HOSPITAL OPERATIONS · CLIENT WORK
Nurse and leader rounding for Elevra, the hospital operations platform at Compass Healthcare Digital. Every diagram is reconstructed, every person and site anonymous.
- Role
- Lead Product Designer
- Team
- Product, operations, engineering
- Status
- Production work, recreated for this site
TL;DR
I picked up a rounding product that had been talked about for six months without an architecture, and the most useful thing I did was stop drawing. The call that mattered was structural, not visual: one module belonged inside a system it had been drawn next to.
The real surface
The product these decisions live inside.
A recreation of Elevra's rounding surface.
Pick a round
Elevra
Sacred Heart Medical
Completed rounds
82%
Patients w/o rounds
4
Rounds w/ opportunity
22%
Rounds
Diane Foster
Age 67 · ICU / 201-A
Open opportunity
Missed drink in meal order
Quick entry, by voice
Elevra
Sacred Heart Medical
Leader Round, ICU / 201-A
Preliminary
Is the patient available for rounding?
Environmental services
Is the room clean and well-maintained?
Food & nutrition
Was the diet order delivered correctly?
Resolving an opportunity
Elevra
Sacred Heart Medical
Opportunities
0 of 1 resolved
Missed drink in meal order
Logged 8:30 AM
Same item, in follow-ups
Elevra
Sacred Heart Medical
Follow-ups
Food & Nutrition queue
Missed drink in meal order
Assigned to Food & Nutrition, logged 8:30 AM
The dashboard, live
Elevra
Sacred Heart Medical
Performance overview
142
rounds
Dietary complaints increased 23% this week across 3 units. This correlates with the menu change on Monday.
Where it came from
These are real screens from the shipped product. Every patient, site, and record shown is demo data.
The nav stack
Flat navigation items in one left rail, no grouping between them.
The card wall
Round cards repeat the same anatomy, whether the round is new or already complete.
The brand panel
One saturated institutional blue, used as a full-bleed fill behind the login form.
What the screens missed
The parts that never made it to a screenshot.
Every screen above is real, but partial. Rounding's actual weight sits in three mechanisms none of them show: what happens after a “No,” what the AI is and isn't allowed to hear, and what the tool already knows about the person being rounded on.
What happens after a “No.”
Fig. 02Service-recovery loop
A round finds a problem
Was the room clean and well-maintained?
Becomes an opportunity, with a clock
Elapsed against the 2-hour window
Closes with an owner and a timestamp
Consent comes before capability.
Elevra
Sacred Heart Medical
AI Summary
PendingWaiting on consent from the panel on the left.
Themes detected
The tool already knows the patient.
Elevra
Sacred Heart Medical
Diet
Allergies
PenicillinLanguages
EnglishThe instrument logs praise, not only problems
1 kudos on this patient's rounds so far
The problem
Six months of conversation, no structure.
Rounding is a hospital operations routine: a floor gets walked on a schedule, and what's found becomes a follow-up somebody has to close.
One module in a wider hospital operations platform, not a standalone build.
Six months of informal conversation. Screens existed. No agreed architecture, no release structure.
Opportunities drawn as their own module, beside a follow-ups system most sites were already running.
The prototype was doing work that requirements conversations should have been doing.
The decisions
Three calls, opened up enough to argue with.
A decision you can only see the outcome of is a claim. These carry what else was on the table and what each one cost, because that is the part a reader needs in order to disagree.
Stopped prototype iteration and mapped operating models instead.
Every review cycle re-opened the same questions about where things lived. The mapping that replaced it turned up the thing that reordered the release: sites did not round one way.
Nothing new to show in a review. A visible pause reads as a real cost in that kind of organization, and it was the right one to pay.
Made an opportunity a state a follow‑up can be in, rather than its own module.
The operating-model mapping showed the two were the same object at different points in its life. Most sites already ran follow-ups, with people trained on them and data in them.
Opportunities lost its own place in the navigation and the stakeholders who had been reviewing it as a distinct module lost the thing they could point at. It also meant reworking designs that had already been through review, which is the least popular kind of rework.
Phased the release around technical dependencies and pushed the patient module to last.
It depended on the hospital admit, discharge and transfer feed, an integration owned outside the team. Everything else in the rounding suite could be built without it.
The most compelling part of the story moved out of the first release. It was read as scope reduction until it was reframed as pilot risk, and reframing it took a conversation rather than a slide.
Fig. 01Release sequence
The round itself
History and question-level detail
The patient module
Nothing waits.
Mandate and outcomes
What was mine, and what came of it.
Anup
Product definition and design
- Module architecture and release structure
- Operating models and site segmentation
Manager
Portfolio roadmap and prioritization
- Roadmap above the module
- Final prioritization across the portfolio
- Handoff plan
Engineering
Implementation from this point forward
- Admit, discharge and transfer integrations
- Estimation and production behavior of follow-ups
What came of it.
The module shipped. Its AI features stayed in early testing, which is the normal speed of an industry that has learned to distrust a new tool arriving with a demo. Everything below is sorted by what kind of evidence stands behind it.
- 01Validated. The module shipped and runs in production. That is the one claim here a real working day has tested, and it is the reason the rest of this page is worth reading.
- 02Directional. The architecture correction was accepted and scoped by the engineers who would have had to build both systems. That is a judgment by the people closest to the cost, which is worth something, and it is not a measurement.
- 03Not measured. Adoption, compliance change, and time saved. None of them were measured, so a number here would be invented, and whether the two operating models hold up on a unit rather than on paper sits in the same column.
Whether the patient-module reframing holds once that feed exists, whether folding opportunities into follow-ups actually cuts duplicate work, and whether the ambient-capture exploration earns the trust an AI feature has to earn in this industry before anyone will use it.
At handoff
Already true
- Six months of informal stakeholder conversation, with screens but no agreed architecture.
- A follow-ups system already running at most sites, and a constrained design library the near-term track had to build inside.
- The decision to build rounding at all.
Still open
- The AI features, still earning their way in on an industry's timeline, not a demo's.
- The external patient module, still waiting on the same feed it had always been waiting on.
In hindsight
The stop should have come sooner.
- Lock the architecture before prototypingEarly on, there was more speculative building than I would repeat. I would settle the information architecture in one structured session before touching a prototype, rather than discovering it through iteration.
- Know which questions deserve a prototypeThe sharpest concern raised against the design, a way two rounders covering the same unit could double-log one finding, surfaced only because the prototype made the scenario visible. The judgment worth keeping is which questions deserve a prototype, and which ones are being avoided by building one.
More work