Bob Knakal has personally sold more than 2,400 buildings across a 42-year career in New York City investment sales. In April 2024, seven weeks after JLL let him go, he started over and named the new firm the world’s first AI-driven commercial real estate brokerage.
The AI label gets the headline. What actually makes BKREA work is what’s underneath it: a 42-year, 2,470-deal proprietary data set nobody else has, and a canvassing project born out of an empty Manhattan during COVID that now wins pitches at a rate few brokerage firms in the country can touch. We sat down with Knakal for the AI for CRE Collective podcast to walk through how he built both, and what any broker running a smaller book could copy.
A generalist model built for a market that doesn’t exist anymore
For most of Knakal’s career, the model was simple. He co-founded Massey Knakal in 1988 out of an 800-square-foot sublet with one partner and a secretary, sold every type of building in a defined geography, and represented sellers exclusively. It worked because market information was opaque. Knowing which buildings were trading, at what basis, and who owned what took a broker with boots on the ground and a Rolodex nobody else had.
That advantage is gone now. Ownership records, comps, and contact data are a search away for anyone. So when Knakal started BKREA, he inverted the model that took Massey Knakal to 250 people and the number one investment sales ranking in New York for 14 straight years before its sale to Cushman & Wakefield in 2014. Instead of a generalist working a wide territory, BKREA sells one thing: vacant buildings in New York City, headed for one of three outcomes, demolition and new construction, a gut renovation to a higher use, or a sale to a user group that will occupy it.
“It’s all about being the market expert,” Knakal told us. “Pick a niche that is narrow enough that you can know every single thing about it, but big enough you can make a great living doing it.”
Hiring a tech co-founder before hiring anyone else
The first person Knakal brought on wasn’t a broker. It was Seth, a co-founder with six years in driverless cars and a background building AI companies, hired specifically because Knakal describes himself as someone who knows nothing about tech. He points to the book Who Not How as the operating logic behind the hire: when an idea is good, the question isn’t how you’ll execute it yourself, it’s who you’ll get to execute it for you.
That decision shows up in three places at BKREA, by Knakal’s own account. The most obvious is content: neighborhood write-ups and marketing copy that used to take an analyst an afternoon. The second is what he calls funneling, the prospecting and deal-execution sequences that move a list from thousands of contacts down to the handful who actually transact, now run through drip campaigns, text programs, and scheduled hard-mail pushes instead of a broker manually tracking who’s due for a follow-up.
The third is the one Knakal says almost nobody in the industry is doing right, and it’s the one BKREA has built its identity around: using AI to find predictive patterns in market data that’s actually trustworthy.
Why 42 years of data beats one aggregator’s spreadsheet
Most CRE data, in Knakal’s telling, is close to useless, not because it’s fake but because nobody agrees on how to count it. He described three different heads of research at his old firm, each tracking transaction volume with a different threshold. One counted every sale over ten million dollars, the next dropped it to five hundred thousand, the third moved it to five million. Comparing those totals year over year, he said, is like tracking a baseball team’s home runs if one season only counts games on Tuesdays and Thursdays.
BKREA’s answer was to build its own data set on one methodology, unchanged, since 1984. The firm spent three years compiling a study of every development-site sale in Manhattan south of 96th Street going back four decades: 2,470 transactions, split into five buckets by end use (residential rental, condo, hotel, office, and a catch-all for healthcare and education), then run against 33 macroeconomic factors, including interest rates, inflation, the S&P 500, and the price of gold.
They didn’t stop at correlation. Because a deal that closes today was usually negotiated months earlier, the team built a lag into the model, so a spike in the price of gold today gets measured against sale prices from roughly when that spike would have actually shaped a negotiation, not the day it closed. The output is a set of findings nobody else has: corner development sites carry a 16.6% premium over mid-block sites on average, and that premium is sharpest for office sites and nearly nonexistent for rental apartments. Compare that to how the rest of the market reports land pricing, as one blended average like last year’s $520-a-foot figure, and the gap in usefulness is the whole point. Knakal’s own comparison: that’s like averaging the price of a peach, a bowling ball, and a two-by-four and reporting the number as if it means something.
220 hours walking an empty city
The second data asset BKREA runs on didn’t come from a spreadsheet. It came from March 2020.
Knakal’s team was sent home when the pandemic hit, and a rumor that Manhattan might be sealed off sent him and his family to Connecticut within the hour. Two weeks later, he drove back into the city for belongings and found the Upper East Side empty. No cars. No pedestrians. Not even a stray dog. New York had no reliable construction pipeline data at the time, and two research firms were citing condo counts thousands of units apart because nobody agreed on when a project starts or stops counting as “under construction.” So Knakal decided to count it himself.
He called his team, had them print copies of the Sanborn tax lot map, and spent roughly 220 hours over the following weeks parking, walking a few blocks on foot, and hand-highlighting every demolished site, every site under active construction, and every site built to less than 25% of its allowable density. He taped the sections together into a single map that now runs 24 feet long. BKREA still updates it today by tracking every demolition, foundation, and building permit the city issues.
Knakal has been told repeatedly to digitize it. He won’t. Clients walking into a room with a 24-foot hand-colored map get what he calls sensory overload, in a way a screen can’t replicate.
Ninety-seven wins out of ninety-nine pitches
Over roughly 10,000 pitches across his career, Knakal’s win rate has averaged 26%. Since building the Map Room, BKREA has given 99 pitches and won 97 of them, a rate close to 98%. The firm now runs about 20 people, none of whom had prior New York investment sales experience before joining, and closed 43 deals worth $1.8 billion last year. It currently holds roughly 70 active exclusives worth just under $4 billion.
What the data still can’t do
None of this replaces what Knakal still calls the number one revenue-generating activity: picking up the phone. He’s blunt that AI won’t replace a broker, because buildings aren’t widgets, and a functioning market still depends on people building trust with other people. What AI removes is the busywork around that relationship, the three-minute gap between a missed call and the follow-up voicemail, text, and email that used to eat a broker’s whole afternoon.
Social media has become the same kind of channel in his account, not a vanity exercise. He closed two pieces of business directly from social media in the week before we spoke, including one seller who told him ChatGPT pointed straight to Knakal’s name when he searched for who handles air rights sales in New York.
BKREA is preparing to make its 42-year land study public. The map stays exactly as it is: on paper, and only in the room.
We publish stories like this every week in the AI for CRE Collective, breaking down exactly how brokers, developers, and investors are building a real edge with AI, not just saving a few hours. Join us to see what’s working before it’s obvious to everyone else.
