The Biggest Risk of AI in Real Estate May Be False Confidence | Opinion

Artificial intelligence (AI) is accelerating the pace of commercial real estate underwriting, yet speed alone does not equate to improved quality of decision-making. As AI transitions from experimental usage to integration within investment workflows, the most significant threat may not be erroneous outputs, but rather it could be the generation of alluring yet deceptive answers that lull investors into overlooking critical underlying assumptions.

This is especially pertinent within a sector where seemingly minor premises can dramatically sway the financial viability of a deal. Progress is swift: Deloitte's 2027 Commercial Real Estate Outlook indicates that 92 percent of surveyed commercial real estate (CRE) entities are still in the research or pilot stages of AI adoption, whereas only 8 percent have incorporated AI solutions into their operations.

Conversely, fewer than half of respondents reported implementing more sophisticated controls such as challenger models, exception handling, or detective controls. The survey, which encompassed 950 executives and their respective subordinates at CRE owners and investment firms, underscores that AI's presence in real estate is indisputable.

The remaining question is whether the industry possesses the acumen to wield this technology judiciously, distinguishing computational efficiency from astute investment assessment. AI excels at handling vast volumes of data. It can swiftly analyze numerous documents, catalog lease information, detect inconsistencies, juxtapose assumptions, and simulate various scenarios – feats surpassing human capabilities in manual processing.

Reports from PricewaterhouseCoopers (PwC) and the Urban Land Institute's Emerging Trends in Real Estate 2026 project indicate a burgeoning AI integration across real estate domains, including research, underwriting, and reporting. Such capabilities hold immense potential. However, they also engender a nuanced risk: the allure of AI-generated outputs, characterized by neat formatting and comprehensive data backing, can amplify their perceived credibility beyond merit.

Underwriting methodologies have consistently represented reality, albeit simplified; AI does not alter this principle. It merely expedites the formulation and scrutiny of models, yet it remains incapable of independently verifying the assumptions embedded within these models against empirical reality. Commercial real estate presents unique complexities not shared by other asset classes.

Properties are not uniform datasets. For instance, a retail venue may appear lucrative based on comparable rents and historical occupancy rates, yet years of tenant turnover might reveal a scarce parking lot that impairs leasing efficacy. Similarly, an office building may seem budget-friendly compared to contemporaneous sales, yet its deferred capital expenditure expenditures could paint a contrasting financial picture.

A property might possess a ground lease with reset conditions that significantly reshape economics. Taxation policies might evolve post-acquisition. Operational expenses that were previously internal could subsequently be borne by subsequent owners. The crux of deal evaluation often resides in less quantifiable elements that are not readily amenable to structured analysis.

This underscores the indispensable role of human discernment. As the pioneer of Sapp Capital Advisors, I have devoted years to dissecting the tenets of real estate investments. This experience has underscored my conviction that robust underwriting transcends mere model construction; it encompasses comprehending the rationale underpinning numerical outputs, scrutinizing the premises supporting them, and recognizing which minutiae could precipitate substantial shifts in outcomes.

AI streamlines these inquiries but does not dispense with their necessity. Consider a relatively uncomplicated scenario: assessing lease agreements. An AI mechanism may rapidly compile a matrix enumerating rents, expiration dates, options, and other stipulations across hundreds of lease documents, significantly economizing time.

Nonetheless, discerning the substantive impact of a lease clause diverges from its mere identification. A model can flag an impending lease expiration. However, an astute underwriter must delve deeper – assessing tenant renewal likelihood, local market predictive indicators, alternative property offerings with more favorable terms, tenant negotiation leverage, and physical suitability of the space should the tenant vacate.

While AI expedites these queries, it does not absolve the need for their formulation. The same tenet applies to market intelligence. In markets characterized by restricted transaction activity, comparable sales may yield diminished relevance. Historical data may not accurately encapsulate prevailing market conditions, particularly in locales undergoing evolutionary transformations.

Interactions with brokers, property owners, lenders, and other industry stakeholders can furnish contextual nuances absent from spreadsheet representations. This does not imply that human judgment is impervious to error; individuals harbor biases, commit mistakes, and may become overly reliant on their preconceived notions. AI can, conversely, unearth such frailties by facilitating extensive scenario testing and exploration of information potentially overlooked otherwise.

The technology's most substantial contribution lies in augmenting rather than supplanting human analytical capabilities. Professional underwriters might leverage AI to process hundreds of leases, as opposed to the limited number previously scrutinized. Testing multiple scenarios, which was once restricted to three, could be expanded to dozens.

Detecting discrepancies between seller disclosures and underlying documents could transpire at a swifter rate. The efficacy emanates from expanding analytical breadth, not eradicating personal accountability. CRE firms contemplating AI integration should thus embed verification mechanisms into their investment frameworks: challenge assumptions, probe scenarios, pinpoint gaps AI cannot discern.