Six months in: how Locum is discovering what AI agents can do in Stockholm’s hospitals
Customer overview
Locum manages and develops healthcare properties on behalf of Region Stockholm — including major acute care hospitals and specialist facilities. With around 300 employees and approximately 800 projects per year, Locum is responsible for ensuring Stockholm’s hospitals and healthcare facilities have functional, sustainable environments around the clock.
Any operational gap — a missed alarm, a delayed work order, an undetected fault — doesn’t just affect a building. It affects patient care. That context shapes everything about how Locum approaches technology adoption: carefully, methodically, and always with safety first.
The challenge
Locum had data. What they lacked was a unified way to see it, act on it, and share it — in real time.
Their property systems — SCADA, facility management platforms, sensor networks — each operated in isolation, generating valuable information that was difficult to correlate or consume across systems. Getting a cross-system picture required manual effort and was rarely possible in real time.
The alarm situation illustrated the problem starkly. Hospitals generate a huge amount of alarms every day. Thomas Ahlberg, Technical Director at Locum, describes their alarm dashboard as extensive and unmanageable over time – an overwhelming backlog that facility technicians struggle to sort through manually. When three faults occur simultaneously, the system has no way to surface that correlation and suggest what actually needs fixing.
Beyond alarms, Locum also faced a subtler but critical challenge: when data stopped flowing from sensors, nobody knew for days. Decisions were being made on stale or absent data without anyone realising it.
The goal was clear: connect all existing systems into a single platform, establish data quality, and use that foundation to work proactively — resolving issues before tenants or clinical staff were even aware of them.
“The AI agent environments came sneaking in, and it has gone incredibly fast over the last three to four months. In November we hadn’t tested anything”
“We had the data before, but it was segmented across different systems. Now we have something that ties it all together so we can use it in a completely different way — working proactively.”
Technical Manager
Solution
Rather than replacing existing systems — a path Locum explicitly ruled out as too expensive and time-consuming — ProptechOS became the connective layer that gives all their systems a common language and shared environment.
The project was structured around approximately 10 defined user stories to verify the platform could support real operational workflows. One fully validated example: sensors in the ProptechOS environment now trigger work orders directly in Locum’s facility management system, Faciliate. The complete feedback loop — from alarm detection to completed action and feedback — has been proven end-to-end.
The sensor foundation is deliberately simple: presence and lighting, temperature, humidity, and pressure in aggregates. Four parameters. But as Thomas puts it, you can do an enormous amount with just those.
The pilot — and the surprise
When Locum started the project, AI and automation were not the focus. The priority was getting data to flow cleanly and reliably. But something unexpected happened.
“The AI agent environments came sneaking in, and it has gone incredibly fast over the last three to four months. In November we hadn’t tested anything.” — Thomas Ahlberg, Technical Manager, Locum
What began as a data integration project rapidly became a live test of agentic AI in one of the most demanding built environments imaginable. During the pilot, Locum will be working with AI agents across four areas:
- Energy and indoor climate.
Energy strategists have gone into testing AI agents to understand how their systems are optimised — including attaching screenshots of aggregate-level data for AI to interpret and suggest actions on. Thomas calls it a fantastic tool, and sees major gains ahead from proactive energy and indoor climate management. - Alarm prioritisation.
With an overwhelming volume of daily alarms impossible to manually triage, AI agents can analyse multiple alarms simultaneously — identifying when several faults occur together, correlating them, and surfacing what actually needs urgent attention. The goal: turn an unmanageable backlog into a prioritised, actionable list. - Maintenance and operations planning.
AI agents can continuously analyse property data to give Locum better grounds for prioritising maintenance — supporting a shift from reactive fault-fixing to preventive upkeep. Operations managers, property managers, and technical managers all stand to benefit from this layer of decision support. - Data flow integrity.
When sensor data stops flowing, AI agents can detect the gap and alert technicians — before it creates an operational blind spot that might otherwise go unnoticed for days.
The project was structured around approximately 10 defined user stories to verify the platform could support real operational workflows. One fully validated example: sensors in the ProptechOS environment now trigger work orders directly in Locum’s facility management system, Faciliate. The complete feedback loop — from alarm detection to completed action and feedback — has been proven end-to-end.
The sensor foundation is deliberately simple: presence and lighting, temperature, humidity, and pressure in aggregates. Four parameters. But as Thomas puts it, you can do an enormous amount with just those.
On safety by design:
Given the critical nature of hospital environments, Locum has been deliberate in defining the boundaries of AI use. Active AI-driven building control is being tested during the pilot — but the vendor itself concludes that live hospital and operating environments are too complex for AI-driven regulation at this stage. AI is used for analysis and decision support, not direct control.
What has changed is the depth of that decision support. Over six months, Locum’s technical facility developers have run AI agents in parallel with human consultants — comparing their analyses of deviations, recommended reconfigurations, and proposed work orders. The output was sufficiently similar to warrant the next step: establishing a permanent organisation of AI agents to monitor and analyse technical building operations, and generate work orders directly into the production FM system.
