MedMetrics · AI-Powered Clinical Management Platform
One patient record. Five roles. Zero manual handoffs.
MedMetrics — a five-role, AI-powered clinical management platform combining EMR/EHR, scheduling, billing, AI-assisted imaging, clinical workflow, and pharmacy management — designed as a single unified system for outpatient clinics and specialist groups.
Format
Design Study
Sector
Healthtech, Clinical Management, Practice Operations
Roles addressed
Patient, Receptionist, Doctor, Health Provider, Admin
Method
Industry research, workflow analysis, design
Year
2026
The 90-second version
MedMetrics replaces the average clinic's eight disconnected tools with one patient record and five role-filtered surfaces — so a vitals flag, a lab result, and a prescription all propagate live, without a phone call in between.
63%
Physicians reporting burnout in 2024. Administrative burden is the leading cause.
AMA, 2024
8
Average number of separate tools a US outpatient clinic uses to run daily operations.
KLAS Research, 2024
40k–80k
Annual deaths in the US attributable to diagnostic errors, most occurring in the gaps between disconnected systems.
BMJ Quality & Safety
01
Healthcare software was built by departments. Patients pay the cost.
A patient with chest discomfort walks into a clinic and triggers a chain of handoffs that no single tool in that clinic can see end to end. The patient books through one system. The receptionist checks them in through another. The doctor opens a third to review the chart, a fourth for the ECG, a fifth for imaging. By the time the prescription reaches the pharmacy, it has passed through six or seven separate interfaces — none of which share a live view of the patient's status.
This is not an edge case. KLAS Research reports that US outpatient clinics run on an average of eight separate tools per day. Physicians spend two hours on administrative tasks for every hour of direct patient care. 63% report burnout, and administrative burden is the leading cause. The system is not slow because the staff are slow. It is slow because the tools were built to serve departments, not the patient journey.
The clinical cost is worse. BMJ Quality and Safety estimates 40,000 to 80,000 annual deaths in the US are attributable to diagnostic errors — most of which occur not because the information did not exist, but because it was not visible at the right moment, in the right system, in front of the right person.
AI has entered the space fast. The FDA cleared over 521 AI-enabled medical devices by 2023, the majority in radiology. But integration remains unsolved. AI imaging tools flag findings in separate portals. Physicians copy results into notes manually. The AI accelerates one step and adds friction to three others. The problem is not capability. It is architecture.
02
Four challenges this project set out to solve.
Challenge 01
Five roles, one patient journey, five completely different jobs
A patient, a receptionist, a doctor, a health provider, and an admin all participate in the same clinical episode. Each needs a different view of the same underlying data and makes decisions that affect what every other role sees next. Most platforms either build separate applications that share no context, or one application that serves no one well. The challenge was designing a unified data layer with five genuinely role-specific surfaces on top of it — where an action taken by the health provider instantly updates what the doctor sees, without a phone call or a manual note in between.
Challenge 02
AI that assists without replacing clinical judgment
AI clinical insights arrive in most platforms as outputs the physician is asked to trust without being able to interrogate. A flag with no confidence level, no source, and no override path is not a clinical tool. The challenge was designing AI assistance that makes the physician faster and more informed without removing them from the decision. Every AI output needed to be transparent, editable, and overridable — with the physician's decision recorded either way.
Challenge 03
Documentation during the visit, not reconstructed after it
The standard clinical workflow produces documentation as a downstream task. The doctor ends the consultation and reconstructs the visit from memory. The Annals of Internal Medicine identified this as the source of the 2:1 EHR time burden. The challenge was designing a consultation workspace where AI Scribe captures the visit in real time, the doctor edits inline during the encounter, and the note is ready to sign the moment the consultation ends — not an hour later.
