While Wall Street waits for AI to revolutionize real estate, a quiet breakthrough in better explanations might finally make it happen.

This 2022 “AI train” comment from Moghadam is notable. It originated from a Costar article (link below) in which Prologis announced an investment in TestFit, ”an AI software startup,” and foreshadowed artificial intelligence playing “a pivotal role across the industrial property giant's business.” Fast forward to 2025, TestFit is making cool software but there doesn’t appear to be actual AI in it, and Prologis may not be using AI for much either as it’s not mentioned at all in its recent 10-K or investor presentation. Maybe AI is a big, slow-moving train?

Moghadam is a pillar in the industry with a long track record of success; this is not a knock on him or PLD, rather a recognition that the transformative potential of AI in real estate, particularly in investment decision-making, at the very least has yet to be fully realized. Investors are reluctant to trust or defer to an opaque analytical machine when making buy and sell decisions.

Well, the “black box” era of AI may be ending. A recent study makes a strong case for a certain type of machine-learning-powered valuation model that investors will appreciate. These models not only outperform traditional approaches but also offer transparency into their reasoning, promising a superior and, crucially, explainable approach to real estate valuation.

Importantly, the overall superiority of machine learning (ML) models is becoming increasingly clear. Researcher Tchuente Dieudonné - the study’s author and a business school professor based in France - reviewed 1.5 million residential real estate transactions covering nine French cities over a seven-year period.¹ The work involved comparing two types of automated valuation models: A traditional “hedonic” model, commonly used by appraisers, brokers and investors, and a machine-learning model.

Hedonic models assign value to each characteristic of a property (e.g. square feet, location, amenities, etc.). They assume static linear relationships (every 100 square feet always adds $X to price), they assume each characteristic is independent from the others (number of bedrooms affects price in isolation), and they are limited to structured data (square footage, number of bathrooms, specific amenities). What’s nice about these models is that there’s a known ‘weighting’ of each characteristic when determining the value of a house. That’s transparency.

The AI-powered models are unshackled. They detect non-linear effects (e.g., the value of an additional 2,000 square feet is not simply ten times the value of an extra 200 square feet), find connections between characteristics (large homes without enough bathrooms are valued lower) and can incorporate unstructured data, like satellite imagery and text from property listings (e.g. “fixer-upper”). They can also determine when a specific characteristic is a large part of a property’s market value in one location but a smaller part in another.

The work confirms that AI models are more robust and thus “provide better real estate price estimations,” the research notes. However, they lack transparency, and “this lack of explainability of the outputs of ML techniques leads to a lack of confidence.” This is why most institutional investors are slow to adopt AI models.

So let’s get transparent. There’s a whole field of AI, called Explainable Artificial Intelligence (XAI), devoted to AI models whose results can be traced back to identifiable inputs. The AI used in the study needed to predict the value of each property accurately by figuring out and telling the user how much each property characteristic (e.g. size, location, year build, school district, etc.) contributed to a final price. To do this, it ran a Shapley value analysis, a scientific approach that works like a "What If?" experiment across all possible characteristic combinations.

To get to Shapley values, during training, the AI sees each home (and its idiosyncratic set of characteristics) and each home’s sale value and then runs hypotheticals in which it adds and subtracts features from each home and calculates what the house price would be with and without each factor. With millions of calculations over millions of houses, the AI learns the average impact of each characteristic across all possible combinations and also how each change in a characteristic can impact the value other characteristics contribute. Once trained, the model can assign a customized Shapley value to each characteristic for a house in a given location to predict the market price and give you both the analysis and a clear explanation of how it got to a value.

The learning: In all, AI adoption is slow to happen for real estate investors, partly because we humans are skeptical when we can’t understand the ‘why.’ Developers know this, and with explainable AI models that provide answers and Shapley values, you can get a clear, local estimate of property value “with a clear quantification of the contribution of each input feature.“ As these types of models become more mainstream, investors, lenders, appraisers and looky-loos can get the quality analysis and transparency they want, and with that the technology will garner more trust and adoption.

The Rake

Three good articles.

• Housing: Tariffs & Labor Shortage Driving Costs Up - Multi-Housing News

  The housing market is facing rising costs due to tariffs on materials and a shortage of skilled labor. This combination is significantly slowing down the construction of new homes, creating obstacles that require immediate attention and innovative solutions to keep housing production on track.

• More Good News for Commercial Capital Markets - BisNow

  Commercial property-backed CLO issuance is booming in 2025, with a 374% increase compared to last year, reaching $7.4 billion by mid-March. This resurgence is driven by improved market sentiment and major players issuing new CLOs, as the Fed's interest rate hikes have paused and securities backed by CRE loans are performing well.

• Site-Selection Deadline Looming. Challenges Emerge - BisNow

  States are aggressively vying for these relocations, seeing them as major economic boosts. However, the rushed timeline and lack of GSA resources raise concerns about logistical challenges and potential cost overruns.

The Harvesters

Someone making real estate interesting. They don't pay us for this, unfortunately.

Who: Nestidd

What: Finds, buys, customizes and manages homes leased to care agencies serving people with intellectual and developmental disabilities. They have more than 700 properties and 50 agency partners in the U.S.

The Sparkle: We don’t need to tell you that acquiring and managing residential real estate requires an evolved skill set, and that when handled by experts can be done profitably and efficiently. That fact maps favorably onto a world where real estate is often an off-mission but necessary part of many business models (e.g. colleges/universities, manufacturing, etc.). Nestidd takes real estate expertise off the list of things care organizations need to do, allowing for better use of resources on the care side and a long-term if nichey investment model on the investment side.

From the Back Forty

A little of what’s out there.

Two thousand years ago a group of people built a circular berm 1,000 feet wide with a small break in the middle perfectly situated to watch the moon at a special moment in its 19-year lunar cycle.

Surprisingly, they did this in Ohio, and maybe less surprisingly, it was turned into a golf course. Thankfully, this extraordinary site is now a protected landmark, inviting us to reflect on the mysteries of the past (and the quirks of the present)

Link: https://www.atlasobscura.com/articles/octagon-earthworks-ohio

Editor’s Note: The Real Estate Haystack believes in sharing valuable information. If you enjoyed this week's newsletter, subscribe for regular delivery and forward it to a friend or colleague who might find it useful. It's a quick and easy way to spread the word.

1  Tchuente, D. Real Estate Automated Valuation Model with Explainable Artificial Intelligence Based on Shapley Values. J Real Estate Finan Econ (2024). https://doi.org/10.1007/s11146-024-09998-9

2  https://www.costar.com/article/1393560409/person-of-the-week-hamid-moghadam-prologis-co-founder-ceo-and-chairman