The 7 Things CRE Firms Are Actually Stuck On With AI (From Real Conversations) | NextAutomation

If you have been quietly wondering whether the thing you are stuck on with AI is just you, this piece should settle it. It is not just you. Across months of real conversations with commercial real estate operators, the same seven blockers came up again and again, in almost the same words. This is the ranked list of what CRE firms are actually stuck on, straight from those conversations.

A word on where this comes from, because it changes how you should read it. This is not a survey and we are not going to dress it up as one. It is two things: de-identified themes from real recorded operator calls, and 150 messages that came in through our website chat across 70 sessions. Every count below is a floor. When we say a stuck-point came up in 13 calls, we mean 13 distinct conversations where an operator raised it explicitly, not an estimate of how many people quietly worry about it, so the real number is higher. We publish the floor because in a category this crowded with inflated figures, the number you can actually stand behind is the more useful one. And we keep the two corpora separate: we never sum a call count and a chat count, because that would invent a total that does not exist. Where a stuck-point shows up in both, we report both, side by side. Every quote below is stripped of anything that could identify the operator or their firm.

Here are the seven, ranked by how loudly and how often operators raised them.

1. Finding the Deal Before It Is Competitive (Off-Market Sourcing)

The pain, in an operator's words: across our recorded calls, the single most common thing operators want AI for is not underwriting. It is sourcing. Finding the deal, and finding the owner, before the property hits the market and the price gets competitive. This came up in 13 distinct real conversations, more than any other theme, and it showed up again in the site chat as deal-sourcing intent in 13 of the 150 messages.

Why firms get stuck here: most operators know exactly which owners they would love to reach. What they do not have is a repeatable way to find those owners at scale and get in front of them first. General-purpose AI is great at drafting and summarizing, but the operator is still the one manually hunting for who owns what, and that manual hunt does not scale past a handful of targets a week. The demand is loud, and the do-it-yourself version stalls fast.

The honest first move: pick one narrow slice of your buy box, one asset type in one market you know cold, and get precise about what an ideal owner-to-contact looks like there. Sourcing is not one problem, it is a chain of small ones (find the owner, verify the contact, reach out, follow up), and the operators who make progress define the target tightly before they try to automate the chain. For the fuller read on where operator demand is actually pointing, see the state of AI in commercial real estate 2026.

2. "How Does This Actually Land in MY Business?" (Delivery)

The pain, in an operator's words: "how does your workflow get into my tool?" This is the largest single theme in our entire site-chat corpus, 38 of the 150 messages, more than pricing and more than any objection. Operators are not asking whether AI works. They are asking how it lands in their shop, on their files, in the tools they already use.

Why firms get stuck here: most AI content shows you a slick demo and stops. It never answers the mechanical question every operator actually has: once this thing works, how does it plug into the way I already run deals? Do I have to change tools? Who sets it up? Does it touch my existing pipeline? That gap between "impressive demo" and "running in my business on Monday" is where the majority of interest quietly dies. It is not skepticism about AI, it is a reasonable demand for delivery mechanics that almost no one bothers to answer. If you assumed the market was full of AI skeptics, this is the correction: the loudest question in our data is not "does it work," it is "how does it get into my business."

The honest first move: before you evaluate any AI, write down the exact workflow you want it to change and the exact tools it would have to touch. Not "we want AI for acquisitions," but "when an OM comes in by email, I want X to happen, feeding into the pipeline tool we already use." Half the delivery confusion evaporates the moment the workflow is written down concretely, because now you can ask any vendor a precise question and judge the answer.

3. Automating the Underwriting and the Pro Forma

The pain, in an operator's words: "I'm using them a good amount, but I guess my wish would be for something that is more of the complete process as much as possible." That is a CRE investor evaluating underwriting AI. This theme came up in 6 distinct real calls: operators who want AI to take a deal from the raw offering memorandum through the pro forma and the analysis, not just answer questions along the way.

Why firms get stuck here: underwriting is the workflow operators most expect AI to nail, and the one where the do-it-yourself version breaks down most visibly. A general chatbot will happily read an OM and produce numbers, but a pro forma is a chain of dependent assumptions, and one wrong extraction upstream quietly poisons everything downstream. Operators get a result that looks complete and is not trustworthy, which is worse than no result. The stuck-point is not "can AI underwrite," it is "can I trust the whole chain end to end without re-checking every cell."

The honest first move: do not try to automate the whole underwrite at once. Break it into stages and automate the boring, verifiable front of it first: pulling the rent roll and key terms out of the OM into a structured format you can eyeball in ten seconds. Get that first pass reliable and you have removed the most tedious hour of the job while keeping your judgment on the assumptions that matter. This is the same underlying blocker as the next one.

4. Hitting the DIY Ceiling in ChatGPT and Claude

The pain, in an operator's words: "I've pretty much implemented everything and I've been able to leverage AI to my advantage. I have templates for different deal types, my virtual assistant uploads the OM, and gets it to a point where the data is filled in." That is a solo investor who self-built his entire stack and is now stuck at reliability. Six distinct real operators told us, on recorded calls, that they are already self-serving with ChatGPT and Claude before ever buying anything, and all six hit the same wall. Another, a principal at a fully-integrated industrial investment firm: "we created a prompt where I can now just put an OM into Claude, and it'll spit out an LOI for me. I'm older, right? I'm early 40s, I'm new to this AI thing, and it's pretty cool."

