The brief said developer portal.

InfoTrax provides back-end operations for direct-sales enterprises on a legacy full-stack system, with every deployment customised. FlexCloud was the strategic bet: one SaaS platform, every service as a subscription API. Phase I was the developer portal. Phase II was the commissions engine.

The third position reorganised everything. If internal teams were the first users, this wasn't a client-facing portal. It was a system internal and external users would both live in, with very different permissions.

In practice: a developer managing five integrations needed the one scoped to their test environment and had no way to filter by environment, owner or status. They kept a bookmark for each environment. That was the interface.

No direct-sales company had done this, so we looked sideways at Shopify, Box and Slack, and built a side-by-side IA comparison between Box's console and FlexCloud's structure. It made the gaps arguable, not just assertable: permission scoping, environment isolation and developer-specific views were all missing.

Thirteen sessions across thirteen roles, then six weeks of synthesis. Each session walked participants through a deliberately hypothetical blueprint, then through the legacy services they actually used. Putting a wrong version in front of people surfaces what nobody thinks to mention.

The tier structure is what made the permission model fall out naturally later. Participants also kept trying to do things the blueprint had no place for: switching organisations, isolating work to an environment, seeing only what their scope allowed. The blueprint assumed a portal. Their corrections assumed a platform. That gap was the finding.

Recommendations were prioritised as must, maybe and ignorable, which turned a list of good ideas into a roadmap the client could act on.

The existing build treated FlexCloud as a flat set of pages. It couldn't handle internal users spanning several organisations, or environments gated by scope. I defined the hierarchy before drawing anything. Once it was settled, the IA, navigation, permission model and environment switcher resolved almost mechanically.

I explored six foundation layouts, each with annotated trade-offs: where organisation and environment switching lived, how navigation collapsed, how far sub-menus reached, and how visual weight signalled depth of scope.

Every workflow was flow-charted before wireframing, including edge cases, error states and micro-interactions. That discipline caught the read-only gate: letting users edit directly without first seeing a summary created accidental-modification risk in live environments. It came from a flow diagram, not a wireframe, and usability testing later surfaced the same gap independently.

InfoTrax had colour values and little else, so I built the system alongside the product. The palette was rebuilt as a semantic set from the brand primary and tested against WCAG AA (4.5:1) and AAA (7:1), because enterprise clients required it contractually. A custom icon library on a 24px grid with a 2px stroke, and full component variant sets, each with error and empty states. Sketch, low-fi and high-fi interactive prototypes; nothing skipped a rung.

Moderated testing with real InfoTrax roles: software developers, the implementations team, and product and R&D leadership. One of us facilitated; I observed and took findings back into the designs. We measured time to complete, number of attempts, and behavioural cues like hesitation and backtracking, over three iteration rounds before handoff.