Artificial intelligence has been adopted by commercial real estate organizations at a pace that would have seemed remarkable just three years ago. Lease abstraction, underwriting support, document summarization, market analysis, the list of tasks that real estate professionals are now delegating to AI tools grows longer every quarter. But beneath the adoption numbers lies a more complicated picture. The industry is using AI extensively. It is trusting AI to make consequential decisions far less often, and the gap between those two things points toward a structural challenge that, when it is finally resolved, could fundamentally change the architecture of real estate technology itself.

“Right now, firms don’t have a proven data strategy so they don’t often trust the AI to make the decisions,” said Mike Sroka, CEO and co-founder of Dealpath, which builds deal management and pipeline software for real estate investment firms. “If AI is used to summarize a document, most clients still go back and verify that it is correct.” That verification step is ubiquitous and not irrational. AI models produce outputs that look authoritative and are frequently correct, but that are also capable of subtle errors that a human reader would catch and a downstream system would not. Until organizations have enough experience with a specific model’s behavior in a specific context to calibrate their confidence in its outputs, verifying everything is the only defensible posture.

Sroka calls what that verification requirement the verification tax. Every time a professional checks an AI-generated summary against the source document, or confirms that a model’s extraction of a key deal term is accurate, they are spending time that AI was supposed to eliminate. The theoretical time savings of AI-assisted work are partially offset by the cost of oversight, and until the verification tax comes down, the true efficiency gains are considerably smaller than the adoption numbers suggest. Reducing it requires something that most real estate organizations have not yet built: structured, reliable, consistently formatted data that AI models can work with confidently rather than tentatively. “I believe that we will get there,” Sroka said. “The models are already really good.” The limiting factor is not the model. It is the foundation the model is working on.

The investment required to build that foundation is significant, and organizations are beginning to understand that data readiness is a prerequisite for AI effectiveness rather than something that can be addressed after deployment. “There is always leverage to having a well-organized database,” said Chase Garbarino, CEO of HqO, which builds workplace experience platforms for commercial real estate occupiers. “AI promises to be able to use unstructured data but so far it still causes problems.” The promise of unstructured data processing has been one of the more seductive aspects of large language model technology, and it is partly real. But the outputs are less reliable, the errors are harder to catch, and the downstream consequences of acting on a flawed synthesis are harder to reverse than they would be with cleanly structured inputs.

The database integration challenge is where the current state of AI in real estate runs into its most practical obstacle. “Large real estate companies have always had a ton of different tech vendors,” Garbarino said. “The question is how is AI going to sit on top of all of those?” Every system of record in a real estate technology stack exposes its data through a different API, in a different schema, on a different update cadence, and with different authentication requirements.

An AI agent that needs to pull rent roll data from the property management system, lease terms from the lease management system, market comparables from a data provider, and financial performance from the asset management platform is navigating four different data architectures before it can even begin its analysis. Agentic AI systems are being designed to handle exactly this kind of multi-system orchestration, but they require careful training on each system’s specific data model and they need to be adjusted whenever any of those systems changes its schema or modifies how it structures its outputs. An update that changes how lease expiration dates are formatted in a property management system can silently break an agent that was correctly reading them before, and the error may propagate through downstream analyses before anyone catches it.

The longer-term vision is a large data lake that AI can draw on without requiring pre-structured inputs. This represents the architectural shift that would finally resolve the verification tax and the integration challenge simultaneously. The idea is that AI sophisticated enough to understand the semantics of different data formats, to recognize that a field labeled “commencement date” in one system and “lease start” in another contain the same information, and to normalize those inputs on the fly before using them for analysis, would liberate organizations from the costly work of data standardization that currently consumes so much of their AI investment. That capability would mean that the data does not need to live in any particular system in any particular format. It just needs to exist somewhere the AI can reach it. And if the AI can reach it and make sense of it regardless of format or source, then the question of which system it lives in becomes considerably less consequential than it has ever been.

That shift, if it happens, would be one of the most disruptive developments in real estate technology in a generation. The software companies that serve the commercial real estate industry have built durable businesses on a concept the industry calls stickiness: the difficulty of switching from one system of record to another once an organization has built its workflows around a platform. Property management systems, lease management platforms, and investment management tools are genuinely hard to leave. The possible disruption to operations during a transition is real enough that most organizations tolerate platforms they are dissatisfied with rather than go through the pain of moving. That switching cost is not just an operational inconvenience. It is a business model. Software vendors in real estate have been able to raise prices, slow product development, and lag on customer service in part because their customers have limited practical ability to respond by leaving.

AI that can abstract away the differences between systems would change that calculus in ways the current generation of software vendors has not had to seriously reckon with. An organization that can switch its underlying property management or lease management system by rewriting a set of AI prompts rather than by rebuilding its data architecture and retraining its staff is a fundamentally more fluid technology consumer than one that is locked in by switching costs. The stickiness that has protected incumbent vendors would erode, not because the systems themselves became easier to migrate but because the AI sitting on top of them made the underlying system interchangeable. I

f the AI can read data from any system in any format and produce the same quality of output regardless of which vendor’s database it is drawing from, then the vendor’s ability to hold customers through switching costs disappears. Real estate companies would be free to be experimental with their technology stack in ways they have never been able to afford before, testing new platforms without the fear of being locked in and abandoning ones that aren’t performing without the pain of a full migration.

That future is not imminent. The models are getting better, the integration tooling is improving, and the data lake vision is becoming more technically feasible with every generation of AI infrastructure. But the industry is still in the structured data phase, building the foundations that the next generation of AI capability will run on. The real estate technology companies are building toward a world where their value comes from the quality of their analysis and the depth of their domain expertise rather than from the difficulty of leaving them. The ones whose entire business model depends on being THE system of record should be paying very close attention to how fast the models are improving and how quickly they can switch out one database for another.

The post AI Could Eventually Make the Real Estate Industry’s Systems of Record Obsolete appeared first on Propmodo.

​