Investing in biotech startups is a laborious business.
Venture capitalists spend months poring over data before deciding whether to back a fledgling company. They scour the scientific literature for signs the core ideas are valid. They call in the founding team for interminable grillings. They may even hire outside researchers to try to replicate key experiments.
Then there’s Correlation Ventures. The San Diego firm has a simple — and audacious — approach: The partners assign a numerical value to every element of a company, from its CEO’s résumé to the track records of its backers.
Those data get punched into an algorithm, and the computer spits out an answer: Invest. Or don’t. The whole process takes less than two weeks, guaranteed.
“It’s kind of like counting cards in blackjack,” Correlation cofounder David Coats said. “You may lose any individual hand, but if you play enough hands, you should win. The odds are truly in your favor.”
Other venture capitalists love Correlation’s database, but aren’t so sure about picking winners by algorithm. Data are “a helpful tool,” said Tim Shannon, general partner at Canaan Partners, “but can you purely model success or failure? Obviously not.”
Here’s how it works: Correlation, which has been in business since 2012, has assembled a sprawling database of more than 80,000 venture capital deals, encompassing roughly 98 percent of all US investments dating back to 1997, Coats said. Every facet of each deal is given a numerical score and fed into the database. Zoom all the way out, and tendrils of data form patterns that connect the winners and freeze out the losers, Coats said.
And that leads to Correlation’s proprietary algorithm, which the firm’s partners believe allows them to predict the future of new companies based on the fates of ones past.
Only about 10 percent of potential investments pass the test. In recent years, the algorithm has led them to make investments — generally between $1 million and $4 million — in a wide range of biotech and tech companies, including a drug developer out to treat cancer, a health IT company hoping to make doctor referrals more efficient, and even an online mattress retailer.
“Everything is entirely empirical,” said Coats, who has been investing in biotech for about 20 years.
But relying on data, no matter how big, in lieu of old-fashioned instinct and intuition remains a divisive idea in the financial world. “If you’re a sports fan, this is reminiscent of the analytics folks in baseball versus the old school,” said Gary Pisano, a Harvard Business School professor who authored a book on the biotech industry.
The promise of a dispassionate investing machine, free of human hubris and its resulting biases, has an undeniable allure. But such a machine could only be as clever as its flesh-and-blood creator, said Christian Catalini, an assistant professor at the Massachusetts Institute of Technology who studies the intersection of tech and finance.
“One of the big challenges of any machine-learning approach to this is you need to make sure you’re not overfitting the data,” Catalini said, warning that a poorly designed model can be fooled into seeing a signal amid a lot of useless noise.
As for the predictive value of deals cut two decades ago? “What may have been true in the past is not necessarily true for the future,” Catalini said.
At Correlation, though, the partners implicitly trust the computer. They say having 20 years’ worth of data means their algorithm accounts for both booms and busts. Their internal rules let them vote “no” when an algorithm suggests they invest, but under no circumstances can they fund a company that doesn’t impress the computer model.
Despite their loyalty to the algorithm, the partners don’t aspire to render human VCs obsolete. In fact, the firm needs them in order to survive.
Correlation is strictly a co-investor, meaning it participates in funding rounds only if a blue-chip VC has already put money down. And the track record of that investment firm is a particularly weighty data point for the algorithm.
So, how’s it working out?
That’s hard to say, because Correlation doesn’t disclose its returns. Coats said the firm is outperforming its benchmarks, which are tied to the stock market. And the firm’s investors seem to be pleased: Putting together the firm’s $167 million first fund in 2012 was a slow process, Coats said, but Correlation speedily closed a $200 million fund-raise in January.
Correlation’s biotech track record includes some sizable successes, at least when seen through the lens of return on investment, rather than scientific breakthroughs.
It invested in Flex Pharma, which pulled off a stellar $86 million initial public offering in 2015 (though the company’s lead drug, to prevent muscle cramps, failed in its first major clinical test). Mirna Therapeutics made a hefty $44 million Wall Street debut the same year (though it later had to halt a Phase 1 trial of its oncology drug because of multiple severe side effects).
But the algorithm isn’t always right. Aldea Pharmaceuticals, which Correlation backed in 2013, later went out of business when its drug to reverse alcohol intoxication didn’t pan out. The same fate befell Enlibrium, another Correlation portfolio company, when its efforts to treat cancer came to naught.
A few flameouts of high-risk, early-stage biotech companies are inevitable, Coats said, adding that individual failures haven’t shaken his faith in Correlation’s approach.
The firm’s technology traces its roots to financial crime. Anu Pathria, Correlation’s head of analytics, is the co-inventor of FICO’s credit card fraud-detection system, and Correlation relies on a similar principle: Just as a computer can learn your habits and discern when you’ve gone on an uncharacteristic spending spree, the firm expects to recognize the patterns that separate promising startups from an ocean of also-rans.
The firm openly shares analyses generated by its model with other investors. Partners at Atlas Venture, Alta Partners, and Canaan Partners said Correlation’s trove of data is a valuable resource.
But none is worried about getting displaced by Coats and his army of numbers.
“Not everyone’s going to turn around and become Correlation Ventures,” said Jay Lichter, managing director of Avalon Ventures, which has co-invested with Coats’s team. “At some point, someone’s got to stick their neck out.”
Then there’s the other work VCs do, like sitting on a company’s board, making hiring decisions, and steering strategy.
“I don’t think you’d ever get that with an algorithm,” said Dan Estes, a partner at Frazier Life Sciences.Damian Garde can
be reached at email@example.com.