Lean Insight & AXN

The context

Lean Business Services builds AI-powered enterprise tools for the Saudi market. When I joined, two products were in active development simultaneously: Lean Insight — a business analytics SaaS — and AXN, a medical coding platform powered by NLP to automate ICD and CPT coding for healthcare providers.

Both products had engineering momentum but no coherent design language, no design system, and a messy handoff process that was creating friction and rework on every sprint.

The problem

Running UX for two AI products at once presented a specific challenge: the products were architecturally unrelated but needed to feel like they came from the same company. And the AI-heavy nature of both meant the UX had to make opaque, probabilistic outputs feel trustworthy and actionable to non-technical users.

Concretely:

• Lean Insight: dashboards with complex data relationships that users couldn't interpret
• AXN: NLP suggestions that clinicians didn't trust because there was no transparency into how codes were generated
• Both: design and engineering weren't speaking the same language, causing 3–4 revision cycles per feature

Discovery

I started with stakeholder interviews across both product teams to map the specific pain points. For Lean Insight, the core issue was information hierarchy — dashboards were showing everything at equal visual weight with no clear path to insight. For AXN, the trust gap was the blocker: clinicians needed to see the reasoning behind a code suggestion, not just the suggestion itself.

I also audited the existing Figma files — there were 12 disconnected component libraries across the two products, with no shared tokens and inconsistent naming.

Design system first

Before touching any product screens, I built a shared design system. Single token library, shared component set for common patterns (tables, modals, inputs, status indicators), and a naming convention that mapped directly to the engineering component names. This was the foundational work that unlocked everything else.

The system reduced the number of Figma libraries from 12 to 3 (one shared, one per product for unique patterns) and gave engineering a stable, predictable handoff structure.

Lean Insight — making data legible

The redesign focused on progressive disclosure: executive summary at the top, drill-down on demand. I introduced a consistent "card → chart → raw data" pattern so users always knew how deep they were and how to get back. Key decisions were surfaced as callouts rather than buried in charts.

I also established a clear visual hierarchy: one primary KPI per card, supporting data in a secondary tier, with colour used only to encode state (positive/neutral/negative), not decoration.

AXN — earning trust from clinicians

The NLP coding interface needed to answer a specific question every clinician had: "Why is it suggesting this code?" I designed an explanation panel that surfaced the source text snippets that drove each suggestion, with a confidence indicator and a one-click accept/reject flow.

This transparency reduced the hesitation to accept suggestions and opened the door to training feedback: rejected codes now fed back into the model pipeline as labelled negatives.

What I learned

Running two AI product design workstreams simultaneously is tractable if you invest in shared infrastructure first. The design system wasn't overhead — it was the thing that made parallel work coherent. Without it, we'd have shipped two products that looked like they came from different companies.

For AI-powered features specifically: transparency is the UX. Users don't need to understand the model, but they do need enough signal to decide whether to trust it. That signal has to be designed in from the start, not added as an afterthought.