AI Won’t Fix How CRE Makes Decisions. It’s Going to Expose It.
88% of commercial real estate investors and owners have started AI pilots, per JLL's global technology survey of more than 1,500 senior decision-makers. Only 5% say they're achieving most of their AI goals. Everyone bought the tools. Almost nobody is getting the outcomes.
Propmodo put a name on the gap this week. In a July 8 piece, editor Franco Faraudo wrote about "decision architecture," the discipline of mapping how decisions actually get made inside an organization before you hand any of them to AI. What counts as a decision. What information feeds it. Who signs off. What runs on autopilot and what gets escalated. As consultant Vincent Dormedy put it in the piece: "Once the decision architecture is created, it prepares an organization for an agentic world."
What's Actually Happening
CRE has made its most important calls the same way for a century: relationship, intuition, and market feel. That worked because humans are good at deciding with fuzzy inputs. AI agents are not. The moment you ask an agent to prioritize leasing outreach or approve an amenity spend, everything implicit has to become explicit. And when firms actually do that mapping, they hit something uncomfortable fast: most of their decision inputs are averages.
Here's why that's a problem, straight from our own portfolio.
Booking volume across our buildings looks about as stable as a metric can look right now. For eight straight weeks, total weekly bookings stayed within roughly 7% of the summer average. A portfolio dashboard would tell you nothing is happening.
Building by building, plenty is happening. Over the last 30 days versus the prior 30, some buildings grew bookings more than 30%. Others in the same portfolio, in the same window, fell 40% or more. A few dropped by well over half. The flat portfolio line isn't stability. It's two opposing moves canceling each other out.
Feed the average to an agent, or to an asset manager, and the rational answer is "do nothing." Feed it the building-level behavior and you get a completely different set of decisions. Double down where demand is climbing. Get someone on a plane to the building that fell off a cliff. Ask why the fastest-growing building is growing (then copy it).
Why It Matters
Leesman's survey of corporate real estate leaders found the same pattern one level up. As of last year, 97% of organizations run hybrid work models. About a third of leaders are confident they've landed on the right approach. Near-universal adoption, minority confidence. Same shape as JLL's AI numbers: 88% piloting, 5% satisfied.
The industry keeps adopting things faster than it builds the decision inputs to manage them. Hybrid first, AI now. The tool was never the constraint. The inputs were.
What to Do
Don't start with the software. Pick three decisions that actually move NOI: which tenants get proactive renewal attention, where amenity capital goes next year, which building gets an operator's time this quarter. For each one, write down what information feeds it today. If the honest answer is portfolio averages, quarterly reports, and whoever spoke last in the meeting, congratulations, you found your AI strategy. It has nothing to do with buying software yet.
Then replace the averages with behavior. Bookings, access, spend, service requests, at the building level, weekly. That's the input layer an agent can act on. Conveniently, it's also the input layer a human asset manager should have been getting all along.
The firms in that 5% didn't out-spend anyone. They mapped their decisions, threw out the averages, and gave the machines something real to work with.
AI isn't going to fix how CRE makes decisions. It's going to expose it. The portfolio average goes first.