UX Case StudyTeaching a hospital
platform to answer back.
Designing the onboarding, consent and interaction model for a new conversational AI capability inside teamplay — Siemens Healthineers' digital health platform for radiology performance management.
This work is covered by an NDA with Siemens Healthineers. What follows is a public-safe account of the design thinking and process — sourced from Siemens' own public materials, without proprietary screens, data, or specifics. Happy to walk through the real thing in a 1:1.
A platform built for decision support —
now learning to converse.
teamplay is Siemens Healthineers' digital health platform — a vendor-neutral foundation that connects data, systems and institutions, and gives radiology departments a clear, real-time view of their operational and clinical performance. It's built to make Siemens' own decision-support software, and a growing marketplace of AI applications, available to hospitals inside one connected environment.
As part of that ongoing AI investment, Siemens Healthineers set out to give radiology staff a way to ask questions of their institution's own data in plain language, instead of navigating between separate dashboards to find the same answer. The teamplay UX team, including myself, took on defining how that capability should be introduced.
The brief wasn't just to design the intelligence — it was to design the relationship: how a hospital user is introduced to an AI feature they didn't ask for, how they come to trust what it does with their data, and how they stay in control of it throughout.
→ teamplay digital health platform, on siemens-healthineers.com
teamplay UX team
Where does an AI feature live in a tool people already trust?
Before any onboarding copy or consent language, we had to settle where the assistant physically sat in the product — because that choice alone changes how intrusive, or how easy to ignore, it feels.
We prototyped both, then combined them: a lightweight introduction moment for first-time visibility, resolving into the embedded panel as the assistant's permanent home — so the interruption happens once, and the tool stays quiet after that.
Consent isn't a checkbox — it's a sequence.
An AI feature that reads hospital data can't just appear. We broke onboarding into four honest moments, each answering one question a clinical user would reasonably have before saying yes.
What is this?
Plain-language introduction to the assistant's purpose and its beta status — set expectations before asking for anything.
What will it touch?
Explicit disclosure of what institutional data the assistant can access, and why — before any permission is requested.
Show me how
A short guided moment demonstrating how to ask a question, with example prompts grounded in real workflows.
Can I leave?
A visible, permanent way to disable the assistant or opt out of the beta at any time — no burial in settings.
From selling a feature to earning permission.
Early rounds
Our first explorations read like a product launch — carousel-style promotional slides selling the assistant's capabilities, with consent tacked on as a final step. It looked polished, but it front-loaded persuasion over information, which felt wrong for a tool making decisions with clinical data.
Where it landed
Later rounds flattened the promotional framing entirely. Data use, scope and consent moved to the front, stated in plain terms with real checkboxes — not styled as marketing, but as a disclosure a hospital IT or compliance reviewer would actually accept.
Contribution, stated plainly.
A defensible consent model
Turned an abstract "should we ask permission" question into a concrete, sequenced flow that legal, clinical and product stakeholders could all point to and approve.
One entry point, not three
Consolidated competing entry-point proposals into a single, testable pattern — reducing engineering ambiguity going into build.
A reusable trust pattern
The four-step onboarding structure was written to be reusable for future AI features shipping on the same platform, not just this one.
Detailed process, artefacts, and specific screens cannot be shared on public domain — this page describes the thinking, not the proprietary product. I'm happy to walk through the full case study, wireframes and design rationale in a 1:1 conversation, within the bounds of my NDA with Siemens Healthineers.