Sage Earth

Refining a complex AI workflow and unblocking delivery.

Parachuting into an in-flight enterprise project to refine, restructure, and deliver a highly complex AI-assisted data categorisation feature in just 3 months.

Scroll to explore

The Problem

An in-flight feature stalled by UX friction.

I was brought in on a 3-month contract to assist the existing Design Lead with a critical, complex backlog item: an AI-driven transaction categorisation feature. The initial designs were struggling to cross the finish line. User testing revealed that the interface was actively confusing users. They didn't understand the difference between AI-sorted data and manual entries, the tab architecture ("To-do" vs "In Progress") was ambiguous, and the core "Edit Category" workflow was causing high friction. The challenge was to bring a fresh perspective, untangle the UX, and get the feature ready for development without blowing up the timeline.

The Approach

Tactical UI restructuring and user flow optimization.

I tackled the feature piece by piece, focusing on micro-interactions and structural clarity.

Component Redesign: I completely reworked the "Applied Category" pills, using visual cues to clearly differentiate between AI-suggested, partially categorized, and fully categorized states.
Flow Overhaul: I redesigned the "Edit Category" sidebar experience from the ground up, ensuring users could effortlessly override the AI at any stage of the journey.
System Feedback: To build trust with the AI, I integrated clear feedback loops, including AI explainer pop-ups, toast notifications upon saves, and robust filtering. Finally, I refined the top-level tab architecture to clearly reflect the user's actual mental model of the task.

The Impact

From backlog to production-ready.

Within the short 3-month engagement, we turned a stalled, confusing feature into a highly usable, intuitive workflow. By focusing heavily on component states, system feedback, and clear user flows, we resolved the friction points identified in early testing.

Key Wins:

Successfully unblocked a critical roadmap item for the Lead Designer.
Transformed a confusing AI interaction into a transparent, user-controlled process.
Delivered a robust set of UI rules, edge cases (like restricted users), and scalable components ready for engineering handoff.