An algorithm as a venture capitalist
The dominant issue of our time is the coming revolution in Artificial Intelligence and its impact on employment. From self-driving cars to surgical robots, no profession is immune to the threat of automation. Yet, the orchestrators of this disruption—venture capitalists (VCs) who fund these start-ups—seem impervious to digital disruption.
For decades, start-up investors have relied on a mix of intuition and chance to spot winners. But is venture investing a skill that’s difficult for computers to master?
On the surface, the venture industry has generated attractive returns. Investors in these funds expect a “venture rate of return” that is at least two or three times the committed capital. But venture returns follow a long-tail distribution, with only a tiny fraction of funds delivering outsized gains.
The Kauffman Foundation, a think tank and investor in venture capital funds, analysed the returns of their investments over a 20-year period. The foundation found that almost two-thirds of the funds in its portfolio failed to exceed the returns available from public markets. Over that time period, Kauffman’s investments generated a return of 1.31 times, well below the venture rate of return. Their benchmark with other funds also showed that the average venture fund barely manages to recoup investor capital after all fees are paid.
Closer home, there isn’t any publicly available data on India-specific funds, but the sheer drought of exit opportunities hints that the numbers will be even more dismal. The historical performance of this industry would indicate that it is ripe for machine intervention.
But what would an algorithm-based venture investor look like? To design a system that replaces, or at the very least augments, a VC’s output, we first need to understand how they think. A VC’s role encompasses multiple tasks: deal sourcing, deal selection, company advisory, managing exits and fund-raising. Of these, the first two are tasks that have a high degree of analytical component. They are also the most important factors of fund success, contributing almost 60% to the overall return, by one estimate. It makes sense, then, to optimize performance in the sourcing and selection of start-ups.
Unfortunately, while venture investors demand metric-driven transparency from their portfolio companies, their own methods are shrouded in secrecy. Luckily, a comprehensive survey of over 800 institutional venture investors by Paul Gompers of the Harvard Business School and his colleagues sheds some light.
Their data shows that almost 60% of the start-up deal flow was generated from the VCs’ networks. Only a small number, 10% of companies, were inbound. In selecting companies, investors have to weigh many aspects of the company: the team, technology differentiation, product traction, market strategy, competition, etc. Despite considering all this, the authors found that VCs placed a premium on the team above all else. Almost half the investors ranked the management team as the most important factor when deciding on an investment. Business-related factors were rated as most important by only a third.
This makes sense. Identifying young, innovative companies isn’t easy. Frequently, the teams are very small and the company has no market presence. Given the limited size of most VC firms, associates and partners limit their funnel to a few geographic (Bay Area, Bengaluru) or institutional (Stanford, Indian Institute of Technology) clusters. Betting on the team is also a good hack since most successful start-ups go through multiple pivots before finding product-market fit.
But machines can do better.
Take deal sourcing. Today, founders who aren’t part of popular clusters find it extremely difficult to get noticed. But the ubiquity of the cloud also means that great ideas can be implemented from anywhere. Machine-learning algorithms can be used to track thousands of new websites that are launched daily. The popularity of these sites or the products they offer can then be compared, starting from basic profiling like web rankings and moving to more complex indicators like product mentions. Companies that score highly on these metrics can then be shortlisted for a closer look by the VC. The pool of investible companies grows from a few hundred to hundreds of thousands.
In deciding which companies to fund, there is scope to move beyond human-centric rules as well. A machine can be trained to identify patterns among start-ups by feeding it data on thousands of start-ups from the past. This training set yields a set of parameters that can be applied to any new company. But rather than simple rules like “if from this college and this past company, then... ”, the algorithm can process hundreds of data points to make a recommendation on viability. Of course, one has to be careful not to replicate human biases by feeding the system partial data.
This type of “robo VC” isn’t far-fetched. Google is famous for using a data-driven approach in its analysis. Funds like Correlation Ventures and SignalFire have built their investment thesis on a machine-human hybrid. Yet, compared with their financial cousins in hedge fund and private equity, most venture capitalists remain wedded to their human intuitions.
Given the mediocre performance of most funds, it’s time for venture investors to imbibe the same mantra of innovation that they evangelize.
Shailesh Chitnis is head of product at Compile Inc., a data intelligence company.
Comments are welcome at firstname.lastname@example.org