Data analytics outperforms top investors in picking startups, MIT researchers claim

Venture capital is often thought of as a practitioner’s business, where hustle, research, technical expertise and a nose for investing combine to sniff out the best startups.

Could be this axiom is ready to change.

That is according to a pair of Massachusetts Institute of Technology researchers, who say they came up with an analytics-driven formula for identifying successful startups that outperforms some of the best investors in the business.

“We didn’t reinvent the wheel here for venture capital,” said Tauhid Zaman, a professor at the MIT’s Sloan School for Management. But the work “shows the power of analytics in the venture capital space.”

What Zaman and graduate student David Scott Hunter developed is a scoring mechanism for startups that they used to build a hypothetical portfolio. The portfolio they put together of top scoring 2011 companies had an exit rate of 60 percent, when more typical venture hit rates are 20 percent to 30 percent.

Zaman says he’s been contacted by several venture capitalists since the work, “Picking Winners: A Framework For Venture Capital Investment,” posted online, including one apparently interested in raising a fund for analytics-driven investing. He declined to name the investor.

The work, which isn’t the first attempt in the industry to apply data analysis and predictive modeling to selecting startups, was submitted in June for publication in the journal Management Science. Over the past five years, firms such as Correlation Ventures and SignalFire have raised considerable capital to test their methods in which they’re mining their databases to identify target companies. SignalFire launched in 2015 and Correlation raised $200 million for its second fund in January.

Zaman and Hunter began their effort by assembling a list of 24,000 companies founded between 2000 and 2016, where reliable information could be determined on seed and Series A rounds. They also gathered information on 558,000 startup employees.

Then they began digging to determine which variables would best predict whether a company succeeds or fails, looking at, among other factors, sector focus, founder experience, founder education and investor track record. Their effort included examining whether founders had the experience of a previous company and the schools they attended.

Then they pumped the data through their analytical engine.

“You can actually do very well picking winners in venture capital,” Zaman said. “It’s a space where data can make a difference.”

One key observation they came up with is not entirely unexpected. The top factor for predicting the success of a company was whether a founder had a previous company that completed an IPO. Second on the list was simply whether a founder had a previous company, followed by whether the previous company was acquired. The fourth most important attribute was whether a founder attended a top school.

The 2011 portfolio the pair assembled based on their work had six companies with exits: Shift, Jibbigo, Nutanix, Friend.ly, Jybe and LaunchRock; and four without: Sequent, PowerInbox, MediaRoost and CloudTalk.

A 2012 portfolio had four exits: Metaresolver, ViewFinder, SnappyTV and Struq, and six companies that did not exit.

Yet for all their success, one hole in the work remains. Zaman said he could not determine the quality of the exits or their return multiples because pricing and other details of small purchases are often kept private. “That’s one drawback to our data,” he said.

Nevertheless, he has big aspirations. He said he is looking for a partner to work with to place bets on startups, eager to prove the formula in the real world.

Long term he hopes to convince GPs to change the way they do business by bringing in analytics.

The result could be more efficient – and profitable – capital allocation.

Photo of crystal ball foretelling future courtesy of VallarieE/iStock/Getty Images