Case study 01Healthcare AI

care.aiSeeing every room is easy. Deciding what a nurse sees is the design.

Ambient AI that watches every hospital room, so nurses and sitters see only what deserves them.

The care.ai Virtual Patient Monitor: a wall of twelve hospital room video tiles, each with the patient name and room controls.
  • RoleCreative Director, UX/UI Design Lead
  • Years2018 – 2026
  • PlatformConsole, wall, mobile, tablet
  • StatusAcquired by Stryker, 2024

01The moment

One nurse. Six patients. Twelve hours.

At 2:40 AM a post-surgical patient gets out of bed alone. Confused from pain medication, he will not press the call button.

NURSETWO ROOMS AWAY2:40 AM · BED EXIT

PatientNurseOut of bed, no call

2:40 AM
the critical moment
1
patient, recently post-surgical
1
nurse, two rooms away
0
calls. He does not press the button.

Nothing about that moment is unusual.

02The problem

Seeing everything is the easy part.

An AI that sees everything drowns a nurse in alerts, unless the interface decides what deserves her attention. That was the brief.

Bed exit · Room 412Act now
Illustration: every detection the AI could surface, and the one that earns an interruption.
72–99%
of hospital clinical alarms are false
350+
alarms per patient per day, reported by one hospital’s ICU

The brain learns to filter them out, and filters out the real one too.

Industry context, not results of this deployment. Figures as summarised by Nurse.org from published research.

03The approach

Decide what deserves a human.

Video: not storedBed exitRoom 412Act nowAn event, not footage
01

The room reports events, never footage.

  • Privacy is the default
  • Video has to justify itself
Act nowBed exit, fall, code eventsAcknowledgeBathroom, turning, pressure intervalInformationalCleaning, provider in room, meals
02

Every event gets a tier.

  • Set with clinical advisors
  • Legible in under a second
REJECTEDColour onlySHIPPEDNotifiedSHIPPEDAcknowledgedSHIPPEDResolved
03

Colour never travels alone.

  • 1 in 12 men has a colour-vision deficiency
  • Every status: a word and an icon

Tier colours and event examples from the product’s semantic palette. Icons are the production webfont.

04The flow

One event. One language. Every surface.

The nurse station and the phone in her pocket show the same three states.

  • Detect

    AMS-M2 sensor: an event, not footage.

  • Notified

    Notified

    Console and phone, same alert.

  • Acknowledged

    Acknowledged

    One tap stops the escalation clock.

  • Drop in

    Rosa D. GoodrowSettingsDrop in

    Live video and two-way talk.

  • Resolved

    Resolved

    Documentation is a by-product.

05Decisions that held

Five choices, and what they changed.

LocationPatientLast ActivityTimeVisualStatusCallRulesRoom 102 / Med Surg / Sacred HeartDon C. ThomasFall Prevention10:41 PMNotifiedRoom 107 / Med Surg / Sacred HeartRosa D. GoodrowRoom Entry12:17 AMAcknowledgedRoom 212 / Med Surg / Sacred HeartIris H. RocheProvider In Room9:00 PMResolvedRoom 217 / Med Surg / Sacred HeartGiuseppe M. PittsPressure Injury12:49 AMNotifiedRoom 109 / Med Surg / Sacred HeartSalvatore A. DillardAssesment2:32 AMAcknowledged

Table, not tiles.

  • One row per room, read in one fixation
  • One screen, not six
  • One click to live video

Before: a page per room, a separate alerts console, video in a third tool.

StatusCallRulesNotifiedAcknowledgedResolvedNotifiedAcknowledgedRulesFall PreventionPatient TurningChair Exit

Rules live on the row.

  • A popover, not a settings screen
  • Next to the alerts they prevent
  • Try it: open the gear, flip a rule
Alayna SimonisCAI70594Dallin QuitzonCAI38315Emely EmmerichCAI98016Leopoldo KlockoCAI32724Bed exitAlayna Simonis · CAI70594Acknowledge

Peripheral alerts, not modals.

