Automated Venture Capital— Can machines systematically pick the best startups?
The rise of the machines in finance has long taken place. So-called quantitative funds increased their assets under management to almost USD 1 trillion in 2018. They fiercely compete to develop the best systematic approach to trade assets based on patterns found in large amounts of data. Some VCs are excited about this idea — but how viable are such approaches really for venture investing?
Machines have only timidly reached venture capital territory due to the lack of fuel they require: large quantities of properly labeled data. In private markets, data is scarce and asymmetrically distributed among participants. Private companies don’t publish standardized data that algorithms can easily process.
Hence the decision which startups get funded is still a human judgment call by intermediaries (VCs) who have privileged information and access to early-stage companies via their trusted networks. But technology is pervasive across sectors and time. VCs know this best since they invest in technology-driven and highly scalable businesses. Since data is increasingly available for private companies, VC firms are investing in software to help them find and, in some cases, pick tomorrow’s winners.
Calling first
Venture firms like EQT Ventures, eventures or SignalFire have hired engineers to leverage data (such as page ranks, social media dynamics, Linkedin status changes, etc…) notably for prioritizing and discovering startups across geographies. These engineers develop tools to categorize companies across sectors to help triage and signal “trending” companies in different categories. This should give these firms the opportunity to engage with the right companies at the right time and start building a trusted relationship with the founders before others investor do so.
This approach helps scale purely network-driven referral systems, inherently limited in size and gives a first small edge to the first investor calling. However, since many strong early-stage companies stay in “stealth mode”, hiding until they launch, this makes it impossible for any sophisticated data monitoring machine to find them in advance.
Picking winners
Ryan Caldbeck from CircleUp, a US firm investing in early-stage consumer brands strongly believes that machine learning (an algo named Helio) can help them invest in the next winners in the consumer goods space more systematically and potentially, one day even automatically (here and here are his tweetstorms on this). Social Capital has experimented with funding startups with USD 0.5m based solely on metrics demonstrating product-market/fit. For that, entrepreneurs have to submit their numbers to the “diligence engine”; no meetings, raw data only, automated “capital-as-a-service”. Unsurprisingly, it also has been reported that G/V (Google’s venture arm) was delegating new and follow-on investment decisions to “the Machine” based on patterns found in historical data and other inputs (such as co-investors, market data, historical valuation, etc…) fed to it by the investment team at G/V.
For systematic picking, two ingredients are required: a) standardized business models and b) large quantities of data. In consumer goods (CPG), where CircleUp is active, the business models are simple (create a branded product that you sell at a margin via retailers) and the amount of historical and comparable data is significant. In Software as a Service (SaaS), the standardization of benchmark metrics (even if these are not publicly and easily available to use) makes the concept of a systematic VC fund realistic, hence the idea surfaces more and more often.
However, at the very early stages, metrics are scarce and business models are new and evolving fast. This makes it extremely difficult for machines to gain a foothold. Last but not least, many of the best VC bets don’t fall into any established historical categories. Returns are unevenly distributed and mostly come from “outliers”.
Technology to support founders
Follow patterns too closely and you miss such outliers. Also, not every pattern is equally helpful. A recent analysis shows that many founders of billion-dollar startups have worked for Google, Oracle or IBM previously. Furthermore, the most frequent star founder names are Rob, John or Brian.
At the seed and early stage, personality traits required to build new large businesses from scratch are scrutinized: resilience, leadership, high learning capacity, intrinsic passion for the problem the founders want to solve. These traits are obviously not easy to test, especially for first-time founders. It is indeed hard to get sufficient standardized information of these traits across founding teams at scale.
Looking at the applications and the requirements for a quantitative VC fund, it seems that consumer goods or Series A and growth SaaS are good candidates. For the rest, having a technology-empowered investment team to source efficiently also outside the traditional paths globally is a convincing way to enrich the origination process.
Ultimately, VCs should leverage technology to help founders systematically uncover and win sales opportunities as well as attract the best talents post-transaction. VCs who truly serve entrepreneurs and do even better with the help of technology will certainly have an edge in the long run. They will most likely get the first call from the best founders, even if a machine directed others to them first.
Thanks to Eugen Stamm who co-authored this article (also published here: https://www.investiere.ch/blog/can-machines-pick-the-best-startups/)
