BGO Is Using AI To Rewrite Its Investment Thesis

California’s Inland Empire industrial sector had been one of the best-performing real estate markets in the world for a decade, and investment manager BGO had built up a $1B portfolio of it. Market consensus was that stellar growth would continue.

But just over two years ago, the firm had a decision to make. 

Its artificial intelligence-enhanced data model was telling it the consensus was wrong: The market was a sell, with vacancy set to rise and rents to flatline. So in spite of its success, BGO sold the lot and reinvested in industrial assets in other metros where AI said growth would be higher. 

“I think a lot of players in the sector saw that as a strange move,” BGO co-CEO John Carrafiell told Bisnow. “Why are you selling the best-performing portfolio and market for the last decade-plus?”

The call proved right, as the Inland Empire cooled. That showed Carrafiell that BGO, one of the world’s biggest real estate fund managers, with $89B in assets under management, was savvy to overhaul its investment processes over the past five years to put AI and data at the center of its decision-making. 

“I think it's really important as investment professionals to park your biases at the door when you walk into work,” he said. “We're not perfect, but it's very comfortable to continue doing what you've been successful doing before. It's not human nature to go against that, whether it's a sector, type of asset, whether it's a market.”

BGO has been one of the most forthright major real estate investors about how it uses AI in its investment strategy — a dominant topic of conversation, given the speed at which AI capabilities are increasing — and Carrafiell gave Bisnow detailed insight into how the technology is boosting its returns. 

“This business is about performance. Period,” Carrafiell said. “For us, performance means generating alpha and reducing risk, and our proprietary data science capabilities have become a transformative competitive advantage.” 

Carrafiell has been investing in real estate for 40 years, and for most of that time, buying and selling has been done in the same way — investors use the same metrics, the same data, the same processes as the late 1980s. 

“It’s not that those metrics aren't important, but it can lead to groupthink,” Carrafiell said.

To avoid hive mind investing, BGO in 2020 began hiring data scientists, including Chris Liedtke, now managing director and chief data scientist at the firm, to work on a data model to guide investment decisions. 

BGO fed tens of thousands of demographic, market and historical data points on every investment it had ever made into a computer model and used it to analyze which factors and metrics determined whether an investment was successful. 

The AI found that for core and core-plus deals in particular, upward of 70% of performance comes from being in the right market. Investment professionals have always known location is important, Carrafiell said, but historically the sector has underappreciated the importance of which submarket you invest in. 

“If we blindly tested investment professionals, they probably would have said market was something like a third, and more important was the in price,” he said. “In price is of course very important, but if you got a great in price and it was a market where rents declined, you had a pretty lousy investment.”

Armed with that insight, BGO has created a model that ranks 400 U.S. metropolitan statistical areas according to their expected performance for different property types. 

If a property is in a low or middling market according to its rankings, the company will only buy at an incredibly low price. 

The model comes into its own when it allows the company to be contrarian, Carrafiell said — where the real estate market generally thinks a region will underperform but BGO’s model says it will be a winner. Or, like the Inland Empire portfolio, the market thinks good performance will continue, but its model indicates fundamentals are about to take a turn for the worse. 

Carrafiell was cagey about the specific data BGO uses, but he gave some examples when it comes to residential investment. It is well-established that the quality of education in an area determines whether people want to live there. But the company can now go further in its analysis, taking into account factors like inward migration forecasts and how educated those people are or what their earning potential is. 

BGO is on the fifth iteration of its model, Carrafiell said, and it continually tweaks the inputs it uses. The increase in computing power over the last few years has allowed it to input more data and produce analyses faster. 

The company has made a significant investment in its data science team, in acquiring data sets and also paying to use data center-based cloud computing systems that use top-end Nvidia AI processing chips.

Even so, the quarterly process of rerating those 400 markets can take an entire weekend. BGO uses a variation of a Monte Carlo simulation, where a computer runs thousands of scenarios through a model to determine the most likely outcome. 

Insights provided by the model are incorporated into decision-making at an early stage, whether BGO is evaluating new investments or when to sell existing assets. 

“You need buy in from the top down and from the bottom up,” Carrafiell said. “This is not a tool that sits on the side of our investment teams’ desk and once a week they might look at it. We've actually seen how powerful it is and how instrumental it is in generating alpha and outperformance.”

Carrafiell said BGO is a step ahead of peers that might only be starting to incorporate data and AI now. And it is hoping to extend the type of analysis its model can undertake. 

BGO is examining ways to use data to more accurately predict how individual assets will perform in the future, in the same way it does now for wider locations. It is looking at ways of taking the data it has for individual real estate sectors and using it to determine the optimum allocation among asset classes to create a long-term core or core-plus portfolio that overperforms.

Increasing computing power means both should be possible soon, Carrafiell said.

At the end of the day, the data model BGO uses doesn’t eradicate traditional investment skills, he said — it just makes sure an investor’s biases don’t get in the way of the facts.

“You’ve got to pound the pavement, you’ve got to walk the market,” Carrafiell said. “You walk the competitive buildings, you do your due diligence on what you're acquiring. You underwrite thoroughly. The day job does not change at all. You just have an extra edge.”