AI in Commercial Real Estate: Why 92% Pilot, Few Deliver

In 2025, JLL found that 92% of commercial real estate teams were planning AI pilots, up from under 5% in 2023. In that same research, only 28% of facilities-management organisations had actively embedded AI solutions into their operations. That 64-point gap is not a technology problem. It is an integration problem, and it lives almost entirely inside building operations.

Property and facilities leaders are under pressure to show AI results in 2026 and present a credible roadmap for 2027. The macro adoption numbers are real, but they describe intent, not execution. The firms that are closing the gap are not doing so by choosing a better model. They are doing so by connecting AI to the specific systems, queues, and workflows where property managers and field technicians spend their days: tenant helpdesks, maintenance ticket routing, work-order management, and asset data feeds. Everything else is a pilot that never graduates.

AI in Commercial Real Estate: The Adoption Numbers Hide an Operational Deficit

The headline figures look like momentum. 92% piloting. Adoption in property management reportedly jumping year over year. But the operational reality is more instructive. JLL's 2025 research showed that 54% of CRE teams cited compatibility with legacy infrastructure as the leading barrier to AI deployment. That figure explains more about the state of the industry than any adoption percentage does. Firms are experimenting at the edge while their core operating systems, work-order platforms, tenant communication tools, and building management systems remain disconnected from any AI layer.

The consequence shows up in a separate JLL finding published in August 2026: only 15% of surveyed CRE leaders had reached the optimising stage of AI adoption, while 78% expected AI to drive significant changes to their portfolio strategy. Strategic ambition is running roughly five times ahead of operational maturity. That ratio is not sustainable heading into 2027 budget cycles.

Among the facilities-management organisations that had embedded AI, the leading use cases were work-order management at 57% and asset-lifecycle insights at 54%. Those two numbers matter because they point directly at the integration layer that separates a working AI deployment from a stalled pilot. Work-order management is not glamorous. It is also where the volume is. A mid-size commercial property portfolio can generate hundreds of maintenance requests per week across HVAC, access control, cleaning, and compliance. Routing those tickets manually, or through a rule-based system that cannot learn, is where time and tenant satisfaction both erode.

Where the Integration Gap Actually Lives in Building Operations

Three integration points account for most of the operational gap in CRE building management. The first is tenant intake. Most property management teams still handle tenant requests across a mix of email, phone, WhatsApp, and portal submissions. Each channel creates a separate queue. An AI agent that sits across all of those channels, classifies the request, checks the asset register, and routes to the right technician or contractor does not require a new model. It requires a connection to the systems that already hold the data: the CMMS, the lease management platform, and the communication history. The model is the easy part. The connective tissue is the work.

The second integration point is maintenance ticket flow. A ticket that arrives at 22:00 on a Friday for a heating fault in a multi-tenant office building has a cost profile that changes dramatically depending on how fast it is classified and escalated. An AI agent that can read the incoming message, match it to the asset record, check the maintenance history, identify whether the fault falls under a service-level agreement, and create a prioritised work order in the CMMS before a human reviews it in the morning is not science fiction in 2026. It is a configuration and integration project. The firms that have done it report measurable reductions in mean time to resolution. The firms that have not done it are still routing tickets by hand.

The third integration point is the building systems layer itself. Deloitte's 2026 infrastructure research found that 44% of infrastructure organisations planned to use AI for energy-demand forecasting over the following three years, and 38% planned AI-enabled workforce optimisation. Both use cases depend on connected asset data. A building management system that does not expose its sensor data to an AI layer cannot support demand forecasting. A workforce optimisation model that cannot read the work-order queue cannot schedule field technicians efficiently. The model sits idle until the data pipeline exists.

Why Roles Matter as Much as Systems

JLL's July 2026 research on workforce change found that 60% of senior business leaders expected AI to reinvent human roles rather than replace them. In building operations, that distinction is operationally significant. A property manager whose job currently involves triaging tenant emails, manually creating work orders, and chasing contractor confirmations is not going to be replaced by an AI agent. But the shape of that role changes substantially when the agent handles intake, classification, and routing. The property manager shifts toward exception handling, relationship management, and quality review. That shift requires deliberate workflow redesign, not just a new software subscription.

The same JLL 2026 research noted that only 31% of CRE leaders were actively preparing to redesign spaces and workflows for human-AI collaboration. That figure suggests that most firms are deploying AI tools without redesigning the processes those tools are meant to support. The result is predictable: the tool gets used for the easy cases, the hard cases still go through the old process, and the efficiency gains never materialise at scale. Workflow redesign is not a change management afterthought. It is a prerequisite for operational AI that actually delivers.

Deloitte's March 2026 AI infrastructure survey found that 64% of large enterprises had begun limited or scaled AI-factory deployments, with 88% expecting to do so by 2028. In CRE terms, an AI factory is the infrastructure layer that allows AI agents to run close to operational data: the lease database, the asset register, the work-order system, the tenant communication history. Without that infrastructure, AI in building operations remains a demo. With it, the integration points described above become buildable in weeks rather than quarters.

What Operational Integration Actually Requires in 2027

For property and facilities leaders planning 2027 programmes, the practical question is not which AI model to use. It is which three integration points to connect first, and in what order. The evidence from the firms that have moved past the pilot stage points to a consistent sequence. Start with tenant intake because it is the highest-volume, lowest-risk integration point. A multi-channel intake agent that classifies and routes requests does not need access to financial systems or lease data to deliver value. It needs a connection to the communication channels and the work-order system. That connection is achievable in a short implementation cycle and produces measurable output from day one.

The second step is maintenance ticket automation inside the CMMS. Once intake is connected, the ticket data starts to accumulate in a structured form that supports routing rules, priority scoring, and SLA tracking. An AI agent that operates inside that data layer can begin to handle escalation logic, contractor assignment, and status updates without human intervention for the majority of standard requests. The property manager reviews exceptions, not every ticket.

The third step is connecting building system data to the operational AI layer. This is the most technically complex step and the one that requires the most coordination with facilities engineering and IT. But it is also the step that unlocks the energy and workforce optimisation use cases that the Deloitte infrastructure data points toward. Firms that complete all three steps have a defensible operational AI programme. Firms that stop at step one have a better helpdesk.

The 92% pilot figure from JLL's 2025 research is not a sign that the industry has solved AI adoption. It is a sign that the industry has solved the easy part: getting a model running in a controlled environment. The hard part is the integration work that connects that model to the systems, roles, and workflows where building operations actually happen. That work is not glamorous, it is not fast, and it does not generate press releases. It is also the only work that closes the gap between a pilot and a programme that delivers.