Last year I wrote about talking buildings. I pictured a tenant asking about air quality or a new building operator asking when a rooftop unit was last serviced, and the building answering in plain English. AI agents are reshaping property management and the real estate industry is finally figuring out the need for a common data structure underneath those agents.

But I think there’s another layer underneath that architecture that deserves attention: the physical building itself.

Instead of “what if buildings can talk,” maybe the better question is: What does the building actually know about itself?

A building may be able to tell us the temperature on the fourth floor. But does it know the verified area of that floor? Does it know which air-handling unit serves it? Does it know when the unit was installed, what has been done to it, and whether the drawing showing it is still current? More importantly, does it know which information can be trusted?

I’ve become convinced that this is one of the biggest practical issues we’ll have to solve before conversational buildings become truly useful.

Data isn’t the same as knowledge 

Commercial buildings accumulate information for decades. Drawings, leases, BOMA calculations, equipment schedules, photographs, condition assessments, work orders, energy bills, specifications, maintenance records, building automation data and now sensor data all become part of the record.

The trouble is that they were created by different people, at different times, for different reasons.

Some information is current. Some isn’t. Renovations happen without every drawing being updated. Equipment gets replaced. Tenants move. Room names change. A spreadsheet maintained by an experienced operator may contain better information than the formal record.

That was manageable when a person could walk downstairs and ask someone who had worked in the building for 20 years. It becomes a different problem when we expect an AI system to provide an answer in seconds.

AI is very good at finding information. That doesn’t automatically make the information true.

Our business has spent decades measuring existing buildings. One lesson from that work is simple: the building in front of you eventually becomes different from the building in the records.

That’s why I think building intelligence has to start with physical ground truth. Modern reality capture makes this far easier than it used to be. LiDAR can establish accurate geometry. 360-degree imagery can preserve a visual record. CAD and BIM can structure spaces and systems. Asset information can then be attached to that physical framework. The important point isn’t the scan or the 3D model. Those are tools.

The value is establishing a reliable digital baseline of what exists today and giving the rest of the building’s information somewhere logical to live.

A digital twin needs more than geometry

In some of my earlier pieces I wrote enthusiastically about digital twins and hyper-intelligent buildings. I still believe in that direction, but our practical work has made my definition of a useful twin more demanding. A 3D model isn’t enough.

Take a rooftop unit. Its shape and location are useful, but an owner really wants to know what it serves, how old it is, its service history, expected remaining life, current performance and replacement cost. The geometry gives the equipment a location. The relationships give it meaning.

The same applies to rooms, tenants, leases, energy consumption and almost every other piece of building information. An intelligent building needs to understand how those things relate to each other.

Experienced operators carry a huge amount of institutional knowledge. They know the pump that starts making a particular noise before it fails. They know which valve was changed but never updated on the drawing. They know why one part of the building always runs warm on a sunny afternoon. When those people leave, some of the building leaves with them.

A persistent digital record gives us an opportunity to change that. A renovation updates it. An equipment replacement updates it. A condition assessment enriches it. Maintenance adds history. Sensors add observations. Over time, the building develops a memory of itself.

Knowing what it doesn’t know

One capability, in particular, will separate useful building AI from a clever chatbot: uncertainty. Suppose I ask when a rooftop unit was installed. One document says 2014 and another says 2016. I don’t want AI to choose the answer that sounds most likely and present it confidently. I want it to tell me there is a conflict. Better still, I want it to flag the item for verification the next time someone is in the building.

An intelligent building shouldn’t only know things. It should know what it doesn’t know, where information came from, and when it was last verified. That’s what will make people trust it. Once that foundation exists, the conversational interface becomes much more powerful.

An asset manager can ask which rooftop units across the portfolio are approaching replacement. A facility manager can ask which equipment has generated repeated work orders. A leasing team can ask whether measured areas still match its records. An acquisition team can ask what information is missing before it buys a property. The user shouldn’t have to know which software contains the answer.

That’s where I see the real opportunity for an intelligence layer: connecting the physical building to the systems and information that already exist around it, then allowing people to interact with that knowledge naturally. One day soon the interface itself may become less important as agents work across systems on our behalf. I agree. But that makes the quality of the information underneath the interface even more important. Before a building can talk intelligently, it has to understand what it is talking about.

The post Before We Talk to Our Buildings, We Need to Teach Them What They Know appeared first on Propmodo.

​