There's a moment in surgery when a robot arm does something a human hand physically cannot: it holds perfectly still. No tremor, no fatigue and no variation between the first cut and the fiftieth. But even with the most advanced robot operating, you still want a surgeon in the room. Not to do the thing the robot is better at, but to catch the thing it will miss.

That is the analogy I keep coming back to when I think about AI in commercial real estate underwriting. The technology is genuinely impressive at the mechanical level. It organizes documents, extracts structured data from unstructured sources, and flags potential issues before they reach a credit committee. 

But the underwriting judgment layer is different. Reading a market, weighing a sponsor's track record, and deciding whether a deal still pencils when two data sources disagree still belong to an experienced human. Value is still a subjective judgment, and not territory AI can navigate alone. 

I recently spoke with Bisnow about how AI is changing commercial real estate underwriting in a more complex macro environment. The piece ran yesterday and raised questions I want to dig into more fully here, because the nuance matters. The industry is at an inflection point where hype and reality are harder to separate, and getting the AI bet wrong in either direction has real financial consequences.

Most CRE Firms Are Piloting AI. Very Few Are Getting Meaningful Results.

The adoption numbers are striking, but adoption alone is not the same as impact. According to JLL, 88 to 92 percent of investors, owners, and occupiers are currently experimenting with AI. That experimentation is happening against a backdrop of rising deal volume and real competitive pressure. The Mortgage Bankers Association projects commercial mortgage origination volume will hit $806 billion this year, up from $633.7 billion in 2025, which means more deals moving through more pipelines with more pressure on the teams underwriting them.

But piloting and deploying it inside a real underwriting workflow are two very different things. Only about 5 percent of CRE firms report achieving all of their AI program goals. That gap isn't primarily a technology problem. It's a trust and workflow integration problem, and understanding that matters if you want to get ahead of it.

Most AI tools in CRE right now are being used for what I'd call ‘document intelligence’ work. That means pulling structured data from rent rolls, organizing lease abstracts, summarizing offering memorandums, and identifying what information is missing from a deal package. Leading AI platforms now achieve 90 to 97 percent accuracy on standard commercial lease terms, and abstractions that used to take four to six hours per document now can be finalized in under 15 minutes. That's a measurable improvement in underwriting workflow.

But document intelligence is not the hardest part of underwriting. Catching the nuanced clause buried in an exhibit that changes your CAM recovery math, or knowing that a particular submarket's employment base is more vulnerable to tech sector contraction than the aggregate numbers suggest, those are the things that separate a good underwriter from a junior one, and they're still beyond what AI handles without human oversight.

The Lease Is the Foundation of CRE Underwriting. AI Is Finally Starting to Read It Properly.

Commercial real estate underwriting starts with the lease. For a single-tenant net lease deal, that's one document that easily could be dozens of pages long. For a large retail center or a multi-tenant office, it's dozens, sometimes hundreds, each with its own escalations, exclusivity clauses, co-tenancy provisions, expense caps, and early termination rights. A missed CAM cap in a complex lease doesn't just affect that tenant's contribution. It changes recovery assumptions and can materially change valuation.

The Appraisal Institute has noted that valuation errors of just 10 percent on a $5 million property represent $500,000 in mispriced risk. In a competitive acquisition environment where deals are typically trading within 5 to 10 percent of underwriting, that kind of error is the difference between winning a deal well and winning it badly.

JLL's own work has documented cases of discovering over $1 million in missed lease clauses after implementing AI lease review. For most firms, that kind of finding covers the platform cost within a single deal. But the ROI framing cuts both ways. If AI tools are implemented poorly or reviewed carelessly, the same scale that accelerates document processing can just as easily accelerate the cost of a missed clause.

The tools CRE has historically relied on for this work, such as Argus Enterprise for institutional modeling and custom Excel spreadsheets for everything else, are powerful within their lanes. Argus is the common language of CRE finance, but it can feel archaic to use and carries significant licensing cost. Excel is flexible but fragile. A single broken formula in a complex model can corrupt assumptions in ways that don't surface until the deal is already in market. 

AI-powered document processing is filling the gap between raw document and clean model input, and it's doing it faster and more consistently than any analyst team can at scale.

How AI Underwriting Tools Earn Trust Over Time

When I think about how AI underwriting tools will achieve real adoption, I think about it as a trust curve built across asset complexity.

It starts with the simplest assets. Single-tenant net lease properties, a Starbucks or a dollar store with a single lease to analyze, standardized terms, and a straightforward cash flow profile. An analyst runs AI-assisted abstraction on the first ten or fifteen deals, checks every output against what they know, and builds a calibration layer. Over time, the accuracy record speaks for itself, and reliance increases.

From there you move up the complexity ladder. Small multifamily, mid-size retail, eventually more complex multi-tenant assets. The AI underwriting tools that win long-term adoption will be the ones that mirror how analysts actually learn, starting with simple verifiable deals and earning expanded responsibility as the accuracy record builds.

Good underwriting teams operate the same way. A junior analyst produces the first pass. A partner reviews it, because the cost of an error at the decision stage is too high to skip double- and triple-checks. AI fits naturally into the analyst role in that structure: faster, more consistent, and able to process far more documents, but still operating inside a review chain. PwC's Emerging Trends in Real Estate 2026 report echoed this, noting that “AI is a solid replacement for a junior analyst” while finding that employment for more experienced underwriters has remained stable or grown alongside AI adoption.

Some firms report handling three to four times more deal applications with the same analyst headcount after implementing AI-assisted underwriting workflows. The leverage from increased output cannot be overstated.