Why firms get stuck here: this is the most important item on the list for reframing whether you are behind. The person you are worried about outrunning you is not waiting for a vendor, they are already running OM-to-LOI prompts in their own Claude account. And they are stuck at exactly the same ceiling you would hit: the prompt mostly works in the chat window, but it is not reliable or complete enough to run a process they can stake a decision on without babysitting it. The DIY ceiling is not a lack of tools, it is the gap between "it works when I nurse it" and "it runs on its own."

The honest first move: if you are already prompting your way through deals, your next move is not more prompting or a bigger plan. It is turning the prompt that mostly works into a process with guardrails, checks, and a defined output you can trust every time. That is a different kind of engineering than writing a good prompt, and it is where the DIY ceiling actually gets crossed.

5. Trusting the Output (Reliability and Hallucination)

The pain, in an operator's words: one operator described current general-purpose tools as too unreliable, requiring constant manual fixes. Another said his self-built stack simply "lacks reliability and tracking." Reliability came up in 3 distinct real calls, and separately in 6 of the 150 chat messages across two independent chat sessions, which is what makes it a recurring objection rather than a one-off. In one chat thread an operator asked, in sequence: what about liability with sensitive data, so you are not liable, and what if your system messes up.

Why firms get stuck here: in real estate, a hallucinated number is not a typo, it is a mispriced deal or a wrong figure in front of an investor. So operators rightly refuse to trust AI output they cannot verify, and once they have caught the tool being confidently wrong a few times, they stop trusting all of it. The stuck-point is a trust problem, and trust does not come back from a demo. It comes back from a system that shows its work and fails safely.

The honest first move: stop asking AI for answers you cannot check, and start asking it for answers it has to source. The reliable pattern is to have the AI pull from a specific document or dataset and show you where each number came from, so verification takes seconds instead of a full re-underwrite. Reliability is not a property of the model, it is a property of how you wire the workflow around it. We go deeper on the build-versus-buy tradeoff of getting there in build vs buy AI for commercial real estate.

6. Getting the Whole Team to Actually Use It

The pain, in an operator's words: "I don't want it to be just me as sort of the sole person kind of running with this stuff. I want to have it permeate the whole company. Find somebody to really sit with our team and dive deeper into the specifics we're doing day-to-day and help us understand how we can do it ourselves." This capability-transfer demand, and the mirror problem where the principals themselves lack the time or interest to learn a new tool, came up in 2 distinct real calls.

Why firms get stuck here: this is the quietest item on the list but one of the most consequential. AI that lives in one person's head is not an asset, it is a bus-factor risk, and it does not become a firm capability until the team can run it without that person. On the other side, some principals will not learn a new tool no matter how good it is, and a rollout that depends on them personally becoming power users is a rollout that dies. Both failure modes are about people, not technology.

The honest first move: decide honestly which mode your firm is in. If your team is willing and able to learn, the move is a capability transfer, someone working alongside your people until they can run the workflow themselves, not a black box handed over the wall. If the principals genuinely will not learn it, the move is a done-for-you setup that runs without asking them to change how they work. The wrong move is picking the mode that flatters your firm instead of the one that matches it.

7. Putting Confidential Deals Into AI (Data Security)

The pain, in an operator's words: one operator wanted the security and architecture, including where data is hosted, "written down" so he could "absorb it" before committing. Data security came up in 3 distinct real calls and in 3 of the 150 chat messages. It is a smaller theme than delivery or reliability, and we are not going to over-weight it, but for the operators who raise it, it is a hard gate.

Why firms get stuck here: CRE deals are confidential by nature, and the fear is concrete: does my sensitive deal data get used to train someone's model, who can see it, and what happens to it after the engagement. Operators who ask this are doing their job. They get stuck because most AI tools answer with reassurance instead of architecture, and reassurance is exactly what a careful operator discounts.

The honest first move: before you put a single confidential deal into any AI, get the data handling in writing: where it is hosted, whether your inputs are used for training, and what the retention and deletion terms are. The right answer is specific and architectural, not "trust us." An operator who insists on the written version is filtering for vendors who actually have one.

What the Ranking Actually Tells You

Step back from the seven and two patterns stand out. First, the biggest blockers are not skepticism about AI. The loudest theme in the chat corpus is "how does this land in my business," and the operators who show up on the calls are, by and large, already sold on AI and already using it. The room is not full of doubters. It is full of believers who cannot see the path from a prompt that mostly works to a process they can trust.

Second, several of these are the same underlying problem in different clothes. The underwriting stuck-point, the DIY ceiling, and the reliability objection are all, at bottom, the gap between "it works when I nurse it in a chat window" and "it runs a process I can stake a decision on." Crossing that gap is the actual work of 2026, and it is the same work whether you are a solo investor with a clever prompt or a firm rolling AI out to a team. Firms that never cross it are usually tripping on the same predictable causes we catalog in why AI projects fail at CRE firms.

If you recognized your firm in more than one of these, that is normal, and it is not a sign you are behind. It is a sign you are exactly where most operators are. The ones who pull ahead are not the ones with the fanciest tools. They are the ones who pick the one stuck-point that costs them the most and cross it properly. If you want to locate yourself more precisely first, run the CRE AI self-check.