Challenge 04
Real-time clinical status visible across all five roles simultaneously
When a health provider records elevated vitals and flags urgent risk, the doctor should see it the moment it is saved. When a lab sample is collected, the patient should receive confirmation automatically. When a prescription is sent, the pharmacy queue should update before the patient reaches the counter. The challenge was designing live status propagation so clinical events move through the system without any role having to chase another for an update.
03
What the research said before any screen was drawn.
Source
AMA, 2024
Finding
63% of US physicians report burnout; administrative burden is the leading cause
Implication for design
Documentation reduction is not a feature request. It is the primary clinical workflow design problem.
Source
KLAS Research, 2024
Finding
US outpatient clinics run on an average of 8 separate tools daily
Implication for design
Consolidation is the value proposition. The platform earns trust by replacing tools, not adding one.
Source
Annals of Internal Medicine
Finding
Physicians spend 2 hours on EHR tasks per 1 hour of direct patient care
Implication for design
Time-on-documentation is the primary design metric. Every consultation surface decision should move this ratio.
Source
FDA, 2023
Finding
521 AI/ML-enabled medical devices cleared; majority in radiology
Implication for design
AI imaging is a present clinical expectation. Design for workflow integration, not novelty.
Source
BMJ Quality & Safety
Finding
40,000–80,000 annual US deaths attributable to diagnostic errors
Implication for design
Clinical decision support must surface information at the moment of decision. Gaps between systems are where errors happen.
Source
JAMIA
Finding
49–96% of medication alerts are overridden by clinicians
Implication for design
Alert fatigue is a design failure. Alerts that explain their reasoning get acted on. Alerts that do not get clicked past.
Source
AMA, 2022
Finding
Physicians spend 13 hours per week on prior authorizations
Implication for design
Automation is the answer. Automation visibility — showing what the system did — is what makes physicians trust it.
Source
McKinsey Global Institute, 2023
Finding
AI-assisted radiology reads reduce time-to-diagnosis by 30–50% in high-volume settings
Implication for design
Speed gains from AI exist only when the output is integrated into the clinical workflow. A separate portal erases the gain.
Source
Forrester Research
Finding
Patients using a connected patient portal have 2.6x higher clinic retention
Implication for design
Patient-facing surfaces are a retention mechanism for the clinic, not just an access mechanism for the patient.
Source
ACAMS, 2025
Finding
67% of healthcare operations staff report burnout; alert volume is the leading cause
Implication for design
Reducing noise is as important as surfacing signal. Defaults should hide what is not immediately actionable.
04
Four principles every screen had to defend.
01
Role-first, not feature-first
The doctor's view of a patient record and the receptionist's view of the same patient's appointment are different products built on the same data. Every surface was designed from the role's primary question outward — not from the data model inward. That distinction is what separates a clinical tool from a database with a UI.
02
Live status over manual communication
Every status update that requires a phone call, a manual message, or leaving the platform is a workflow failure. Clinical events — vitals flagged, labs collected, imaging ready, prescription dispensed — propagate to every relevant role in real time. No role has to ask another what is happening.
03
AI as augmentation, not automation
Every AI-assisted action keeps the clinician in the decision loop. AI Scribe produces a draft the doctor edits before signing. AI imaging flags show confidence level, source region, and a mandatory override path. The physician is faster and better informed. The clinical decision is still theirs.
04
Documentation during, not after
The consultation workspace, the telehealth session, the imaging review — all produce documentation in the same interaction, not as a downstream task. If the doctor has to reconstruct anything from memory after the fact, the design has failed.
05
What this project covered and what it did not.
