AI Is Finally Cracking Commercial Real Estate. Here Is What Changed.

CRE was the industry AI forgot

Residential got Zestimate in 2006. Commercial got spreadsheets.

For a decade, CRE data lived in too many formats. Every deal had its own quirks. The old guard had no reason to change how they worked. So AI companies went where the data was cleaner and the checks were easier.

That flipped in late 2025. A few startups shipped tools that understand commercial leases, cap rates, and rent rolls as structured data, not flat text. Not perfectly. But well enough that the early adopters are getting mean looks from competitors who waited.

What broke the logjam

Three things, all at once.

One: the models got better at reading documents. A 200-page commercial lease used to be a nightmare for AI. Too many cross-references, too much legal nesting. The new generation parses it without making up the square footage. That was the table stakes problem and it got solved.

Two: a couple of big REITs put AI adoption numbers in their earnings calls. Real numbers, attributed to margin. That lit a fire. Nothing moves commercial real estate like a competitor's margin improvement.

Three: inference costs cratered. You no longer need a seven-figure tech budget to run document AI at scale. Mid-market firms can afford it. That changed the addressable market overnight.

What is working right now

Lease abstraction is the one to watch. The old way: a paralegal spends three days pulling key dates, renewal options, and expense clauses from a stack of leases. The new way: AI does it in minutes, the paralegal spot-checks the weird ones. Tools like LeaseUp and Prophia are leading here. The error rate on routine leases is now below what a tired human produces on their fifth review of the day. I have seen the output. It is faster than any human at the same accuracy level.

Portfolio optimization is the second real use case. Platforms like Reonomy and CompStak layer AI over market data to flag mispriced assets, upcoming vacancy exposure, and submarket trends that have not hit the quarterly reports yet. If you are running a portfolio of 20+ properties, the stuff you miss is costing you more than the software.

Deal sourcing is getting sharper too. Instead of brokers dialing every owner in a zip code, AI looks for distress signals, loan maturity walls, and ownership changes that correlate with a willingness to sell. It is not a crystal ball. It is a filter that cuts the noise by 80%. The phone calls still have to happen.

Do not expect residential levels of automation

A 50-unit apartment complex and a single-family ranch are different animals. The data is messier, the deals are more bespoke, and the stakes per transaction are an order of magnitude higher. CRE will never be as automated as residential. Anyone telling you otherwise is selling something.

But the gap is closing. The firms that figure out lease abstraction and portfolio AI right now are building a lead that will be expensive to close in three years. I am watching a few mid-sized operators in the Midwest do exactly this. They are not announcing it at conferences. They are just getting faster while everyone else is still reading whitepapers.