AI in Production - Healthcare: From Data to Decisions Stockholm MLOps #30, (Stockholm, Sweden | April 9, 2026)

Insights from the Stockholm MLOps community

“We track 80+ trauma variables — but the only outcome we use is death.” — Dr. Gunnar Sandersjöö, Karolinska Universitetssjukhuset

That single line captured the reality of the evening.

AI in healthcare is not limited by intelligence.
It is limited by how systems are measured, integrated, and trusted in practice.

At Stockholm MLOps #30, clinicians, engineers, and system leaders came together to discuss what actually happens when AI meets real healthcare environments.

Summary — Key Insights from the Event

  • Healthcare struggles to measure success beyond survival

  • AI fails at workflow integration, not model performance

  • Trust is built in data pipelines, not in models

  • Procurement and integration costs block most innovation

  • Data is abundant — actionable insight is not

What This Event Was Really About

Most industries use data to optimize outcomes.

Healthcare is different.

It operates as a complex adaptive system treating complex adaptive systems — patients.

“Healthcare is a complex adaptive system treating complex adaptive systems.” — Steve Mackenney, Mission 10x

The implication:

AI in healthcare is not just a technical problem.

It is:

  • a systems problem

  • a measurement problem

  • a trust problem

  • and a coordination problem

Key Insights from the Event

1. Healthcare Measures Death — Not Recovery

“We track 80+ parameters in trauma, but the only outcome we use is death.” — Dr. Gunnar Sandersjöö
“We know if patients die — we don’t know how they live after.” — Dr. Gunnar Sandersjöö

Healthcare lacks meaningful outcome measurement beyond survival.

Without measuring recovery or quality of life, AI cannot optimize what actually matters to patients.

2. AI in Healthcare Fails at Workflow, Not Intelligence

“The workflow is not A to B to C — it branches constantly.” — Dr. Gunnar Sandersjöö

Clinical environments are non-linear, dynamic, and high-stakes.

AI systems fail when they are designed outside real clinical workflows.

3. Trust Is Built in Pipelines, Not Models

“The most critical component isn’t the model — it’s the de-identification layer.” — Arnold Anderson, Region Stockholm
“You can’t just trust a box labeled ‘de-identification’ — you have to inspect it deeply.” — Arnold Anderson

Trust in healthcare AI is defined by data handling, privacy, and system design — not model accuracy.

4. AI Dies in Procurement Before Production

“It took us almost a year just to get approval for a system.” — Stefan Billton, Södersjukhuset
“If integration costs 1 million SEK and takes a year, it won’t happen.” — Stefan Billton

The biggest blocker to healthcare AI is not technology — it is procurement and integration.

5. Healthcare Has Data — But Not Signal

“We are flooded with data, but we don’t know which data matters.” — Dr. Gunnar Sandersjöö
“We measure what’s easy, not what helps the patient.” — Dr. Gunnar Sandersjöö

Healthcare systems generate massive amounts of data — but very little actionable insight.

AI’s role is to create clarity, not more data.

6. Infrastructure Matters More Than Models

“We keep data on-premise for control, but we need the cloud for compute.” — Arnold Anderson

Healthcare AI systems depend on hybrid infrastructure:

  • on-prem for control

  • cloud for scalability

Infrastructure design is part of the clinical system.

7. Outcomes Drive Everything — If You Measure Them

“Most complications are predictable — and many are avoidable.” — Dr. Graham Copeland
“Death is easy to measure — everything else is difficult.” — Dr. Graham Copeland

Healthcare outcomes can improve significantly — but only if they are measured correctly.

8. If You Can’t Express It Financially, It Doesn’t Exist

“If you can’t express impact in money, decision-makers won’t act.” — Helena Mueller, Grant Thornton
“We need to speak CFO language if we want change.” — Helena Mueller

Healthcare innovation must translate into financial impact to drive real adoption.

Patterns Across Talks

Across all speakers, a few patterns stood out:

  • Measurement is fundamentally broken

  • Trust lives in systems, not models

  • Workflow integration is the real AI challenge

  • Procurement and integration block innovation

  • Data exists, but is not operationalized

What This Means for AI in Healthcare (Sweden)

AI in healthcare is not limited by intelligence.

It is limited by:

  • measurement

  • trust

  • infrastructure

  • execution

Sweden is addressing these challenges under strict constraints:

  • regulation

  • data sovereignty

  • system complexity

This is forcing a shift:

Less focus on models.
More focus on systems that work in production.

Join the Community

👉 Full event details: https://www.meetup.com/stockholm-mlops-community/events/313881409/
👉 Explore all events: https://www.meetup.com/stockholm-mlops-community/

Event Details

  • Location: Stockholm

  • Date: April 9, 2026

  • Speakers: Karolinska Universitetssjukhuset, Södersjukhuset, Region Stockholm, Grant Thornton, Mission 10x

  • Topics: AI in healthcare, MLOps, data infrastructure, clinical systems

  • Event: Stockholm MLOps #30

Previous
Previous

Sovereign AI in Production | Stockholm MLOps Event #31 | April 23 2026

Next
Next

Inference Optimization in Production | Stockholm MLOps #29 | March 12, 2026