AI-assisted patient matching

AI-Assisted patient matching flow

Overview

My role & goal

As the Senior UX Designer, I led the design of improving how administrative staff in an Electronic Medical Record (EMR) system match uploaded documents to the correct patients. Today, admins receive large batches of documents, often containing multiple patients within a single file. Matching and splitting these documents manually is repetitive, time-consuming and highly error-prone.

The goal was to streamline this workflow using AI: automatically read uploaded files, identify which content belongs to which patient and present admins with clear, actionable matches.

The team

  • Senior UX Designer

  • Project Managers

  • Business Analyst

  • Developers

  • Quality assurance

Understanding the problem

User group: Clinic administrators and EMR support staff

Methods: Workflow exploration, comparative analysis with another EMR’s patient matching feature

Our PM presented the initial problem: admins were spending significant time manually splitting documents, identifying patient information and confirming matches with the system, all tasks that could benefit from automation.

To deepen my understanding, I reviewed:

  • The existing manual workflow

  • How a different EMR within the company handled patient matching (even though their flow differed, it served as a useful reference point)

Defining key design goals

I was then able to established the following high-level goals:

  • Reduce the time spent on matching documents with patients.

  • Simplify handling for documents containing multiple patients

  • Minimize errors that could result in misplaced information

  • Provide a clear and trustworthy review process for AI-generated matches

Ideation & exploration

I explored several conceptual flows:

  • How to introduce AI recommendations early

  • How to keep admins from performing unnecessary manual steps before AI runs

  • How to visually indicate match confidence

  • How to handle edge cases (e.g., documents manipulated before upload, causing the AI to fail)

Iterative design

Method: Internal design reviews with the PM, BA, and developer — including team members from another EMR’s AI Matching project

Approach: Low-fidelity mockups → interaction flows → clickable prototype → iterative refinements

I worked closely with internal team members who had experience with patient matching in another EMR. These review cycles offered quick, relevant feedback and helped refine the flow.

Key insights from iteration:

  • Users needed to understand why the AI couldn’t match a document

  • Confidence indicators needed to be clear and non-technical

  • The flow for multi-patient documents required special attention

  • Admins should be encouraged to use the AI before performing any manual steps

  • Clear next steps (select, review, fix and confirm) to help reduce uncertainty

Final design dolution

✓ AI-generated patient matches: A clean review interface where admins see suggested matches alongside confidence indicators.

✓ Multi-patient document handling: A clear workflow for verifying and correcting documents that include multiple patients.

✓ Error / No-match scenarios: Guided interactions explaining next steps if the AI could not confidently match a document.

✓ Manual override: A streamlined way for admins to adjust or correct AI matches without losing context.

✓ Intelligent workflow entry points: The system encourages users to let the AI process documents first, rather than beginning with manual tasks.

Outcome

Even without direct user testing, internal validation confirmed that the AI flow reduced manual effort, clarified key workflows and improved transparency.

When I left the company, the feature was in active development with engineering, supported by the full set of designs and detailed flows I delivered.

Reflection

This project showed how important it is to design AI workflows that feel transparent and trustworthy. I included it because AI-driven efficiency is becoming essential in healthcare and this work reflects my ability to design for complex data-sensitive processes.