Artificial intelligence (AI) is making commercial real estate underwriting faster, but faster is not the same as better. As AI moves from experimentation into investment workflows, the greatest danger may not be that the technology produces obviously wrong answers. It may be that it produces convincing answers that cause investors to stop questioning the assumptions underneath them.
That distinction matters in an industry where a seemingly minor assumption can materially change the economics of a deal. A different expense forecast, a missed lease provision, an incorrect financing assumption or an overly optimistic exit valuation can alter an investment thesis long before anyone realizes the model was wrong. The technology is moving quickly.
Deloitte's 2027 Commercial Real Estate Outlook found that 92 percent of surveyed CRE organizations were still in the research or piloting phase of AI adoption, while only 8 percent had integrated AI solutions. At the same time, fewer than half reported implementing more advanced controls such as challenger models, exception handling or detective controls. Deloitte's survey covered 950 executives and their direct reports at commercial real estate owners and investment organizations.
That tells me something important. The question is no longer whether AI will enter real estate. The question is whether the industry will learn how to use it without confusing computational speed with investment judgment.
AI is exceptionally good at processing information. It can review hundreds of documents, organize lease data, identify discrepancies, compare assumptions and test scenarios far faster than any person working manually. PricewaterhouseCoopers (PwC) and the Urban Land Institute's Emerging Trends in Real Estate 2026 report found that AI use is expanding across real estate, including in research, underwriting and reporting.
Those capabilities can be enormously valuable. But they also introduce a subtle risk: when an output arrives quickly, neatly formatted and supported by thousands of data points, it can feel more authoritative than it deserves to be. An underwriting model has always been a representation of reality, not reality itself.
Extract — continue reading at the source.
