Skip to Content

An Algorithm to Pick Startup Winners

A venture capital firm throws out intuition and uses computer models to determine investments.

Aldea Pharmaceuticals, a biotechnology startup developing an emergency treatment for alcohol poisoning, seemed like an attractive investment to venture capitalist David Coats. But he didn’t rely on a hunch—he consulted the computer model he’d built.

Investment model: Wenjin Yang, research vice president at Aldea Pharmaceuticals, got funding thanks to software suggesting that his company’s method for speeding up alcohol metabolism was a good investment.

Two weeks and a few phone calls later, he cut the company a $1.25 million check. “A decision like that would have normally taken a minimum of three months,” says Tim Shannon, a partner with Canaan Partners, the firm that had led Aldea’s $7 million fund-raising round.

The $1.25 million was a follow-on investment from Correlation Ventures, which calls itself a “new breed of venture capital firm”—one driven by predictive analytics software built over the last six years by founder Coats and his partner Trevor Kienzle. The effort adds efficiency to the investment process. And for entrepreneurs, it means far faster answers: rejections come in as little time as two days.

To run its model, Correlation Ventures, which is based in San Diego and Palo Alto, California, asks startups to submit five basic planning, financial, and legal documents. It enters these into a program similar in function to credit rating software.

A top-ranked score leads to a 30-minute interview with both the startup CEO and the outside venture firm leading the investment, plus a quick legal review and background check. As a co-investor, Correlation Ventures always relies on some vetting by the primary investor.

Correlation Ventures will then often deliver a check from its $165 million fund, closed last November, in less than two weeks. “That’s unheard of in the venture industry,” says Coats.

Once it makes an investment, Correlation backs off and doesn’t take a board seat. That policy is itself data driven: the firm’s analytics show that companies with more than two VCs on the board are less likely to be successful.

What’s not yet clear is whether this system works. Correlation Ventures has so far invested in 26 companies in diverse sectors but says it is too early to report successes or failures.

None of this might have been possible a decade ago. Harvard Business School professor Matthew Rhodes-Kropf, who advises Correlation Ventures and is now an investor in the fund, says the venture capital industry has only recently worked through enough business cycles to look for subtle trends.

There was also no complete, accurate, public set of venture capital data, so Correlation Ventures hustled for it. To build and maintain its database, it partnered with Dow Jones, scoured the Internet, signed nondisclosure agreements with more than 20 venture funds to see their internal statistics, and called hundreds of companies.

While so-called Big Data companies have attracted plenty of investors, the reputation-driven venture capital industry itself has yet to embrace their tools. (There are exceptions, such as Google Ventures, which uses quantitative analysis to help guide decisions.)

Rhodes-Kropf says venture capitalists are good at identifying companies that will have the best chances at success but not as good at predicting which will be the next Facebook. And one finding from Coats’s research is that while top-tier firms do invest in a disproportionate share of “winning” companies, the majority of successful investments are led by firms that don’t even crack the top 50. So it makes logical sense for Correlation Ventures to focus equal time and energy on many companies and co-invest with a diverse set of venture capital firms, he says.

To explain his project, Coats cites Moneyball, the book and movie about how Oakland Athletics general manager Billy Beane rejected the conventional wisdom on evaluating baseball players and built a winning franchise by letting a computer tease out variables that others overlooked. He believes the averages will work out. “We’re not claiming to have a magic crystal ball,” he says. “We’re tilting the odds a little in our favor with each investment.”

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

The problem with plug-in hybrids? Their drivers.

Plug-in hybrids are often sold as a transition to EVs, but new data from Europe shows we’re still underestimating the emissions they produce.

How scientists traced a mysterious covid case back to six toilets

When wastewater surveillance turns into a hunt for a single infected individual, the ethics get tricky.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.