AI-POWERED UX REDESIGN · 2026

From support tickets to self-service

A major villa construction platform in the UAE had a journey so unclear that 40% of active users filed a support ticket just to keep moving. They were not asking for help with edge cases; they were asking what to do next.

ROLE
Product Designer + Product Owner
DISCIPLINE
CX, UI/UX, AI delivery
STACK
Claude Code, Claude Opus 4.6, Figma MCP, Playwright MCP, FigJam, Jira
AT A GLANCE
40%
OF USERS FILED A TICKET TO CONTINUE
57
AUDIT ISSUES ACROSS 17 FLOWS
9
FLOWS REDESIGNED
165+
SCREENS DELIVERED
THE PROBLEM

“What's next?” The journey was so unclear that 40% of active users filed a support ticket just to keep moving. We ran raw user testing results and customer service data through Claude to quantify the pain.

  • 40%
    Unclear requirements18 of 45 respondents said service requirements were not clear to them.
  • 24%
    Upload restrictions11 of 45 could only upload in specific formats, with no guidance on supported types.
  • 9%
    Navigation difficultyOnly 4 self-reported it, but behavior data told a different story. Actual and perceived usability rarely match.
THE APPROACH

One pipeline, six stages, with human review between every step so quality compounded instead of resetting.

Production app + Figma → Heuristic audit → Stakeholder synthesis → Design system → Wireframes → Figma capture

Claude Code and the Figma MCP ran the pipeline. Each step generated the artefacts that gave later stages the best possible context.

THE ARTEFACTS

True UX is not just how it looks; it is how it works. The existing documentation captured neither, so I created the missing artefacts to get a strategic view of the product's current state.

  • 01
    The logic: PRDsFlow-by-flow product requirements, built with Claude and the Playwright MCP reading the live app.
  • 02
    The architecture: user flowsEnd-to-end task flows exposing exactly where journeys broke.
  • 03
    The enhancement: UX auditThe 57-issue Nielsen audit, structured as a living document in Jira.
PRDS IN ACTION

The focus was two things documentation almost never captures. Unrealized acceptance criteria: the microfeatures that make a flow feel finished but never get specced, like skeleton states while prices calculate on slow networks, a budget-warning modal with a financing link when costs exceed the wallet, and offline caching that syncs when the connection returns. And edge case definition: errors, timeouts, and empty states defined before any code is written.

Why it matters commercially: one PRD explicitly required a full VAT breakdown on booking confirmations. That single acceptance criterion reduces anxiety and prevents the exact category of “hidden fees” support ticket the redesign set out to kill. This rigor was encoded into custom Claude skills for PRD generation, so every PRD came out consistent and right-sized for its stakeholder.

Let's build
the next one.

Robert.t.marsch@gmail.com ← Back to all work