The model remains human-reviewed. Currently, all AI-generated work orders are evaluated and approved before being assigned to an operator — whether in-house, a consultant, or an Executor Agent. As confidence grows, the scope of autonomous action can expand. The threshold for that expansion is earned, not assumed.
Planning for a structured agent organisation
The agentic operations model they are planning for at Danderyds Sjukhus Emergency Care Building is not a single AI tool — it is a structured team of agents with defined roles and responsibilities.
Four Analyst Agents — covering heating, cooling, ventilation, and electricity — monitor building systems daily against 85 KPIs derived from 113 data points, drawing on 200 operational commissioning specifications (the “Driftkort”) and years of FM operations history including work orders, error reports, and alarms. Each analyst is prompted to behave as a domain expert: knowing what to monitor, what the ideal state looks like, and what deviations matter.
A Supervisor Agent reviews, quality-assures, and filters the analysts’ output before it reaches the FM system — providing a layer of coherence and catching noise before it generates unnecessary work.
Work orders that don’t require a physical site visit are handled by an Executor Agent, closing the loop from detection to resolution without manual handoff.
The result is a continuous operational intelligence layer — one that doesn’t sleep, doesn’t have a backlog, and doesn’t require a consultant to be scheduled before an issue gets looked at.
“One of the most important things in this pilot is being able to resolve faults before the customer has even discovered they exist.”
Technical Manager
Results & value — the early findings
The pilot is ongoing, with full evaluation planned for late 2026. Even at this early stage, what Locum has test-driven has shifted their expectations significantly.
Speed that changes expectations.
The dashboard that now gives Locum’s leadership real-time property data across their portfolio — Thomas built the initial prototype in 2 minutes. It was then refined and published as an app within ProptechOS. Today, logging in and opening live data takes under 90 seconds. He demonstrated it to leadership the same day in February 2026.
Specialist output, in a fraction of the time.
Using AI tools including Copilot, one of Locum’s specialists now produces regulatory inspection reports in 4 hours — work that previously required a 2-day consultant engagement, plus the lead time of ordering and waiting for the work. The expectation is that AI agents will eventually bring this further still. As Thomas notes: “Then we don’t need the framework agreement next time.”
Integration speed transformed.
Across the project, building ProptechOS integrations is now happening approximately 60 times faster than a year ago. A two-way integration against a new system was completed in 3 hours — with no human writing the code, only reviewing and guiding it.
A foundation ready to build on.
The agentic AI developments of the past 3–4 months have, in Thomas’s view, demonstrated that scaling to all acute care hospitals in Stockholm is a realistic ambition — not a distant one. The data foundation is in place. The next challenge is organisational.
Tenants who want more.
Locum has shown tenants a portion of what is now possible via the AI agent dashboard. The response has been strongly positive — and on data they already had.
“Building a system from scratch, from zero, would cost too much and take too long. Here we’ve found a good way to leverage the systems we already have.”
Technical Manager
Why it matters
Thomas is candid about the bigger picture. When Locum first looked at digital twins seven, eight, nine years ago, vendors would promise everything and demonstrate nothing. Nobody could show a concrete example that was actually working for a real customer.
That has changed — and Locum is now on the right side of that line.
The pace of change is the part that gives even Thomas pause. If the last four months of agentic AI development continue at the same rate, the question of where things stand in two years is genuinely hard to answer. What’s clear is that organisations that have built a clean data foundation — as Locum has — will be positioned to take advantage of each wave as it arrives.
The deeper organisational challenge Thomas identifies is one every property company will face: it’s not the technology that will be the problem. The technology works, if the data is good. The harder work is helping teams, managers, and service contractors adopt new ways of working. The role of the facility technician is already shifting — from someone with a wrench in their back pocket to someone who reads data, interprets it, and acts on AI-generated recommendations. The whole industry is in the middle of that transition.
Next steps
Locum is planning a full evaluation of the pilot in late 2026, followed by a formal procurement process. The ambition is to scale the solution to all acute care hospitals in Stockholm. Thomas is direct about the expectation: he would be very surprised if the agentic AI developments of the past few months don’t prove to be the deciding factor that makes that rollout viable.
That has changed — and Locum is now on the right side of that line.
The pace of change is the part that gives even Thomas pause. If the last four months of agentic AI development continue at the same rate, the question of where things stand in two years is genuinely hard to answer. What’s clear is that organisations that have built a clean data foundation — as Locum has — will be positioned to take advantage of each wave as it arrives.
The deeper organisational challenge Thomas identifies is one every property company will face: it’s not the technology that will be the problem. The technology works, if the data is good. The harder work is helping teams, managers, and service contractors adopt new ways of working. The role of the facility technician is already shifting — from someone with a wrench in their back pocket to someone who reads data, interprets it, and acts on AI-generated recommendations. The whole industry is in the middle of that transition.
Ready when your data is.
ProptechOS connects what you already have — and puts agents to work on it.