Pilot Purgatory: Everyone's Testing AI, Almost Nobody's Finishing

Pilot Purgatory: Everyone's Testing AI, Almost Nobody's Finishing

JLL surveyed 500+ senior real estate decision-makers across 15 markets last fall, and the results are the most honest thing published about AI in this industry: roughly 90% are running AI pilots, and only 5% say they've achieved all their AI goals. Meanwhile 87% are raising tech budgets specifically to buy more of it. Put together, that's an industry sprinting into a room it hasn't figured out how to furnish.

The money agrees. Proptech venture funding hit $16.7B in 2025, up 68% and past the old 2019 peak. But $11.2B of that went into $100M-plus mega-rounds, and 31 companies captured about 72% of the total. Capital is piling into a handful of winners while everyone else runs pilots that never graduate.

Where it actually works right now

Two places, mostly.

Leasing and property management. EliseAI raised a $250M round in August at a $2.2B valuation and now runs the AI leasing agent for roughly 1 in 10 U.S. apartments — Brookfield, AvalonBay, and Equity Residential are on the roster. Tours, lease audits, rent collection, maintenance triage. This is the clearest "it saves real hours" story in the sector.

Feasibility and entitlement — the part that matters if you develop in LA. Tools like TestFit generate massing, parking, and unit-mix scenarios straight from zoning constraints in real time; Archistar screens plans for code compliance before you burn a cycle with the city. Not magic — but it collapses the "can this site even pencil?" question from a week to an afternoon.

> 5% — the share of real estate firms that say AI has actually delivered everything they hoped. The other 95% are still in the pilot.

The cautionary tale

AI pricing just hit a wall. In November, the DOJ settled its case against RealPage: no more real-time lease data feeding the algorithm, inputs must be at least 12 months old, and no sub-market geographic modeling. Whatever you think of the case, the lesson for operators is concrete — "let the model set the rent" is now a legal question, not just a software feature. If your AI touches pricing, your compliance team just joined the product team.

The honest read

The bottleneck isn't the models. It's your data. JLL's own finding is that 60%+ of firms aren't technically ready to scale — messy rent rolls, PDFs nobody parsed, five systems that don't talk. The firms getting value aren't the ones with the fanciest AI; they're the ones that cleaned up a single workflow, pointed a tool at it, and shipped.

So skip the moonshot. Take the most painful, most repetitive thing your team does — lease abstraction, feasibility screens, comp pulls — get that data clean, and deploy one thing all the way. Finishing one pilot beats starting ten.

Sources