How Commercial Real Estate Firms Successfully Adopt AI

If you’re reading this, someone at your firm decided AI was important and then sent it to you.

What usually happens (at least in real estate) is a CEO comes back from a conference where AI was discussed, or an LP or client asks how your firm is using AI, and all of a sudden you’ve been volun-told to figure this AI thing out!

For some in your seat, that mandate comes with an established budget and a clear definition of what success looks like. But most, unfortunately, are largely left to figure it out on their own.

So you do the obvious thing: you pick an AI harness, whether Claude, ChatGPT, Copilot, etc., and you sign up. You send out an announcement, people begin logging in, and a handful of folks (usually your in-house Excel nerds!) really lean in. But most use it as a faster, more expensive Google.

And so, progress stalls, and you’re now sitting there asking yourself: now what?

Consider this, as it frames everything else hereafter:

This matters because it illustrates the next misstep most people in your seat make.

The pressure to get results from AI pushes you to take action, and the most obvious action is to train your people. So you find an AI training program. Most programs are run by well-meaning people with short-lived careers in CRE (if any CRE experience) and an interest in AI.

And given how nascent this technology is, an “interest in AI” seemingly qualifies someone to teach AI!

So, your people take that AI training. They learn how an LLM works, a few tricks for getting more out of AI, and they see a demo or two of Claude using an agent skill. And while they leave the training impressed, and maybe even energized, three weeks later nothing has changed.

That’s the failure we see most often. It’s not that the team is lazy, ill-equipped, or untrained. It’s that the training didn’t connect what they actually do to the key ingredients for success with AI.

As we implement AI in our firms and others, the difference between positive ROI from AI and spinning your wheels comes down to three actions.

Three actions.

The next few sections lay out what those three actions are and how to tell whether your firm has actually made them.

Read it end to end. It’s fifteen minutes, and it arms you with an answer to “now what?”

— Spencer Burton and Joe Prendergast

Most firms start off wrong. They ask, “What can AI do for us?” which in turn either a) leads to a series of demos with point solution vendors (e.g. an AI lease abstraction tool, an OM generator platform, or an AI underwriting app) or b) leads them to ask Claude to do something like reconcile an estoppel to a lease or build an offering memorandum without properly teaching it how.

Route A leads to app fatigue and an unsustainable tech stack, while route B leads to outputs that aren’t quite right and/or can’t be trusted.

The firms getting real leverage start from the other end. They start by auditing the work they do and flagging the work that, if AI handled it, would produce the greatest lift. They then assess which high-impact tasks are the most feasible for a single AI harness to handle, assuming it is properly tuned to perform those tasks.

They then focus their efforts on those high-impact, high-feasibility tasks.

For example, we recently had the pleasure of working with a top investment management firm that really took this action to heart. In a four-week period, a pilot group of 25 team members identified and trained Claude to do 25 tasks on their behalf. Over the next year, just from that first four-week sprint, the firm will save an estimated 3,000 hours, freeing those people from 25 monotonous tasks to focus on higher-value work.

Or take a nimble investment sales team we’re working with. Three producers and an analyst. Their stated goal was to double fee revenue over the next two years without adding headcount.

They asked themselves, what are the tasks we’re not doing that, if we had the time, we would do? With an emphasis on dollar-productive tasks, they identified four simple, but powerful, prospecting tasks that AI could do. Since training AI (Manus in their case) to do those things six months ago, they have doubled their BOV volume!

Should do vs. can do

In practice, the high-impact, high-feasibility tasks turn out to be remarkably consistent from firm to firm.

Creating the list of things AI should be doing is the easy part. But knowing which of those tasks AI can do is the hard part.

A task should never be assigned to AI unless the AI can produce the deliverable for that task as well as or better than a person can. And how difficult it is to get AI to do that is what we refer to as ‘feasibility.’

Feasibility is a function of the AI having three things. Flip them on and off for a task you have in mind:

This is a critical action that many firms are delaying, or ignoring altogether. But it is the action that will produce the greatest edge in the coming years.

Here’s the uncomfortable truth. At this point, almost every one of your competitors is “using” an AI harness (i.e. Claude, ChatGPT, Copilot Studio, Manus, Perplexity, etc) and some have even moved beyond those basic harnesses and are using autonomous agents like Grok Bot or Hermes!

Whatever advantage comes from adopting an AI harness will be gone a year from now. There is no moat in licensing AI and, if anything, it’s now a matter of those who don’t use AI dying a slow death.

Maybe you have 10 years of T12s sitting in your email inbox, cloud drive, and/or local folders. Or maybe you have a rent or sale comp database sitting in Excel files.

