Back to selected work
Healthcare AIEnd-to-End UX2025

HIP One — Designing for trust in healthcare AI

Redesigning medical review around a single principle: AI can accelerate evidence handling, but the human reviewer must retain visibility, context, and accountability.

40%Faster turnaround
65%Fewer manual-entry errors
92%Reviewer AI trust
4.2×First-year ROI
Challenge

Three kinds of friction were compounding.

Process delay

PDF- and fax-heavy review created long turnaround times and repeated manual steps.

System fragmentation

Separate review, analytics, and compliance tools created silos, duplicate work, and inconsistent context.

Human cognitive load

Dense evidence, audit pressure, and opaque AI recommendations reduced reviewer confidence and increased manual overrides.

Competitive gap

Move from extraction to decision support.

HyperscienceStrong OCR; limited clinical decision context and role-based review.
RossumDocument automation optimized for finance rather than clinical workflows.
NanonetsFlexible ML extraction, but developer-centric rather than reviewer-centric.
HIP One directionAI + workflow + compliance in one audit-ready review experience.
Primary research

Research designed to reduce clinical and adoption risk.

01Stakeholder interviews

Aligned compliance constraints, workflow goals, and non-negotiables before interface design.

02Contextual inquiry

Observed manual PDF/EHR workflows to expose workarounds, interruptions, and high-friction handoffs.

03Artifact mapping

Reviewed spreadsheets, forms, and operational artifacts to uncover hidden workflow logic.

04Early IA validation

Tested navigation and information grouping with representative users before high-fidelity design.

Jobs to be done
Medical reviewer

Needs fast access to patient context, risk evidence, and source material to make a defensible decision.

Operations lead

Needs workload, task status, SLA risk, and bottleneck visibility to prevent escalation.

Quality analyst

Needs early risk signals and traceable evidence to support compliance and audit readiness.

Design principles
Information overload→ Progressive disclosure

Summaries first, evidence on demand, hierarchy shaped around scanning behavior.

Low AI trust→ Explainable AI

Confidence, provenance, and human confirmation visible before consequential actions.

Compliance anxiety→ Invisible guardrails

RBAC, validation, audit trails, and secure data handling embedded in the flow rather than bolted on.

Key executions
Reviewer dashboard

Priority, type, status, and SLA cues make workload state instantly scannable.

Document intake

Drag-and-drop upload, real-time validation, and required metadata prevent incomplete cases entering the pipeline.

AI-assisted review

Source document and AI decision panel remain side-by-side to reduce context switching and support verification.

Explainable summary

Low-confidence fields, accuracy signals, and targeted validation states direct attention where it is most needed.

Information architecture

Five legacy tools → one role-based workflow.

DashboardReview listExamine & updateDocumentsMedicare analytics

Navigation follows the clinical task sequence while permissions determine what each role can see and do.

Outcomes
40%Faster medical review completion
65%Reduction in manual-entry errors
92%Reviewer confidence in AI workflows
4.2×First-year return on investment
  • Role-based interfaces reduced reviewer cognitive load.
  • Five fragmented tools were consolidated into one workflow model.
  • Audit-ready controls were embedded across critical steps.
Email copied