Four key data risks with AI in real estate

The upsides of using artificial intelligence in real estate are manifold, enabling private real estate managers and investors to access data and make better-informed investment decisions more quickly and efficiently. But AI adoption also poses myriad risks for property owners. We highlight four of those risks below.

1 Too much data?

Reliable, comprehensive real estate data can be difficult to access, both because of the fragmentation of the property market by geography and sector and because of the lack of publicly available information. The largest property owners in the world, however, have a lot of internal data on their portfolios – including information on leasing, operating performance and financial metrics – to help supplement the external datasets they obtain.

In addition, more real estate owners are using alternative data, which can encompass data on foot traffic patterns, credit card spending, supply chain data and the location of transit hubs, demographic patterns and hybrid working patterns.

For APG Asset Management, alternative data can also include how many people are within a 30-minute driving distance of a logistics property, or the average Google ratings for stores in a retail mall to gauge how well the asset is performing.

A major consideration, however, is how much data Dutch pension investor wants its digital portfolio manager Samuel’s “mind” to ingest. “There’s vast amounts of data available that we can incorporate into our investment decision process,” says Rutger van der Lubbe, head of global real estate investment strategy. “But all that data has to be maintained, and we need to update our processes, etc. So there’s always a balancing act. In the end, we have to face and realize that making investment decisions will always entail a degree of uncertainty. It’s really just deciding on the trade off.”

2 Cybersecurity

Meanwhile, for Bouwinvest Real Estate Investors, the greatest risk with the use of AI is cybersecurity. “One of the threats is that non-publicly available data becomes available,” such as confidential information or personal data being hacked or accidentally released into the public domain, says Stephen Tross, its chief investment officer of international investments.

In response, Bouwinvest has developed guidelines for the responsible use of AI within the organization. For example, although the use of ChatGPT and other tools from artificial intelligence company OpenAI is permitted, employees can only provide publicly known information while using those tools.

3 Data integrity

A third AI-related risk is data integrity. “You still need to understand what has been generated and how good the information is,” Tross adds. “There’s still a lot to learn in that respect as well.”

To be able to effectively analyze AI-generated information, employees need to be well-trained on AI tools, says author Dror Poleg. “The employee has to have an understanding of the technology, of how that specific tool is built,” he says. “You need to be able, obviously, to be qualified enough to verify, but also to understand where it is coming from, to even probe it with certain questions, and to converse with it in a certain manner.”

Meanwhile, APG has created safeguards in terms of the data and the rules for which Samuel is the repository. “I think it’s also important to understand that it is not a black box, we feed Samuel with our investment principles, and we feed Samuel with the data that we ultimately want to leverage into our investment process,” van der Lubbe says.  

In contrast with an AI tool like ChatGPT, which can either “hallucinate,” meaning make up false answers, or provide answers with no attribution, Samuel also has been developed to cite data sources in its responses, adds Kevin van der Wees, senior quantitative portfolio manager at APG. “Because you want to know what is behind Samuel’s suggestion, Samuel can also give context, ‘I’ve used this information from this source, I’ve used this principle, and that’s how I came to this conclusion,” he explains. “And that’s how you evaluate the answer of Samuel to be correct or not, or sufficient or not.”

4 Bias creation

In September, a class action lawsuit was filed in Seattle alleging property management software company Yardi Systems and 18 property management companies colluded to fix apartment rents in the US with the use of Yardi’s automated pricing software, RENTmaximizer. The lawsuit came less than a year after a class action against pricing software company RealPage and some of the largest residential real estate companies – including managers Greystar Real Estate Partners and Cortland Partners – also alleged the large apartment owners were using RealPage’s algorithm to set the highest possible rents.

The Yardi and RealPage lawsuits point to a lesser-known risk of AI – the use of data or analysis being to create conscious or unconscious bias in decision making, says Vaibhav Gujral, senior partner at New York-based management consulting company McKinsey & Company. “Do you end up inadvertently causing a supply side shock to the system, because everyone is using similar modeling techniques to set pricing?” he asks. “If everyone is using similar models to determine which markets or submarkets are attractive or not, does it lead to a crowding of investment capital and therefore compression of cap rates in certain markets versus others?”

But while there needs to be more transparency around real estate pricing models, “we just haven’t had any form of either industry-led or regulation-led guidance yet on what is appropriate versus not on the use of models,” Gujral adds.

Part of the challenge is real estate being a fragmented, local business. “The competition is so varied in each submarket,” he says. “And we need to be at a point where, it’s hard to precisely say, but at least the majority of transactions are getting informed by [AI]. So we’ve still got a way to go from the low levels we’re at today.”