On top of that, every day you have conversations packed with valuable data, and you’re looking at OMs, market reports, and tenant financials that help you make investment decisions. In a pre-AI context, the cost and time of structuring and securely storing that data internally was not worth the incremental benefit of having that data at your fingertips when making decisions.

But in the era of AI, your edge, in part, comes from structuring the data you’ve already collected and building an engine to consistently structure and recall data going forward. And AI lowers the cost and complexity of doing that dramatically.

Additionally, the quality and usefulness of an AI output is a function of how that AI is instructed to produce that output. Over your firm’s history, immeasurable value in terms of process, insights, and “street smarts” has accumulated in the minds of your people. Encode that into a format an AI can use (i.e. your Methods encoded in an Agent Skill), have a disciplined process for growing that knowledge base, and pair it with your proprietary data, and you build a meaningful and distinct edge from AI.

A useful way to hold this, and one we teach, is that Knowledge (which includes data and methods) plus Instructions plus Tools equals Optimal Output. Most firms spend their entire AI budget on the Tools component and then wonder why everything that comes back from the AI is so generic (i.e. AI slop).

It’s generic because they didn’t give the model any distinct knowledge to inform its work!

Three tiers of data

Think about your data in three tiers.

For instance, we recently worked with a shopping center investor who had collected tens of thousands of leases over their history and continues to see hundreds of new leases every month. We helped them build a structured database of those rent comps, connected it to their Claude, and built an engine to keep the database updated weekly with new comps.

Effective AI adoption, one that generates a positive return on AI spend, requires what we refer to as “practical AI governance.” Practical in the sense that it’s not overly bureaucratic and encourages high-ROI AI practices, while also protecting the firm’s data, reputation, and standing.

Right now, across our industry, there are hundreds of thousands of CRE pros using AI on real work in a way that hasn’t been mentioned to anyone. They aren’t necessarily being reckless. They simply don’t have a practical standard to follow, and a “no AI” or limited AI policy is NOT practical governance in the era of AI.

So, those CRE pros are making a judgment call that the safest path forward is to stay quiet about it.

That’s what the absence of practical governance actually leads to. Shadow usage with uneven outputs and no way to learn from the things that are working. And the reflex fix, where legal or IT steps in to write an “AI policy” completely detached from how real estate work actually gets done and largely unaware of where the technology is and is going, just makes it worse.

Restriction doesn’t stop the work, it just moves it somewhere corporate can’t see. People open a personal account, pay the twenty a month themselves, and build their own setup around it.

We’ve seen this at the senior leadership level, someone who had moved their real thinking work onto a personal account because the firm harness couldn’t handle it. The policy didn’t reduce AI usage. It just puts firm data in a consumer account nobody is administering, and a policy written to reduce risk ends up manufacturing it.

Three principles

Practical governance does the opposite job. Its purpose is to make adoption faster by removing the question of whether something is allowed, while grounding adoption in principles that support productive activity. In our experience, there are three principles in every firm winning today with AI:

That’s it.

You now have the beginning of an AI roadmap: three high-impact tasks with a good sense of which you should start with and a view into the knowledge and data that could create an edge as well as the basic guardrails needed to move forward.

Now notice you didn’t need another AI demo or generic AI training session. You just needed a framework to follow and the confidence that, at the end of the process, you’d have something that made the process worth it!

That’s the whole method, and it’s the reason generic AI training lands flat. Generic training starts with whatever AI tool is en vogue and hopes that learning to use it will magically translate into AI results.

But the reality is, it never does.

The three actions above are what we’ve seen separate the firms that are truly getting leverage from AI from the firms that are spinning their wheels.

You now have the what. The gap between reading this and having it working inside your firm is the how, and the how is where it gets slow if you’re doing it alone.

A few things worth knowing about us before the last page.

For over seven years, our A.CRE Accelerator program has been the industry standard for real estate financial modeling training, with over 30,000 members and 3,100 graduates across top institutional investors, brokerages, developers, and lenders.

We know how this industry works because we’ve spent our entire careers in it. Spencer has 25 years in CRE, has underwritten more than $30 billion and closed more than $5 billion in real estate. Joe has over a decade in the industry and has engaged with more CRE firms than almost any other industry professional.

We’ve trained over 3,000 CRE professionals on practical AI application in real estate, work directly with dozens of firms to help them implement AI, and use the framework ourselves daily across our AI and real estate businesses.

And when discussing this with enterprises, the arithmetic tends to end the debate about whether it’s worth doing.