Institutional real estate investors have spent the past several years approving AI budgets and building AI into how they source, screen, underwrite, and manage deals. With that work well underway, the question facing investment leaders has shifted from adoption to impact: is AI actually changing how their firms invest?
To answer it, Dealpath commissioned independent research firm Censuswide to survey 103 investment and technology professionals, from analysts to the C-suite, at institutional real estate firms in the United States and Canada managing $500 million to more than $40 billion in assets. Fieldwork was completed in 2Q 2026, and the results are published in Dealpath’s report The 2026 State of AI in CRE Investing: Adoption Without Impact.
The short answer is “not yet,” and not because of the AI models themselves. The gains respondents reported were modest and uneven; many said the time spent checking AI’s work cancels out the time it saves, and fragmented deal data was the most-cited reason AI fell short. The findings below cover four areas: the gap between adoption and results, the cost of verifying AI’s work, how firms’ confidence in their data compares with what it delivers, and where trust in AI breaks down.
1. Adoption Has Outpaced Measurable Outcomes
Nearly all respondents (97 percent) said AI is integrated into their firm’s investment process, and 52 percent described it as fully integrated. Yet the top outcome AI improved most for these teams, cited by just 36 percent of professionals, was reduced time on manual and repetitive tasks.
The second highest outcome named was faster deal cycles, at 35 percent. The top two outcomes both reflect AI accelerating individual tasks, such as extracting figures from documents or summarizing diligence materials, inside workflows that are otherwise unchanged. Improvements to the investment decisions themselves ranked last: just 19 percent cited more precise decisions.
2. Verification Absorbs Much of the Time Saved
Verification is close to universal: 94 percent of respondents reported checking AI outputs most of the time or always. An analyst who uses AI to extract a T-12, for example, still confirms every line, because a single transposed expense flows into net operating income and then into valuation. The time this takes is what Dealpath calls the verification tax: the cost of confirming AI outputs are accurate enough to act on, which rarely shows up as a line item. In effect, AI has moved much of the effort from performing the task to checking it.
The cost shows up in the time data. Only 51 percent of respondents said AI saves them time once verification was counted, while 41 percent said work involving AI now takes longer than building the same analysis manually.
The burden varies by type of work. Among respondents in credit, where covenant-heavy loan documents demand line-level precision, 51 percent said AI makes their work slower. Among those in dispositions, whose work leans on summarization that is quick to check, 59 percent said it saves time.
3. Self-Assessed Data Readiness Does Not Match Results
While 83 percent of respondents rated their firm’s data infrastructure as mostly or fully ready for AI, 90 percent said data quality or fragmentation had limited AI’s impact at their firm, a contradiction the report calls the confidence gap. A quarter called data the primary constraint, and only 8 percent said it had not limited AI at all. When AI outputs fall short, incomplete, inconsistent, or fragmented data is the most-cited cause, named by 43 percent, ahead of hallucinated outputs at 36 percent.
The gap likely reflects a higher bar for what counts as usable data. Firms have historically judged their data by whether it is accurate enough for a trained professional, one who knows the data and how it is organized, to act on. AI requires more: every field must also be structured, current, and connected to the deal record. A professional can work around the gaps; a model inherits them and fills them with guesses rather than flagging them.
The constraint is felt most acutely closest to the deals: 77 percent of investment professionals said data has significantly or moderately limited AI, compared with 56 percent of technology and data leaders.
4. Trust Declines as Tasks Move Toward Valuation
Trust in AI varies predictably by task, along what Dealpath terms the trust spectrum. Informational tasks, whose outputs are easy to check against a source, earned the most trust: 55 percent of respondents highly trust AI to summarize diligence documents, and 53 percent to screen opportunities for baseline fit. Judgment tasks closest to valuation earn the least: just 35 percent highly trust AI to score and rank opportunities, and 36 percent to recommend comps, the two lowest of any use case.
Underwriting shows what is at stake. It had the lowest AI adoption of any lifecycle stage, with 48 percent of respondents using AI there, yet 23 percent named it as the stage where AI has delivered the most measurable value, second only to portfolio management at 31 percent. The pattern is consistent with the survey’s root-cause findings: the more a team can rely on the data beneath an output, the further along in the deal cycle it is willing to use AI.
Outlook: As AI Models Change, the Underlying Data Will Dictate Returns
A year earlier, Dealpath’s 2025 State of AI Readiness survey, a separate study of 100 institutional investors, found 98 percent naming data infrastructure a top priority for the following 24 months and 61 percent expecting AI to deliver faster deals. In the 2026 survey, just 35 percent named deal speed among the outcomes AI had most improved, and the data foundation investors prioritized a year ago was the most-cited reason AI falls short. The industry has deployed AI faster than it has built the foundation beneath it.
That leaves every firm paying one of two costs. The verification tax recurs with every output built on fragmented data. The data toll is a one-time investment to structure, connect, and govern deal data so outputs can be trusted. The survey leaves little doubt which is cheaper: firms should pay the data toll once rather than the verification tax forever.
Waiting for better models will not remove the need. Already, 71 percent of respondents reported using general-purpose large language models directly, and every firm can license the same ones. When the models are shared, a firm’s advantage moves to the proprietary data these models can reason over: the opportunities the market has brought to a firm, and the firm’s decisions about them. Most firms already have more of this data than they can use, but it is scattered. Offering memorandums and broker emails sit in individual inboxes and team spreadsheets where no colleague or model can query them, and the reasoning behind a pass is often lost to turnover because it was never written down.
Timing matters. A document archive can be reprocessed as models improve; a decision that was never recorded cannot be reconstructed by any model. Firms that capture deal flow and decisions as the work happens will build a record that deepens with every deal they screen, so that each deal informs the next decision. Those that wait will have better models working on the same fragmented data. In 2027, the return on AI will depend less on which model a firm uses and more on the context firms can give that model.
—Dealpath
