Arthur Rock, one of the legendary pioneers of the US VC industry, believed in investing in people. He wore out a lot of shoe leather finding founders who had “fire in their belly” and uncommon intelligence.
“I’m not enough of a technologist to be able to understand what most of these entrepreneurs are about technically. And the way I went about it was to spend a lot of time with these would-be entrepreneurs,” he told a Californian audience at the Computer History Museum in 2007.
Rock’s instinctive methodology didn’t serve him badly. Between 1961 and 1968 his VC firm returned $100m on the $3m it had invested.
But Rock’s approach is almost the exact opposite of what some pioneering AI-powered VC investors are doing today. Indeed, this new generation of investors explicitly rejects any reliance on personal chemistry and human intuition. One of the most extreme examples is QuantumLight, which has to date backed 17 startups selected entirely by its algorithms. Last week it closed a $250m funding round.
Launched in 2022 by Revolut founder Nik Storonsky, QuantumLight aims to eliminate human judgement from the investment process. It claims to be the world’s first truly systematic VC and growth equity firm having built its own proprietary model to track 700k VC-backed companies and analyse 10bn data points.
“We just believe that machines are able to do it better. Not only do they have perfect memories, they are also not swayed by emotion, by fear of missing out on a certain hyped deal,” Ilya Kondrashov, QuantumLight’s CEO tells me in a Zoom call from Dubai.
The objective is not to find some sort of magic variable that perfectly predicts the success or failure of companies, Kondrashov says. Instead, the fund optimally weighs up all the data it can find on a company and compares them against previously successful companies to see if they look similar. “So, it’s a kind of pattern recognition game,” he says.
To date, the fund has invested about $80m but is doing deals at a rate of one a month and so will probably exit the year with about $150m of capital deployed. Crucially, it tends to invest at a later stage than Rock after startups have created a “digital footprint.” It typically invests about $10m in a Series B round.
QuantumLight is not alone in playing this algorithmic game. Pretty much every VC fund is now crunching the numbers to see if they can find signal among the noise. But imperfections in the data sets would appear to be one big limitation. As we know, the startup world in the US and Europe skews towards the privileged and the networked. Might these data sets not just exclude individual and corporate outliers who come from non-traditional backgrounds?
“Just because you have a lot of data does not mean you will be successful. It’s about the quality of the data and how you analyse it,” says Ilya Strebulaev, professor of finance at the Stanford Graduate School of Business and author of The Unicorn Report. “By itself AI is not a panacea.”
Given all the blood, sweat and tears involved in launching a startup, there’s something almost spooky about an algorithm inhumanly determining who does and does not receive VC funding. The whole premise of QuantumLight, as Kondrashov explains it, is: “Whether we like the founder or not, we should just do what the statistics tell us to do.”
However, QuantumLight still strongly believes in the power of people once it has made its investments. Storonsky is keen on sharing the experience he gained from scaling Revolut to more than 10k employees in a decade. QuantumLight is now making the Revolut “operating system” both visible and replicable and encouraging founders who take its money to follow its driving high performance playbook. It is now releasing its second management manual on hiring top talent.
Clearly, it is too soon to judge the success of QuantumLight’s approach - although it claims its investment performance stacks up well against other funds of the same vintage, based on Cambridge Associates benchmarks.
But quant investing has proved it can sometimes work spectacularly well in data-rich public financial markets. Can it do the same in comparatively data-poor private markets? It’s going to be interesting to see.