In scope
- Five role-based dashboards: Patient, Receptionist, Doctor, Health Provider, Admin
- Patient: appointment booking, telehealth entry, prescriptions, lab reports, billing and payments, AI Health Assistant, medical records, referrals
- Receptionist: check-in queue, appointment management, insurance verification, billing support, telehealth coordination, referral management
- Doctor: patient queue, EMR/EHR consultation workspace with AI Scribe, prescriptions, lab and imaging requests, telehealth surface, AI clinical insights, referrals
- Health Provider: patient queue, vitals recording with live monitoring triggers, lab workflow, imaging uploads, medication support, task queue
- Admin: user and role management, department configuration, billing and revenue, inventory, lab operations, analytics, audit logs, system settings
- AI Scribe: real-time transcription and SOAP note generation during consultation
- AI Imaging: three-state finding detection with confidence level, source region, and override workflow
- AI clinical risk insights on the doctor dashboard
- Live clinical status propagation across all five roles
- Pharmacy and Medicine Management: prescription fulfillment, drug interaction check, inventory monitoring, low-stock alerts
- Billing and insurance: co-pay calculation, insurance verification, prior authorization workflow, patient in-app payment
- Design system token architecture for PHI fields, consent states, audit trail, and AI draft surfaces
Out of scope
- Native mobile app beyond responsive web
- Wearable and continuous monitoring device integration
- Revenue cycle management and denial management beyond billing queue visibility
- Claims adjudication
- Multi-clinic network and enterprise federation
- E-prescribing to external pharmacy networks
- Localization beyond English
06
Five roles, one journey.
Patient
Primary question
“When is my appointment, what do I need to prepare, and what has my doctor decided?”
Primary action
Book appointment, join telehealth session, view prescriptions and lab reports, make payment, communicate with care team
Modules
Dashboard, Appointments, Telehealth, Prescriptions, Medications, Lab Reports, Medical Records, Billing & Payments, Referrals, AI Health Assistant, Notifications, Profile & Settings
Receptionist
Primary question
“Who is arriving today, what is blocked, and what needs my attention before the queue backs up?”
Primary action
Check in patients, verify insurance, manage appointment queue, coordinate telehealth sessions, handle billing exceptions, process referrals
Modules
Dashboard, Appointments, Patient Registration, Check-In Queue, Billing & Payments, Insurance, Referrals, Telehealth Coordination, Notifications, Profile & Settings
Doctor
Primary question
“What do I need to know about this patient before I speak to them, and what do I need to complete before I move to the next one?”
Primary action
Review pre-visit summary, consult with AI Scribe active, diagnose, prescribe, request labs and imaging, review AI clinical insights and imaging flags, complete and sign note
Modules
Dashboard, Appointments, Patients, Consultation (with AI Scribe), EMR/EHR, Prescriptions, Medicine Management, Lab Tests & Imaging, Referrals, Telehealth, AI Assistant, Notifications, Profile & Settings
Health Provider
Primary question
“Which patient needs me next and what tasks are waiting for me to complete before the doctor can proceed?”
Primary action
Record vitals and trigger monitoring alerts, prepare patients for procedures, coordinate lab collection, upload imaging, support medication workflow
Modules
Dashboard, Patient Queue, Procedures, Vitals & Monitoring, Lab Workflow, Imaging Uploads, Medication Support, Medicine Inventory, Tasks, Alerts, Notifications, Profile & Settings
Admin
Primary question
“Is the operation running, and where is it at risk right now?”
Primary action
Monitor operational health, manage users and roles, review billing and revenue exceptions, oversee inventory, audit compliance, configure system settings
Modules
Dashboard, Users & Roles, Departments, Appointments, Billing & Revenue, Medicine Management, Inventory, Lab Operations, Analytics, Audit Logs, System Settings, Notifications
07
Five decisions, five forks, five calls this project would defend.
Decision 01 of 05
One patient record, five role-filtered views
What I considered
Build separate patient record modules per role. Lower build complexity, clearer scope boundaries per team.
What I chose
One patient record. Role determines which sections are visible, editable, and in what order they appear. All five roles are looking at the same underlying record.
Why
Fragmentation was the problem being solved. Rebuilding it inside the platform because role-scoped modules are easier to scope would have reproduced exactly what clinics already suffer from. A single record also means that when the health provider flags Rahim's vitals as critical, the doctor sees it in the chart already open — not in a separate notification from a separate system.
