VIOLET and MAGENTA each produced a numerical prediction of fertility outcomes — the result of a computer-vision model trained on thousands of IVF cycles. That prediction had real clinical value. But the reports reading it out led with comparative language: where a patient's result ranked against a reference population, at a moment when ranking was the last thing they needed.

The downstream signal was hard to ignore. Clinicians were softening, narrating around, or quietly setting aside the report rather than handing it to a patient. A tool built to build confidence in the science was instead creating a moment of dread — and the workaround was the clinician's voice, not the design.

I started with a structured audit of both reports: layout, language, information hierarchy, and the sequences clinicians were actually using them in. Two failure modes emerged. The first was comparative framing: leading with a percentile positioned patients against a benchmark rather than toward an outcome. The second was information hierarchy: the single number that mattered most — an absolute live-birth probability — was buried in the detail rows.

Both had the same root: the reports were organised around what the AI knew, not around what a clinician needed to say. Before removing anything, I mapped what was working quietly — the things users never filed tickets about: a cumulative multi-cycle view patients had come to rely on across appointments, and the VIOLET/MAGENTA colour signal that gave specialists an at-a-glance read. The redesign would protect both.

I redesigned VIOLET and MAGENTA together for the first time — as one report system with shared information architecture. The lead changed from a percentile to a plain-language, absolute outcome: what a patient's result means for their specific treatment path, not where it ranks. The structure that followed built toward the counselling conversation, not the raw data.

Tone was a live design surface. In a regulated clinical product, the words are as much the interface as the layout — so every sentence went through the same clinical review cycle as the visual structure. What emerged was a report a clinician could hand across the table. The cumulative view and the VIOLET/MAGENTA signal stayed intact; the comparative framing and chart-junk that made the reports feel dated were stripped.