CRE giant BGO uses AI to find undervalued assets in unlikely areas

A version of this article first appeared in the CNBC Property Play newsletter with Diana Olick. Property Play covers new and evolving opportunities for the real estate investor, from individuals to venture capitalists, private equity funds, family offices, institutional investors and large public companies. Sign up to receive future editions, straight to your inbox.

John Carrafiell, co-CEO of BGO, a global real estate investment manager with $89 billion in assets under management, takes great pride in the fact that he sits right next to his chief data scientist. 

Investment strategy, whatever the market, has always relied on research and data, but artificial intelligence has taken that to a whole new level, transforming investment research models developed just a few years ago and putting them on steroids. 

Carrafiell, who has been in the real estate business for roughly 40 years, said he was increasingly frustrated by the sector’s research and data methodologies, which he said really hadn’t changed at all over those years. Everyone seemed to be looking at the same information and coming up with the same conclusions. The question he said he kept asking himself was, “How do we really outperform?” 

The answer, he found, was to analyze all of his firm’s past deals going back 20 years, using just a computer model and taking the human element out of it. What the model found was that outperformance or underperformance was determined fully by the local market that was chosen for the investment. 

That may sound trite — given that real estate’s mantra has always been “location, location, location” — but the results told his team to focus almost entirely on local market fundamentals when choosing its future investments, and not so much on property pricing and national economic trends. 

There are, of course, research firms that analyze and rank local real estate markets, but BGO found their results to be somewhat random, according to Carrafiell. Instead it looked to its own past and built a model that backtested exactly what drove its best and worst performance. The model includes all sorts of local market data points, including demographic and supply trends unique to each location. AI then gave that model increased data volume and velocity. 

“We have taken thousands of data inputs, many that are free from the government, many we have to buy from, for instance, telecom providers, great data. We have found the key,” said Carrafiell. “And we know it’s accurate because we backtest it.” 

BGO used its data science to inform a decision to invest in an industrial development in Las Vegas with partner Northpoint Development. Other data models suggested it wasn’t a particularly good investment. 

Carrafiell said the “best research out there” indicated the investment would be mediocre in terms of performance and returns. 

“But our model was screaming, it is going to explode. We underwrote $5.88-per-square-foot rents. We’ve gotten rents in the $9-per-square-foot range,” he said. “That does not happen in commercial real estate. That is not luck.” 

The model, he explained, saw that the Inland Empire of California was getting too expensive, then analyzed logistics routes. It found that companies could save big by being in Las Vegas instead, where both the rents, taxes and labor were cheaper. 

“So you had an extra two-hour drive, but you saved like 60% on your total cost, and that’s what the model saw,” Carrafiell said. “The tenants we have there are serving an entire region. They’re not serving Las Vegas.”

BGO ran similar analytics for investments in Florida and the Rust Belt, resulting in big returns on its investments. 

“We think our performance has materially increased as a result of this model,” said Carrafiell. 

But he admitted that although the model’s accuracy is improved dramatically by artificial intelligence, it can never be totally accurate, hypothesizing, “Boeing can move out of Seattle, and the model can’t predict that, right? There could be idiosyncratic things.”

While BGO’s investing team focuses on the upside models for potential properties, its lending team looks at the downside modeling, because therein lies its risk. 

New iterations of the research model down the road will include asset allocation to different sectors of commercial real estate. The model would ideally suggest an optimal portfolio mix. The possibilities are still growing, which is why Carrafiell says he’s dialed into the data like never before. 

“AI is an enhancer and an accelerator that allows us to do so much more, but it’s really data science,” he said. “It’s [like] a six-person, dedicated data science team that is sitting next to your CEO and next to your asset management and acquisitions team.”