Decision 02 of 05
Consultation workspace and AI Scribe as one surface, not two
What I considered
Note editor opens after consultation ends. Standard pattern across most EHR systems. Clinicians already expect it.
What I chose
The note editor is embedded inside the consultation workspace. AI Scribe runs during the visit, populating SOAP sections in real time. The doctor edits inline during the encounter. The note is ready to sign the moment the consultation ends.
Why
Post-consultation documentation is the primary source of the 2:1 EHR time burden. Documentation from memory is slower and less accurate. Designing the note editor as a parallel workflow inside the consultation — not a separate downstream step — is the single highest-leverage design decision in the entire clinical surface.
Decision 03 of 05
Three AI imaging states, not a binary finding flag
What I considered
Flag or no flag. A red indicator when the AI detects a finding, nothing when it does not.
What I chose
Three states — Finding Detected, Review Recommended, No Finding. Each state shows confidence level, scan region, AI model version, and an explicit override path that requires a brief clinical note before proceeding.
Why
JAMIA documents medication alert override rates at 49 to 96%. Binary flags produce the same behavior — they get clicked past. A finding the physician cannot interrogate does not get acted on appropriately. The three-state system matches actual clinical reasoning. The override note requirement, adapted from clinical decision support research, reduces inappropriate overrides significantly without blocking legitimate ones.
Decision 04 of 05
Live status propagation, no manual handoff
What I considered
A notification center and in-app messaging between roles. Staff communicate when something changes.
What I chose
Clinical events propagate automatically. When the health provider records Rahim's vitals at 94% SpO2 and 112 bpm, Dr. Carter's dashboard updates instantly. When the lab sample is collected, Rahim receives a confirmation. When the prescription is sent, the pharmacy queue updates before Rahim reaches the counter.
Why
A notification center still requires a human decision about when to message and what to write. In a clinical environment where those readings indicate urgent cardiac risk, that decision loop is a delay the design should not introduce. Live propagation removes the manual step. The status is always current. The role that needs to act sees it without being asked.
Decision 05 of 05
Admin dashboard as real-time operational monitor, not reporting tool
What I considered
Build the admin dashboard as a reporting surface. Charts, trend analysis, end-of-day exports.
What I chose
The admin dashboard leads with real-time operational health — queue status, staff workload, inventory risk, compliance alerts, revenue flags. The primary action is identifying what is at risk right now and acting on it inline — not reviewing a summary of what happened after the clinic closed.
Why
Reports are useful for analysis. They are not useful when a queue delay is building at 10 AM and the admin does not know until the 4 PM summary. The admin's job in a live clinic is triage and response. The dashboard was designed to surface the problem before it becomes a crisis.
08
Inside the AI scribe
The consultation workspace was the most complex surface in the platform. When Dr. Carter opens Rahim's consultation at 10:32 AM, he already has the complete pre-visit summary assembled automatically — chief complaint, vitals, ECG upload, AI cardiac risk insight, insurance profile, and prior visit history. He does not reconstruct context. He starts from it.
As the consultation begins, AI Scribe activates. It captures the dialogue, maps content to SOAP sections, and populates the note in real time. Dr. Carter sees the note building as he speaks. He edits inline — adding the ICD-10 code for Angina (I20), adjusting the assessment, confirming the treatment plan. By the time he ends the consultation, the note is complete, reviewed, and ready to sign.
Three specific decisions shaped the Scribe output surface. First, the output is always shown as a draft, never as a complete note. The visual treatment — a distinct draft state with an explicit Review and Sign action — makes it unambiguous that the AI has done the first pass and the doctor is completing it. Second, SOAP sections appear in order of clinical priority, not structural order. Assessment and Plan come first because that is where the high-stakes edits happen. Third, AI clinical insights — in Rahim's case, "Possible unstable angina detected. Recommend immediate cardiac evaluation" — appear in the consultation sidebar, not embedded in the diagnosis field. They inform the doctor's thinking. What Dr. Carter types in the diagnosis field is his clinical judgment. The AI surfaces evidence. The distinction is preserved in the design.
