Designing an Explainable AI Triage Dashboard — A UX Case Study

AI · 6 min read

The challenge: clinicians needed quick, actionable recommendations from an AI model while also requiring understandable reasoning for clinical governance. The product team mapped the clinical workflow and identified three interaction goals: speed, trust, and auditability. They translated those goals into interface constraints that prioritized a single, time-critical recommendation on the primary screen with optional deep-dive explanations.

The UX pattern used layered explanations: a one-line rationale and confidence band visible at glance, a tappable “Why?” modal with feature-level contributions, and a downloadable audit trail for compliance. Designers collaborated with data scientists to surface explanations that mapped cleanly to clinical features (e.g., vital sign trends) rather than opaque model internals.

During a 12-week pilot across two hospitals, clinicians accepted AI recommendations 72% of the time when explanations were available versus 53% when only a score was shown. Time-to-decision increased by an average of 4 seconds with explanations present — an acceptable tradeoff given higher acceptance and fewer escalation calls to specialist teams.