The Challenge
A large enterprise manufacturing company was migrating from a 15-year-old Oracle ERP system to a modern cloud-native SaaS platform. With 150,000+ active users across 40+ countries, the stakes were enormous: any UX misstep would cascade into millions in lost productivity, employee frustration, and massive support costs.
I was brought in as Lead UX Designer to own the complete user experience layer — from initial discovery research through to shipped design system and final UI. The scope covered 12 primary modules: Inventory, Procurement, Finance, HR, Manufacturing, Distribution, Reporting, Admin, and 4 industry-specific extensions.
What We Were Solving
The legacy Oracle system had accumulated 15 years of inconsistencies, workarounds, and technical debt — all of it baked into the user interface. A baseline audit surfaced critical issues that were costing the business real money every day.
Understanding 150,000 Users
Before touching a wireframe, I spent 6 weeks deep in discovery. The research scope was ambitious: we needed to understand not just current pain points, but the mental models, daily workflows, and success definitions of wildly different user groups — from warehouse floor workers to senior accountants to C-suite executives.
Key Research Findings
Three critical insights shaped the entire design strategy:
Finding 1 — The "7 window problem": Warehouse supervisors routinely had 7+ browser tabs open simultaneously to complete a single purchase order — because information was siloed across disconnected modules. This drove our "unified workspace" design principle.
Finding 2 — Mobile is not optional: 42% of our users worked primarily on mobile or tablet in the field, yet the existing system was completely unusable on anything smaller than a 1440px desktop. Mobile-first became a hard design requirement, not a nice-to-have.
Finding 3 — Training doesn't scale: The 3-week training programme existed because the UI was so unintuitive — not because the work itself was complex. Fixing the UX could eliminate 80% of the training requirement.
How We Designed It
The project followed a modified double-diamond process adapted for enterprise ERP complexity, with tight iteration loops and continuous stakeholder validation built into every sprint.
Crucially, testing wasn't a final gate — it was woven throughout. We ran usability sessions after every major milestone, including paper prototype tests early on that let us validate information architecture before a single pixel was designed in Figma.
From Chaos to Structure
The information architecture phase was the most complex of the entire project. We had to rationalize 12 separate module structures into a single coherent navigation system — while respecting the existing mental models 150,000 users had built over 15 years.
We ran card sorting sessions with 24 participants across 4 user roles to validate our proposed navigation taxonomy before committing to the information architecture. The results challenged several assumptions from stakeholders — particularly around how finance users conceptually group "reporting" features vs. how engineering assumed they would.
The final IA solution used a dual-level navigation model: a persistent global shell for cross-module navigation, and contextual secondary navigation that adapted based on which module the user was in — giving power users fast access to deep features without overwhelming new users.
The Visual Language
With the structure validated, I moved into high-fidelity UI design. The visual language needed to feel professional, trustworthy, and calm — enterprise users spend 8+ hours daily inside this system, so excessive visual noise or aggressive colour use would cause fatigue and errors.
Key design decisions that defined the visual system:
Colour as signal, not decoration: Colour is reserved exclusively for status communication (green = good, amber = attention, red = critical) and primary actions. Background and surface colours are deliberately neutral, keeping users focused on data.
Typography hierarchy that communicates density: Large module headings create clear orientation, while data values use monospaced numbers for alignment and scannability — critical when reviewing lists of inventory quantities or financial amounts.
Building for 8 Product Teams
The single most impactful deliverable of this engagement wasn't a specific UI screen — it was the enterprise design system built to support it. With 8 product teams working across 12 modules, consistency without a shared component library would have been impossible.
Measurable Business Impact
The project shipped in phases over 14 months, with each module launched to a controlled user group before full rollout. Post-launch metrics were tracked across a 6-month period.
Beyond the quantitative metrics, the qualitative impact was equally significant. Post-launch user interviews revealed a fundamental shift in how users felt about the system — from "a tool I'm forced to use" to "something that actually helps me do my job better."
The design system became a foundational platform investment — within 6 months of launch, 3 additional product teams outside the original scope had adopted it for their own module work, extending the ROI well beyond the initial engagement.
What I'd Do Differently
Every large project leaves lessons. The three most significant ones from this engagement:
Start the design system earlier. We began building the system in Month 3, but should have started in Month 1. The early screens we designed before the system existed created technical debt and inconsistencies that took two sprint cycles to clean up.
More continuous testing, less big-bang testing. We ran major usability test rounds at months 4, 8, and 12. In retrospect, fortnightly lightweight testing with 3-4 users would have surfaced issues faster with less wasted design effort between rounds.
Involve engineering from day one. When engineers joined the process at wireframe stage (Month 4) rather than at pixel-ready handoff, the quality of design-to-development translation improved dramatically. Cross-functional pairing from the very start would have saved significant rework.