09
Inside insurance & billing
Billing in a multi-role clinic is not a back-office function that happens after the visit. It is a live constraint on what the receptionist can check in, what the doctor can order, and what the patient sees before they leave. MedMetrics treats coverage verification, co-pay calculation, and prior authorization as first-class workflow states — visible on the patient record, not buried in a separate billing module.
At check-in, the receptionist sees insurance verification status on the queue card before the patient reaches the desk. Failed verification surfaces inline with the specific field that failed — member ID, group number, eligibility date — so the exception is resolved without opening a second system. Co-pay is calculated from the verified plan and shown on the same card the receptionist uses to generate a token.
During the visit, when Dr. Carter orders imaging, the order surface shows estimated patient responsibility alongside clinical context — not as a separate billing screen the doctor has to hunt for. Prior authorization requests initiated from the order surface propagate status back to the receptionist queue and the patient record automatically.
After the visit, the patient sees itemized charges with insurance adjustment and amount due on the same mobile surface they use for lab results. Payment is one action from the billing summary — not a redirect to a third-party portal.
Workflow area
Check-in & registration
System handles
Real-time eligibility verification, co-pay calculation, coverage tier display
Patient sees
Verified badge or specific failure reason on appointment card
Receptionist action
Resolve exceptions inline, generate token, flag for manual review if needed
Workflow area
Clinical orders (labs, imaging)
System handles
Coverage check against ordered service, prior auth trigger if required
Patient sees
Estimated out-of-pocket on order confirmation
Receptionist action
Monitor auth status on queue; notify patient if delay affects wait time
Workflow area
Consultation & documentation
System handles
ICD-10 coding suggestions, charge capture from signed note
Patient sees
Nothing during visit — billing assembles silently
Receptionist action
No action unless billing exception flagged post-visit
Workflow area
Post-visit payment
System handles
Itemized bill with insurance adjustment, payment processing, receipt
Patient sees
Amount due, payment method selection, confirmation
Receptionist action
Handle walk-out exceptions, payment plan requests, refund initiation
Workflow area
Prior authorization
System handles
Auto-submission where supported, status tracking, deadline alerts
Patient sees
Auth status on appointment and lab report surfaces
Receptionist action
Follow up on pending auth, coordinate rescheduling if denied
10
Inside pharmacy & inventory
When Dr. Carter signs Rahim's prescription, three things happen simultaneously: the pharmacy queue updates with the order and drug interaction check results, the inventory system decrements stock for dispensed items, and Rahim's patient record shows the prescription as sent — before he reaches the counter.
The pharmacy surface was designed around the dispensing workflow, not a list of orders. Each prescription card shows the drug, dosage, interaction flags, insurance coverage status, and stock availability on one surface. A low-stock alert on a commonly prescribed item appears on both the pharmacy queue and the admin inventory dashboard — so the admin can reorder before the pharmacy runs out mid-day.
Drug interaction checks run at prescription creation, not at dispensing. The doctor sees interaction severity inline in the prescription editor with an override path that requires a clinical note — the same pattern used for AI imaging flags. By the time the order reaches the pharmacy, interactions have already been reviewed and resolved or explicitly overridden by the prescribing physician.
Inventory monitoring extends beyond pharmacy to clinical supplies — vitals equipment, lab collection kits, imaging contrast. Admin sees low-stock risk alongside queue and revenue flags on the operational dashboard, because a stockout on lab collection tubes affects the health provider's task queue the same way a staff shortage affects the receptionist's wait times.
11
Token architecture.
Layer 01
Primitive
Raw values. Never referenced directly in components.
