JLL surveyed more than 1,500 senior commercial real estate decision-makers for its 2025 Global Real Estate Technology Survey. The headline finding is uncomfortable: 92 percent of CRE teams have started piloting AI, and only 5 percent report achieving most of their program goals. The industry has no shortage of ambition or investment. It has a systems integration problem.

The Gap Is Not About the AI

The most common explanation for failed AI pilots is that the model underdelivered: it hallucinated, its predictions were inaccurate, or it misunderstood industry terminology. These complaints exist, but they are not why 87 percent of CRE AI pilots fall short of their goals.

According to a 2026 survey from Building Engines, a JLL company, conducted with BOMA International across more than 350 commercial real estate professionals, 45 percent of respondents now understand how AI supports property management, double the prior year's figure. Only 28 percent have actually implemented AI in building operations. The implementation gap is not knowledge. It is infrastructure.

Commercial real estate operations run on layered systems that were never designed to communicate with each other. Building management systems from vendors like Siemens, Johnson Controls, or Honeywell control HVAC, lighting, and access. Property management platforms track leases, work orders, and financials. Tenant experience tools handle communications and service requests. Energy monitoring systems collect utility data. These systems often use different data models, different APIs, and different update frequencies, held together by custom connectors built years ago by teams that no longer maintain them.

AI can work across all of this data. But first, someone has to make it accessible.

Where AI Is Actually Delivering in CRE

The use cases where commercial real estate AI has achieved measurable outcomes cluster around workflows with clean, high-frequency data.

Energy management is the clearest success story. According to analysis compiled by AI Building Tools, AI building management software that continuously models occupancy patterns, weather forecasts, and utility rates typically achieves fifteen to thirty percent energy savings through automated HVAC and lighting optimization. The data streams are consistent. The optimization loop is closed. The return on investment shows up on a monthly utility bill.

Maintenance is the second clear winner. Predictive maintenance models trained on equipment sensor data can identify failure signatures before they become emergency repairs, reducing downtime and extending asset life. The tenant communication workload is another significant opportunity. According to the Building Engines and BOMA survey, more than half of property management teams spend five or more hours each week handling tenant service requests delivered by email. AI routing and triage, even without full automation, cuts that time meaningfully.

The pilot failures cluster around a different set of workflows: portfolio-level forecasting, lease negotiation automation, and cross-property tenant behavior prediction. What these have in common is a requirement for unified data from multiple systems that were never connected. The AI has no clean input to work with, so it cannot produce reliable output.

The Software Stack That Actually Works

The CRE organizations reporting AI success share a common architecture pattern. They invested in a data unification layer before they evaluated AI applications.

The unification layer ingests from every building system, normalizes the data into a consistent schema, and exposes it to AI models and analytics tools through a single interface. Commercial platforms like ProptechOS, Honeywell Forge Connected Solutions, and Willow Inc provide this layer as a managed product. Custom implementations built on BACnet, MQTT, and REST APIs are also viable for organizations with in-house development capacity.

Once that data layer exists, AI applications can be added incrementally. Energy optimization first, because the data is clean and the ROI is fast. Predictive maintenance second, because sensor data is high-frequency and the failure labels are learnable. Tenant communication triage third, because email volume is high and routing rules are definable. Each layer builds on the same unified data foundation without requiring another integration project.

The organizations stuck in pilot purgatory typically skipped this sequence. They evaluated an AI vendor, ran a successful pilot against a single building's data export, then discovered that scaling to the portfolio required integrating every system they had not connected yet. The pilot proved the AI works. It did not prove the data infrastructure works.

What To Do About It

1. Map your building data stack before evaluating AI vendors. Document every system that generates operational data: building management, property management, work order, energy, access control. Identify which systems have APIs, which require manual exports, and which have no external data access at all. That map is your AI readiness assessment.

2. Prioritize a unified operational data layer. Before committing to any AI application, decide whether you need to build a data unification layer first. Commercial platforms reduce that investment significantly for larger portfolios. For smaller portfolios, a lightweight integration built on MQTT or REST may be sufficient to start.

3. Start AI deployments with closed-loop workflows. Energy management and maintenance prediction are closed loops: the AI generates a recommendation, the building system executes it, and the result feeds back into the model. Closed loops are easier to validate, easier to measure, and easier to explain to stakeholders than open-ended forecasting models.

4. Build portfolio-level ROI into your measurement framework from day one. The 28 percent implementation rate partly reflects difficulty demonstrating that a single-building pilot scales. If your business case is built on per-building results, you will restart the justification process every time you expand.

HRIM's Take

The CRE industry's AI execution gap is a data infrastructure problem wearing an AI label. The models are capable. The use cases are validated. According to McKinsey, AI could generate between $110 billion and $180 billion in value for real estate. The organizations that capture that value are not going to be the ones that signed the most AI vendor agreements. They are going to be the ones that spent 2026 building the unified operational data layer that makes AI actionable across a real portfolio. The pilot is not the product. The integration is.