What AI Will Not Replace in CRE Underwriting

I want to be clear-eyed here, because I think the hype cycle around AI does real damage when it creates false confidence. The macro environment we're operating in now, with interest rates, tariff uncertainty, consumer challenges, and shifting employment patterns all feeding into deal risk simultaneously, is exactly the kind of environment where qualitative judgment matters most.

AI is already strong at quantitative analysis. It can run scenarios, stress-test assumptions, flag when a deal's expense ratios are outliers relative to comparable assets. What it can't do is read a room, evaluate a sponsor's judgment or character, or weigh the significance of something a broker mentioned off the record that doesn't appear anywhere in the documents. Those inputs don't have a data field. They live in experience and relationships.

CBRE's Michael Riccio put it well in the Bisnow piece: AI takes friction out of the process by cleaning data, running scenarios, and pressure-testing assumptions, so analysts spend less time building spreadsheets and more time understanding real risk. That framing is exactly right. The value of AI lives in removing the mechanical work that keeps the expert from doing their best work.

The best investment decisions I've seen come from people who combine deep data literacy with the ability to make judgment calls when the data is incomplete or ambiguous. AI accelerates the data literacy side of that equation. The judgment part is harder to systematize, and I suspect it always will be.

The Speed Advantage of AI Underwriting and What It Means for Deal Flow

At Crexi, we process thousands of deals through our marketplace daily. That volume gives us a clear view of where underwriting speed creates competitive separation. When an investor can upload an offering memorandum and extract structured insights in minutes rather than hours, they're both saving time and narrowing a pipeline of opportunities down to the ones worth serious attention before a competitor has finished reading the first document.

Speed in underwriting also changes what investors are willing to screen, review, and pursue. The manual process creates a natural filter where deals that require extensive preliminary research get deprioritized, not because they're bad deals, but because analyst bandwidth is finite. AI changes that constraint by expanding how many deals can be screened before deeper diligence begins. Investors can cast a wider net at the screening stage and apply deeper due diligence to a larger pool of screened opportunities.

This is especially meaningful in secondary and tertiary markets, where deals often sit longer because fewer institutional buyers are looking at them early. AI-assisted market analysis can surface comparable market characteristics across geographies that a traditional research team might never consider, extending their reach into markets they might never have considered. 

The platform infrastructure we're building at Crexi is specifically designed to close this gap. The ability to move from raw documents to actionable deal intelligence, and from deal intelligence to market context, within a single integrated workflow is something the industry hasn't had before. It's what allows a broker or investor to focus on the right deals faster, rather than spending their best hours on document triage.

What CRE Underwriting May Look Like in Three to Five Years

The PwC/ULI Emerging Trends report framed it well: AI adoption in CRE is moving from experimentation to operational deployment. The real estate investment software market is projected to grow from $5.6 billion in 2025 to $9.8 billion by 2030. Proptech funding reached $16.7 billion in 2025, a 68 percent year-over-year increase, with AI-centered tools growing at 42 percent annually. The capital is pointing at this problem.

My view is that within three to five years, the underwriting workflow will look fundamentally different at firms that integrate AI infrastructure well. Document ingestion and abstraction will be automated for standard asset types. Market context, submarket-level analysis, comparable cap rate ranges, and supply pipeline data will be available on demand rather than as a research project. The analyst’s job will shift further toward interpretation and decision-making, not document triage and organization.

The firms that get there first will have a structural advantage in deal velocity and screening capacity. Cushman & Wakefield's 2026 AI Impact Barometer identifies lease intelligence as one of the top five use cases driving measurable ROI in CRE this year. That signal is consistent with what we're seeing in deal activity at Crexi, where speed of insight is becoming a baseline expectation, not a differentiator.

That said, I'd temper the full-automation narrative. End-to-end autonomous underwriting, where a system ingests a deal package and produces a final buy or no-buy recommendation without human review, is not where this industry is headed in the near term. High-stakes capital decisions require a human accountable for the outcome, to trust the advice. That isn't going to change because AI gets more accurate. It may change when AI earns decades of trust across thousands of deals. We're not there yet.

How Investors and Brokers Should Use AI in CRE Underwriting Today

If you're trying to figure out where to start with AI in CRE underwriting, here's how I'd approach it.

• Start with document-heavy underwriting workflows. Lease abstraction and offering memorandum review are mature use cases with clear ROI. If you're still doing these manually at scale, you're paying an opportunity cost on every deal.
• Build a calibration baseline first. Run AI outputs in parallel with manual review on the first 10 to 15 deals and identify areas for improvement. That iterative process builds the trust foundation that lets you rely on AI outputs with more confidence later.
• Treat underwriting speed as a competitive variable, not just an efficiency metric. Faster preliminary underwriting means a larger screened deal pipeline, which means more opportunity to find the right deal before the market does, but only if the results are accurate.
• Keep the review chain intact. Use AI at the analyst stage and human judgment at the decision stage. This is the right structure for decisions at this scale, with millions of dollars on the line.
• Look for integrated underwriting workflows, not isolated point solutions. The firms seeing the best results aren't stitching together five separate tools. They're onboarding platforms where document intelligence, market context, and deal execution live in the same environment.

A Final Note on Trust and Adoption

The robot surgeon analogy isn't pessimistic about AI. Robotic surgery has meaningfully improved outcomes for patients. It's a standard of care for certain procedures. But it took years of parallel operation, data collection, and trust-building before surgeons stepped back and let the system act under supervision.

CRE underwriting is on a similar arc. The tools are currently good enough to materially improve the underwriting process. They are not good enough to replace human judgment on a $20 million investment decision. Both of those things are true at the same time, and keeping them in mind is what allows firms to deploy AI in ways that improve outcomes rather than create new risks.

The industry is moving. The real question is whether you're building toward this future systematically or assuming you can catch up later. In my experience, later always costs more than you expect.