  • Up to 16 rooms, 8 to 12 hours, dim light
  • Pulse on the border. Wall #1A202C.
  • A modal blinds the sitter to fifteen
TaskPatientStatusElapsedRoundingDon C. ThomasCompleted6:33RoundingRosa D. GoodrowDeferred34:27DischargeIris H. RocheCompleted32:47IntakeSalvatore A. DillardOverdue19:43IntakeThomas E. GarberIn Progress28:23RoundingCarl E. JacksonIncomplete48:26RoundingWendell A. LockwoodScheduled3:33RoundingTravis J. LockhartOverdue54:3145 min

The elapsed-time column.

  • How long a task sat in its state
  • A list becomes a bottleneck detector
  • Pilots: stalled work visible without asking
Reset–+Default framing

A wheel, with a way back.

  • Eight zones, not arrows
  • Hub resets to the approved framing
  • Try it: press a zone, then Reset

06The interface

Dense, calm, readable at a glance.

Each tab plays its own click-through. Click anything to take over.
CTVirtual Patient MonitorLayoutSmallLabelActivityFullscreenAdd Room© 2024 care.aiRelease NotesHelpAlayna SimonisCAI7059411:55:34Add any notes:Patient Is doing good. We are wellDallin QuitzonCAI3831511:55:34Add any notes:Patient Is doing good. We are wellEmely EmmerichCAI9801611:55:34Add any notes:Patient Is doing good. We are wellLeopoldo KlockoCAI3272411:55:34Add any notes:Patient Is doing good. We are wellMs. Arlene HoegerCAI9345011:55:34Add any notes:Patient Is doing good. We are wellMiller BotsfordCAI7119011:55:34Add any notes:Patient Is doing good. We are wellDr. Amanda BraunCAI4327811:55:34Add any notes:Patient Is doing good. We are wellHayley NolanCAI4105911:55:34Add any notes:Patient Is doing good. We are wellDamon MayerCAI9091911:55:34Add any notes:Patient Is doing good. We are wellEunice Stracke IVCAI9784311:55:34Add any notes:Patient Is doing good. We are wellJacky MorarCAI3668511:55:34Add any notes:Patient Is doing good. We are wellEmiliano HermannCAI3849211:55:34Add any notes:Patient Is doing good. We are wellFailed to connect to the roomReconnectAdd any notes:Patient Is doing good. We are wellAlayna SimonisCAI70594Live, two-way talkAdd any notes:Patient Is doing good. We are wellBed exitAlayna Simonis · CAI70594NotifiedAcknowledgedResolvedNothing to act on.This room is quiet.AcknowledgeDrop inResolveAdd RoomSelect FacilitySelect AreaRoom 103 / Med Surg / Sacred HeartRoom 200 / Med Surg / Sacred HeartRoom 103 / Med Surg / Sacred HeartRoom 228 / Med Surg / Sacred HeartRoom 124 / Med Surg / Sacred HeartRoom 114 / Med Surg / Sacred HeartCancelSave

Swipe the wall sideways

The live sitter wall plays itself. Click anything to take over.

The real product, three moments

The Virtual Patient Monitor wall. Tiles show an alert message, a patient-not-in-room timer, ambient monitoring rules, add notes and create preset, each as a small popover on its own tile. The Add Room dialog over the wall: a list of rooms with sensor state shown by icon, facility and area selectors, and a checkbox on each room. One room expanded to fill the screen, with the other rooms stacked in a column on the right.
Tile tools open in place. Alert message, patient not in room, rules, notes, presets: each is a popover on its tile, never a new page.

The tabs above are the product redrawn as vector with demo data. Numbered marks point to the design decisions. Drill-downs, rules, calls and Drop In are illustrative behaviour. The strip below the tabs, and the hero, are real product screenshots with demo data.

07The system

One system for five distances.

A hallway sign at four metres and a phone in a corridor each got a step on one scale, never their own rules.

255 icons, one set

Hover an icon
  • One versioned webfont
  • The production font: every surface renders identical marks
  • Drawn from the font’s own outlines, not an image

Colour is a clinical decision

Act now#E5544B · bed exit, fall, code eventsAcknowledge#E8A33D · bathroom, turning, pressure intervalInformational#4CAF7D · cleaning, provider in room, mealsBrand cyan#25C3E0 · interaction only, never statusMonitoring dark#1A202C · 8 to 12 hour low-light shifts
  • Act now: solid disc, fast pulse
  • Acknowledge: ring, slow pulse
  • Informational: plain ring

Try it: hover a row.

