You’re back from a showing. You glance at your phone and see three voicemails, an email asking where things stand on two sites, and a partner texting about a deal you’re both on. You know all of it. It’s in your CRM – technically – scattered across notes you typed in a parking lot, a status field someone updated six weeks ago, and a comment thread that cuts off right before the part the client is asking about. Pulling it together into a reply will take twenty minutes you don’t have.
Keeping track of the deal is its own job, on top of running the deal itself. And that’s where AI is genuinely useful right now. It’s not glitzy, but it’s true, and it’s the most impactful AI use case in CRE brokerage today.
Most CRE AI announcements are demo software
Deal probability scoring, automated underwriting, off-market opportunity surfacing: all of it is compelling in a controlled presentation and much less so on your actual deals with your actual data. Skepticism is appropriate for anyone who has watched software sold into this industry for twenty years.
AI-based software misses the mark when it pretends to know more than the broker. AI isn’t better than you, it doesn’t know what to do next, and it certainly can’t close your deals for you. General-purpose chatbots have plenty to say about commercial real estate as a subject. None of it touches your book.
In practice it looks unremarkable
An AI assistant with structured access to your deal records can tell you what happened on your deal last Thursday, who the other rep is, which follow-ups are past their review date, and what commission events are due next quarter. You describe a call in plain language, something like “spoke with the landlord rep, they’re holding firm on TI but offered to extend free rent, running it by our client before we counter,” and it lands in the activity log on the right deal with a follow-up date set.
No record to navigate to, no form to fill. You talk about your deals, and the records stay current.
And when the client emails asking where things stand, instead of twenty minutes of digging, you have the answer in ten seconds.
The same thing applies before a client call. Ask what happened on a client’s three active sites last month and you get back a summary from the actual deal log. Not what you half-remember from the drive over. What was actually recorded: who said what, where things stood, what was still unresolved. The call starts on the right foot instead of with a gap you’re hoping the client doesn’t notice.
Stale follow-ups are another version of the same problem. When you’re working 30 or 40 deals at any given time, you’ll flag things for review and then watch those dates slide by quietly, because you’re busy running deals, not auditing your own calendar. Ask which follow-up dates have passed and you get back a concrete list: the specific comment that was flagged, when it was set, and which deal it lives on. Run that check Monday morning. You’d rather be the one who noticed.
The pattern holds across all of these, and it runs deeper than follow-up reminders. Hand the AI a call summary and it logs it on the right deal. Hand it a PSA and it reads the document, pulls the effective date, the inspection period, and the closing timeline, and lays them out for you to confirm into your deal record. No form to open. In every case, the AI is doing the transfer work: moving what you know, or what a contract says, into the system where it can actually be used. That’s a narrower claim than most AI vendors are making. It’s also the one that holds up.
What doesn’t work
In my experience, today’s AI tools aren’t reliable enough to predict deal outcomes. The probability scores being sold right now are pattern-matching on whatever training data the builder happened to have. I’ve seen no evidence it beats a broker’s read of their own deals.
AI writing your client emails is marginal. Ask it to summarize what happened and it will do that accurately. Ask it to hit the right temperature with that specific client and get the details right, and you’ll spend as long fixing the draft as writing it.
Both failures come from the same root problem. An AI assistant knows exactly as much as the system it’s plugged into. Point it at your deal records and it knows your deals. Point it at a comps database and it knows comps. Point it at nothing in particular and it will still answer you, fluently and at length, out of nothing at all.
The unglamorous truth
When a tool promises AI valuation or rent analysis, ask what it’s reading, make them name the source, and ask to see it run on your deal with your numbers in front of you. A confident answer with nothing underneath it is worse than no answer, because you might act on it.
The brokers who get real value out of AI will be the ones who point it at the tedious work first, at the logging and the status tracking and the fifteen minutes of digging that stands between a client’s email and your reply.
The bigger AI promises may well arrive eventually. The boring work is what’s ready now, and the boring work is where the hours go.