- color-clinical-red-500: #DC2626
- color-clinical-amber-500: #D97706
- color-clinical-green-500: #16A34A
- color-neutral-900: #0F172A
- space-4: 16px
- font-size-base: 14px
- font-size-clinical-label: 12px
Layer 02
Semantic
Intent-based aliases. Components reference these, not primitives.
- color-finding-detected
- color-review-recommended
- color-vitals-critical
- color-consent-required
- color-phi-field
- color-audit-trail-entry
- color-ai-draft-surface
- color-live-status-pulse
Layer 03
Component
Scoped to specific UI patterns.
- color-ai-scribe-draft-background
- color-ai-flag-border-high
- color-ai-flag-border-moderate
- space-consultation-panel-padding
- color-phi-input-background
- color-override-required-indicator
PHI token note. All fields containing Protected Health Information reference color-phi-input-background. Any global change to PHI field presentation — for dark mode, high-contrast, or audit mode — is a one-line token update, not a component-level sweep. Consent states, override requirements, and audit trail entries each hold dedicated semantic tokens so they are never resolved to a generic status color.
Image Placeholder
Design system — visual reference coming soon. Figma overview showing color tokens organized by layer, typography scale, spacing scale, and key component snapshots (AI scribe draft state, severity score, patient record card, KPI strip). File name: design-system-overview.png
12
What I'd be lying if I said was finished
- The pre-visit summary surface — what the doctor sees before speaking to a patient — was designed based on clinical workflow logic, not tested against actual physician reading behavior. Rapid prototype testing with practicing clinicians would sharpen the section order and information hierarchy significantly. That is the next validation priority.
- The Health Provider role has the most complex task queue in the platform. Vitals recording, lab coordination, and imaging upload are designed. The procedure support workflow, medication assistance surface, and the handoff from Health Provider to Pharmacy need a dedicated design pass at higher fidelity. That is the next module.
- The Admin analytics layer was scoped to real-time operational monitoring. Historical reporting — department-level trends, billing forecasting, compliance audit exports — was deliberately deferred. That becomes its own design project once the real-time layer is in use and generating data worth analyzing over time.
- The patient mobile app exists as responsive web layouts, not as a native experience tested against actual patient usage patterns. Appointment booking and lab result viewing are designed. Push notification behavior, offline access to records, and payment flows on low-connectivity networks need dedicated mobile research.
- The full patient journey composite — Rahim's episode from booking to prescription collection across all five roles — was mapped in workflow diagrams but not yet rendered as a single visual narrative. That composite is the highest-value artifact for stakeholder alignment and remains on the list.
13
Shipped screens.
patient-mobile-dashboard.png — Patient mobile app: upcoming appointment, lab results, prescriptions, billing summary, AI Health Assistant entry. File name: patient-mobile-dashboard.png
lab-collection-workflow.png — Health provider lab collection: sample labeling, patient confirmation, chain-of-custody status. File name: lab-collection-workflow.png
billing-payment-surface.png — Patient billing and payment: itemized charges, insurance adjustment, payment method selection. File name: billing-payment-surface.png
pharmacy-dispensing-queue.png — Pharmacy dispensing queue: prescription card with interaction flags, stock status, and coverage. File name: pharmacy-dispensing-queue.png
patient-journey-composite.png — Full patient journey: Rahim's clinical episode from booking to prescription collection across all five roles. File name: patient-journey-composite.png
Related Projects
This study extends the multi-role healthtech patterns from Karaz Care — Healthcare Platform UX Case Study, where five clinician dashboards share one diabetes-care record.
The AI scribe and imaging-assist flows here build on the trust layer I explored in the AI Customer Support Copilot case study— transparent confidence, editable outputs, and human override paths.
For consumer health in the same ecosystem, see Karaz Health — Diabetes Mobile App UX Case Study, where daily logging and clinical data meet in one patient app.
If you are hiring for a senior product designer role in healthtech and want to discuss this work or anything in my portfolio, reach me at hey@shahriarsultan.com.