One type scale

  • Room 412Door signage · 4 m
  • Fall PreventionWall display · 2 m
  • Bathroom AssistanceConsole · 60 cm
  • Room 102 / Med Surg / Sacred HeartDense table
  • AcknowledgedStatus, never smaller

How it was governed

1234
  • One library, one ownerA module could not fork a colour.
  • Versioned in the fontNobody used last year’s set.
  • Written downSpecs, not just design files.
  • Reviewed before releaseA good argument changed the system.

State, in the product

DevicesRulesRoom 102FPPTRoom 104FPPTRoom 212FPPTRoom 217FPPT
  • Devices: grey off, green on, red on alert
  • Rules: cyan when on

The most valuable thing I left behind was not a screen. It was the rule that a nurse moving between four products should never have to ask what a colour means.

08Leadership

Designing the system meant designing the connections.

AI/ML engineersClinical advisorsSolution architectsC-suiteHospital clientsData-labelling teamFront-end developersUI designersDesign
  • AI/ML engineers
  • Clinical advisors
  • Solution architects
  • C-suite
  • Hospital clients
  • Data-labelling team
  • Front-end developers
  • UI designers

Who I led

  • UI designers
  • Front-end developers
  • Data-labelling team

Set quality benchmarks, ran UX review, mentored designers to own modules end to end.

Who I translated between

  • AI/ML engineers
  • Clinical advisors
  • Solution architects
  • C-suite

Each defines “done” differently. I held one definition across all four.

Where I defended it

  • Executive reviews
  • Hospital client sessions
  • HIMSS
  • MGMA
  • HLTH
  • NAHAM

Also art-directed the product films and animated walkthroughs used to sell the platform.

The recurring argument was less.

Engineering wants more detections, more data, more alerts. Alarm fatigue is what happens when nobody wins that argument.

09A correction

The call I got wrong.

I filed the AMS-H1 bedside unit as hardware. It is a complete patient app: thirteen screens for elderly users in assisted living and home care.

Filed under: deviceBelongs under: userProvisioningHardware UXDensity as styleElderly usersAssisted livingHome careAMS-H113 screens

How I had been thinking

  • Grouped work by device and platform
  • Clinical density as the house style
  • Bedside unit as a provisioning surface
  • Information per square inch

What the patient app needed

  • 150 × 150 pt tap targets
  • Icon before text
  • Full-screen confirmations
  • Home on every screen
HomeCall a nurseLightsSoundNurse calledEvelyn has been calledto your roomHome

Try it: press Call a nurse.

A toast is correct by convention and wrong in practice. A frightened patient needs a whole screen: “Evelyn has been called to your room.”

I was categorising by device, not by user. Correcting that changed how I audit my own scope.

10Outcome

Acquired by Stryker.

care.ai now sits inside Stryker’s Smart Care business.

12 Aug 2024
Acquisition announced
17 Sep 2024
Acquisition completed
2024–2026
I stayed through the acquisition and into integration.
100+
deliverables accepted with minimal revision

Who runs on it, as published

Houston MethodistEnterprise rollout, from 2022
Henry Ford HealthVirtual nursing pilot, then enterprise-wide across 13 locations (Stryker)
Parkview HealthVirtual nursing and sitting

Customer programmes as published by those organisations and Stryker. Context for the scale the design had to survive, not a claim of individual credit.

What I took from it

  • Restraint is a deliverable.
  • Design for the user, not the device.
  • A system is a rule people can rely on.

Good design is knowing what to leave out.

Sources: Nurse.org, care.ai announcement, Stryker completion, Stryker on Henry Ford. Product screens are redrawn as vector with demo data, except the hero and the real-product strip, which are screenshots with demo data; the live wall, drill-downs and Drop In are illustrative behaviour, not telemetry. WCAG 2.1 AA targeted, demos keyboard operable, motion off under reduced-motion.