Cracking the Code on AI Lease Abstracts: Production Data CRE Teams Can Trust

CREModels has delivered lease abstracts since 2010. Now its AI lease abstraction turns dense lease files into reliable, production-grade data faster, with expert analysts signing off on every field.

Madeline Miller, head of CREModels’ lease abstraction practice, recently kicked off a project covering about 200 retail tenants the way many projects now begin. She loaded a single zip file, hundreds of unsorted documents pulled straight from the due diligence vault, into Mercury, CREModels’ proprietary AI-assisted abstraction system.

By the time she came back, Mercury had handled much of the front-end organization work. Tenant folders had populated. Financial statements, offering memoranda, and property photos had been sorted into their own folders. And the part she was really waiting on, the first-pass lease abstracts, were already running.

“Three or four years ago I was doing all of this by hand. Unzipping files, figuring out which documents went with which tenant, organizing everything before I could even start,” said Miller, who has spent five years on lease abstracts and personally overseen tens of thousands of them. “Now I drop in the zip file and Mercury takes care of the organizing. By the time I sit down to it, the documents are sorted and the first pass is already running, so my time goes to the actual analysis instead.”

That project wrapped in under two weeks. A pre-AI team might have needed up to six weeks for a portfolio that size, roughly one-third the turnaround time.

Since late 2024, CREModels has used this AI-assisted workflow across multiple production client projects, including closed transactions and completed portfolio onboardings.

Lease abstraction has always been one of the most painful bottlenecks in commercial real estate. Before anyone can underwrite a property, close an acquisition, onboard a portfolio, or report to investors, someone has to know what the leases actually say. The work is high-stakes and, frankly, tedious. A single tenant file can run 200 pages across an original lease and ten or more amendments, assignments, estoppels, and option notices. Across a portfolio, that complexity compounds quickly.

The one click that starts Mercury is the easy part. What makes the output trustworthy is everything CREModels built behind it, and the expert review that still comes after.

It’s Still the AI Wild West

When real estate professionals hear AI claims these days, they should bring some skepticism, said CREModels Cofounder and COO Mike Jaworski, a 20-year commercial real estate veteran who has worked on more than $20 billion in deals.

“It’s the AI Wild West right now in commercial real estate,” he said. “People are using general, off-the-shelf models for lease abstraction and getting inconsistent results. They get miscalculated financial data that can lead to bad decisions, and they’ll often get a different answer every time they ask the same question.”

Generic AI tools can be useful for quick summaries, but production-grade CRE lease abstraction requires more than document extraction. A multi-lease, amendment-heavy project that supports real financial decisions requires document organization, amendment chronology, structured outputs, real estate judgment, and quality control against the source documents.

“In our testing, general-purpose tools were more likely to miss items like common area maintenance, rent abatement, base years, timelines, and financial data,” Miller said. “Once you feed them several documents for one tenant, the risk of mixing up the material terms gets much higher.”

“Our clients tell us they’ve been getting 75% or 80% accuracy with some of the AI tools and companies they’ve tried,” Jaworski said. “But in the world of lease abstracts, 80% accuracy is no better than 0% accuracy.”

Delivering Real Results, Not Pitching Product

This is where CREModels’ AI lease abstraction differs from the apps flooding the market. It doesn’t just sell a tool and leave the client to run it.

In practice, that means the client isn’t buying an adoption headache. They’re buying an outcome: clean, structured, reviewed lease data they can actually base decisions on. Mercury compresses the heavy, repetitive first pass. CREModels’ analysts supply the judgment. What clients get is a finished, decision-ready result, not a login and a learning curve.

“Clients don’t come to us for abstracts because they want another tool to manage,” Jaworski said. “They come to us because they need reliable lease data. AI lets us deliver that work faster, but the responsibility for the final output is still ours.”

How AI Lease Abstraction Works

What sets Mercury apart from a general chatbot is that it was built to read a lease the way a CRE analyst does, not as a standalone book report. Once it has sorted a portfolio, it groups them by tenant, works through the amendment history in order, and abstracts the leases and amendments together so it knows when a later amendment supersedes the original and which terms are actually material. Retail tenants make this especially valuable. They often go through years of expansions, contractions, renewals, and assignments, and by the tenth amendment a file can be genuinely hard to follow.

“By the time you reach the tenth amendment it can get really confusing,” Miller said. “Mercury can work through that chronology quickly.”

The output isn’t a loose summary. It’s structured data. Abstracts can be grouped by portfolio, exported to Excel, or converted to rent rolls. Because the data is standardized, it drops cleanly into the underwriting and cash flow models teams already use. Another benefit is that the structured output also makes high-quality context for downstream AI workflows.

“Say you’re in due diligence,” Jaworski said. “Instead of a person reading every lease, summarizing it, and keying it into a model, you drag and drop the documents in. In about an hour you can have 50 leases abstracted and ready for QC. That’s the structured foundation for a larger cash flow model that reflects what’s actually in those leases.”

Human Review Is the Final Word

The speed is real, but it isn’t the whole story. After Mercury’s first pass, a CREModels analyst on Miller’s team reviews the output against the source documents, corrects errors, and resolves the questions that take actual real estate judgment. Which amendment takes precedence? Was the prior rent schedule superseded? Do the recoveries account for the right exclusions and base years? Does a co-tenancy or termination right matter to the underwriting?

The review is what lets clients trust the result, and it’s why the service is fast without being reckless. The AI first pass is quick, and the expert quality-control pass is much faster than it used to be. The speed comes from compressing the review, not skipping it.

“We go through each document to confirm everything is there and accurate,” Miller said. “Is it still a manual step? Yes. But it’s far faster than before, with no sacrifice in accuracy, and accuracy is the entire point of lease abstraction.”

Building Trust in AI Lease Abstraction

The market is ready for this, but cautious. A May 2026 study from First American Data & Analytics and DealGround found that 66% of CRE professionals now use AI weekly or daily, yet only 5% trust it enough to inform real deal decisions. More than half use it strictly for support.

That gap is the whole opportunity. Lease data drives valuation, financing, reserves, lender diligence, and investor reporting. If the abstract is wrong, everything downstream can be wrong, which is why most professionals won’t let AI near the decision itself.

“Our approach gives clients the confidence to actually reap the benefits of AI,” Jaworski said. “When a document abstraction project is materially faster and you can still trust the information, that’s a real change. A lot of the tools out there are just faster with an asterisk on quality.”

There’s a longer-term payoff, too. A reviewed abstract is also a clean, organized layer of context. Once a firm’s leases are organized this way, that data becomes a durable foundation for asset management, reporting, portfolio analytics, and future AI work. Raw lease PDFs are a poor starting point for many AI initiatives. Reviewed abstracts stored as structured data are much better.

Today, CREModels uses Mercury internally on client projects. Over time, the company expects to bring AI-assisted abstraction deeper into CRE Suite and offer clients more flexible delivery options, including AI-generated first-pass abstracts, analyst-reviewed deliverables, and structured lease data for underwriting, reporting, asset management, portfolio onboarding, and AI-enabled workflows.

“Depending on the project, people can have us review the output,” Jaworski said, “or they might use the AI versions for smaller or inline tenants and have us do the full QC on anchors and big boxes. So they can close confidently and get moving on their next deal.”

The real shift in AI lease abstraction isn’t a button or a demo. It’s a service CRE firms have always needed, now delivered faster and still backed by analysts who answer for the output.