There is a version of the AI-in-real-estate conversation that focuses entirely on what agents can do once they are running: optimising energy, detecting faults, answering portfolio-level questions in seconds. That version skips the harder question — what has to be true about your buildings before any of that is possible.
The answer is onboarding. And in 2025, the story of onboarding has changed significantly.
What onboarding actually means
In proptech, “onboarding” is an overloaded word. For some platforms it means getting a user account. For ProptechOS, it means something more specific and more consequential: taking raw data from the systems inside a building — BMS, FM, IoT, metering, ERP — and modelling it into a structured, standardised format called Real Estate Core.
The output is a digital twin. Not a single twin for the building, but individual digital twins for every sensor, device, asset, and space inside it. Each one positioned, labelled, and linked to its physical location — so that when a sensor detects an anomaly, an agent does not just know the reading; it knows the floor, the zone, the system it belongs to, and what other data points are nearby.
This level of structure is what makes AI-driven analysis possible. Without it, data from a Schneider system in Stockholm and an MOA system in Oslo is incomparable. With it, they become part of the same standardised portfolio view — regardless of vendor, country, or how old the underlying system is.
The technical shift: from manual to automated
For most of ProptechOS’s existence, building onboarding was skilled, manual work. The first building the team onboarded took close to a month. Tools improved, and that timeline came down to days — but it remained labour-intensive, driven by engineers working through thousands of data points one by one.
The turning point was the development of Systema Machinae, an AI model trained on millions of manually modelled building data tags collected since 2018. Rather than hard-coded rules, the model works with what is called an embedding space: a 1,536-dimensional map of how building data points relate to each other, built from real examples across buildings on five continents.
When a new tag list arrives, Systema Machinae places each tag into that space and finds its nearest known neighbours. A tag close to millions of known supply air temperature sensors gets a high-confidence mapping. One that sits in unfamiliar territory gets flagged for human review. The result: a process that previously took weeks now takes minutes, with human attention reserved for the genuinely ambiguous cases.
“Once it is modelled in Real Estate Core, it becomes completely comparable — regardless of what the underlying data is, whether it is a Schneider system or a Bacnet system or a system no one has touched for 25 years.” — Per Karlberg, CEO, ProptechOS
In a live demonstration during the webinar, a tag list of roughly 3,500 points from a system the model had never seen before was processed end-to-end. Around 2,800 tags were mapped automatically. Approximately 600 were flagged for review — a targeted subset, not the full list.
What the process looks like now
The deeper change is not just that Systema Machinae is faster. It is that onboarding has become agentic — the model is now a tool that AI agents call, rather than a UI that engineers operate.
A ProptechOS onboarding agent receives a task, a tag list, and a reference to a project management ticket. It checks what already exists in the system, runs Systema Machinae, applies its own reasoning to low-confidence tags, models everything into Real Estate Core, validates the output, and logs the result — without a human touching each step.
In the webinar’s demonstration, this process ran in under 17 minutes for a building with more than 24 rooms and multiple system types. The agent worked through a 16-step process autonomously, creating and validating digital twins as it went.
The practical implication: a portfolio of 17 buildings with a homogeneous BMS setup was onboarded in approximately three hours. A more complex portfolio of 15 buildings — with two systems per building and several BIM models — was completed in three days.
Where the real friction still lives
With the modelling work largely automated, the main source of delay has moved upstream. The constraint is now access: access to the building systems, access to the network infrastructure, access to the right person at the system vendor.
“What used to be 80% onboarding work and 20% working with partners and customers has just shifted to 95% working with getting access and understanding what should be done.” — Per Karlberg, CEO, ProptechOS
In practice, this means property owners and asset managers need to think about onboarding differently. The technical work has been abstracted away. What remains — and what determines how quickly a portfolio can become agent-ready — is an organisational question. Who has the credentials? Who manages the VPN? Who is the contact at the BMS vendor?
Teams that have answers to these questions before onboarding begins move quickly. Those that encounter them mid-project do not.
The implication for real estate leaders is straightforward: readiness for AI in real estate is not primarily a technology procurement decision. It is a data access decision, and it involves stakeholders across facilities management, IT, and vendor relationships.
The security layer that cannot be skipped
One aspect of the webinar’s demonstration is worth highlighting specifically: the agent that performed the onboarding had no standing access to ProptechOS before the task began. It requested precisely the permissions needed to onboard digital twins to a specific building, pointed to the project management ticket as justification, and had those permissions revoked at the end of the process.
This is called zero standing privileges, and it matters. As AI agents take on more operational tasks in real estate — not just onboarding, but fault detection, energy optimisation, lease management — the question of what each agent can access, and under what conditions, becomes a governance question that portfolio owners are accountable for.
“You are always still responsible for what your agents are doing. That part is not automated.” — Per Karlberg, CEO, ProptechOS
The capability to act is only part of the equation. The policy framework around that capability is equally necessary.
What this means for real estate leaders
The gap between “we have building data” and “our buildings are agent-ready” is narrowing. The modelling work that once took months can now be completed in hours or days. The AI tooling exists, and it is improving rapidly.
What has not changed is the requirement for data to be connected, accessible, and clean before it can be useful. The buildings that will generate value from agentic operations are the ones whose data is structured, validated, and integrated — not as a future project, but now.
For real estate organisations evaluating proptech investment, the most useful question to ask is not which AI agent to deploy. It is: how quickly can we make our buildings readable?
Key takeaways
- AI agents can only perform operational work on structured, validated building data — onboarding is what creates that foundation
- ProptechOS’s Systema Machinae model, trained on millions of real building data tags, automates the bulk of the modelling work that once took weeks
- A portfolio of 17 buildings was onboarded in approximately three hours; a more complex 15-building portfolio in three days
- The main source of delay in onboarding is now access — system credentials, network infrastructure, vendor contacts — not technical modelling
- AI agents operating within real estate systems require clear permission frameworks; zero standing privileges is the standard ProptechOS applies
