks: site assessment is the hardest part, gardeners think in temporal phases (prep / plant / maintain), and most ecological metrics can't be honestly estimated by AI.
THE REFRAME
The vibe-coded version was the wrong starting point
The AI was an afterthought — sitting beside a survey that was doing all the heavy lifting at exactly the part of the gardening process users found hardest: planning and preparation.
The single-page output didn't match the temporal phases users actually worked in. The AI was decorative, not functional.
This was the moment AI stopped being a chatbot-on-the-side and became the core value proposition — exactly aligned with what Accenture needed the showcase to demonstrate.
TRANSLATING THE DECISION ACROSS THE TEAM
A design decision is only as good as the team's ability to execute it
Once the structure was agreed on, the next several weeks were less about Figma and more about making sure SME, PM, devs, and brand designer were building the same thing. I authored three documents that did most of the heavy lifting — but the IA work happened on paper first.
JTBD SCOPING LIST
With the PM, I helped crystallize our feature deliverable list from the lens of users.
Each feature mapped to a specific user job, prioritised against the three-phase structure. Turned the project from a feature pile into a roadmap.
AI PROMPT STRUCTURE DOCUMENT
The AI needed to generate specific, accurate site assessments.
I wrote the prompt framework, vetted by Elliott, that mapped user inputs to scientifically honest outputs. This was as much design work as anything I did in Figma.
DESIGN-DEV COMMUNICATION
Whiteboard sessions, messages, and full plan page specification with 3 devs
Phase logic, AI behaviour, progress tracking, card structures, shopping list filtering. Three developers aligned through constant communication, not a single handoff doc.
Someone needed to hold all five perspectives at once — SME, client, PM, dev, user — and over time, that ended up being me. The documents I authored along the way became as much a product of the project as the screens.
ITERATIONS & TESTING
Three rounds. The first two-thirds was navigation and discovery. The final third was where craft caught up
ITERATION 1 — FEB 15
Happy path usability test, 5 participants.
AI-led site assessment flow (buggy). Plan page was a dev placeholder. Concept test — the right method since users had no actual garden to execute against.
ITERATION 2 — MAR 12
Open exploration usability test, 20 participants.
Improved site assessment, freely explorable. Plan page implemented from my design handoff. Mapped all feedback onto an impact/effort matrix.
ITERATION 3 — FINAL
Impact/effort matrix of the 2nd usability test drove all decision.
Major visual refinement of the plan page — unified type system, color system, spacing, shadow, micro-interactions. End of project timeline; no further user testing.
REFLECTION
Most of the design work happened outside Figma.
I spent maybe a third of this project in Figma. The other two-thirds was SME sessions, user interviews, field visits, prompt structures, JTBD frameworks, dev handoffs, and stakeholder alignment. That balance wasn't a quirk of this project — it's what design leadership on a complex enterprise engagement actually looks like.
I carry this forward as a default: on a project with this many inputs and stakeholders, the design lead's job is to hold every perspective simultaneously and make decisions that honor all of them. The screens are the visible output, but the alignment work is the actual design.
If I could continue: the site assessment flow didn't receive the same visual refinement the plan page did. I designed five improvement screens addressing the usability and visual gaps — worth a next iteration.
What I'd measure if continued:
Provincial plant count accuracy. Self-reported data drifts — users forget to update, abandon mid-season, log plants that don't survive. If the numbers inflate, the whole impact story collapses.
Phase transition rate. The three-phase structure is the core bet. If users complete Prep and never return, the checkpoints aren't checkpoints — just a fancier single-use output.
AI suggestion follow-through. "Completed the assessment" isn't the same as "followed the recommendations." Goal completion per phase would tell us which.
The honest gap: everything measured at delivery is pre-behavior. None of it touches whether a single garden actually got planted — the one outcome the impact story rests on.
The client took the team out to dinner at their office near the end of the project. That meant more than any testing metric — it was the signal that we'd understood what they actually needed, not just what they'd asked for.
