Artificial Intelligence and Venture Capital

How being data-driven is changing the investment process

“On your way out, make a comment about automating Venture Capitalists” (Robbie Allen).

I. The rationale

Great startups very often stem from a simple idea and a couple of people that identify a specific need and a solution to address an existing market gap. On the other side of the fence, there are angels and venture capitalists (VC), who provide the capital and sometimes the necessary help to make that dream become a reality.

If you are an investor then, there are two things you want to know all the time: where the good companies are (scouting abilities) and how to know whether a specific company will be a good investment or not (cherry-picking ability). The scouting abilities are usually refined through years of networking and branding efforts, while cherry-picking skills are often emerging from pattern-matching, i.e., seeing company over company year over year and trying to intuitively infer and deduce why a startup succeeds while others do not.

As it is clear, the VC investment process is incredibly slow, labor-intensive, inefficient, expensive, and often even biased (Sheehan and Sheehan, 2017). Many venture capitalists suffer indeed from common psychological biased such as overconfidence (Zacharakis and Shepherd, 2001); availability biases (over-weighting information that comes easily to mind because memorable while underweighting information that is less exciting); information overload (Zacharakis and Meyer, 2000), meaning that more information often leads only to greater confidence and not to greater accuracy; halo effect (how similar this company is to previous exits I had?); survivorship biases; representativeness, which means ignoring statistical information in favour of a narrative; confirmation bias (accepting information that support pre-existing beliefs); and similarity biases (meaning not simply that entrepreneurs with similar educational and professional path are preferred, but also that VCs with a history of working with startups tend to overlook the potential of entrepreneurs with a background in established firms, and vice versa — Franke et al., 2006).

Given then the incredible degree of uncertainty that this business embeds, it might be worth to look for some external help — let’s say a more “automatic” one. However, men and machines are equally bad at predicting success outcomes (Mckenzie and Sansone, 2017), and although I am not even sure that this effort is a feasible one, I am convinced that something valuable can be found in previous (partial) attempts to understand the basics of a successful business.

This article is a first attempt to map the existing findings and players in the space of “automatizing venture capital”. As always, if I missing something or talking nonsense, feel free to reach out.

Image Credit: faithie/Shutterstock

II. Previous studies

The literature on venture capital is quite vast, and it covers a variety of sub-topics ranging from investment choices and exits to organizational issues, relationships, contracting, post-investment, and much more (Da Rin et al., 2013).

My goal here is only to focus on studies that investigate the impact of certain variables on the likelihood of success of an early-stage company, either direct or indirect. In order to do it, I am grouping different studies in clusters that represent the source of a specific competitive advantage that increases the likelihood of an exit: personal and team characteristics, financial considerations, and business features.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Personal and team characteristics

https://unsplash.com/photos/3y1zF4hIPCg

This group concerns all the traits that are strictly related to the entrepreneur or to the founding team. Starting with mere demographics, McKenzie and Sansone (2017) show that people in their 30s — and in particular male that scored highly on ability tests — are more likely to succeed (findings confirmed also in McKenzie D., Paffhausen, 2017). Moreover, being previously unemployed and being married/cohabit are respectively negatively and positively correlated with a higher success rate (Miettinen and Littunen, 2013).

Previously Gompers et al. (2010) showed that being a successful serial entrepreneur increases the success rate in future new ventures, although this is no longer true for entrepreneurs that were previously unsuccessful. In particular, a good serial entrepreneur seems to be better than average at picking the right industry and the time to start a new company (i.e., timing the market). Hsu (2007) does not only support the findings of Gompers et al. (2010) but also provides the evidence that serial entrepreneurs get better valuations.

A lower valuation from a highly reputable VC is on average preferred to obtain a higher valuation from a low-reputation investor.

However, it is not only the experience that matters in this industry. Bengtsson and Hsu (2010) indeed show that ethnic similarity, as well as attendance of the same top universities, increase the likelihood of receiving funding (findings supported also by Sunesson, 2009), while Shane and Stuart (2002) prove that social capital (i.e., having direct or indirect ties with reputable venture capitalists) improves your chances when fundraising. Hsu (2004) even shows that getting a lower valuation from a highly reputable VC is on average preferred to obtain a higher valuation from a low-reputation investor.

Social networks are also very relevant, whether online or offline. In this fashion, Nann et al. (2010) find that the success of a company is intrinsically connected to its founder’s network robustness. Hence, if a founder comes out from a top university and consistently maintain her links with alumni from that same university, she is more likely to become successful. Gloor et al. (2011; 2013) instead analyzed the entrepreneurs’ email traffic and social media activity to understand whether this correlates with the startup success and eventually found that centrality to the network increases the probability of an exit rate. Finally, whether those networks encompass only strictly personal relationships or external business and more formal ones, it does have an impact depending on the stage of the company (respectively, in the first four years external networks have a positive impact on company performance while the same is true in the following four years for internal networks — Littunen and Niitykangas, 2010).

Eesley et al. (2014) focus instead their attention on team composition finding that a diverse team exhibits a higher performance. However, this does not happen all the time but only when in a competitive commercialization environment. On the other hand, technically focused founding teams are more effective when in a cooperative commercialization environment and when pursuing an innovation strategy. However, on the other hand, Mueller and Murmann (2016) investigated the complementarity of skills in the human capital base of a startup (i.e., co-founding team and employees) finding that the mix of business and technical skills has an exponential impact on the company performance only when the founder has technical knowledge and employs additional business experts (not the other way round, and neither when business and technical skills are balanced within a founding team).

Sometimes though, the founder is not the right person to keep running the company after a certain stage. Ewens and Marx (2017) indeed prove that replacing the founder/s with experienced managers can often improve the company performance.

Finally, there are traits that slightly more “esoteric”, in the sense that is quite hard to understand whether a non-spurious correlation exists between a certain factor and the company performance. Entrepreneurship literature has indeed given attention even to the signal originated from calling a company after the owner name. If this results into a more successful company for reputational cost reasons (Belenzon et al., 2017) or into a non-performing firm because of its not ambitious growth-oriented mindset (Guzman and Stern, 2014) is still to be decided.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Financial considerations

Image Credit: F.Schmidt/Shutterstock

This group embeds every research I could find on the impact of funding and financial variables on predicting the success of a company. From a financial point of view, Miettinen and Littunen (2013) studied the impact of equity share (i.e., the capital owned by the entrepreneur as a percentage of total assets) finding that it has some predictive power over the probability that the company will do better than its competitors.

On the funding side instead, Puri and Zarutskie (2012) showed that VC-backed companies are more likely to go public or be acquired and less likely to fail, which is also supported by the findings of Hsu (2006), Nahata (2008), Sorensen (2007) and Inderst and Mueller (2009).

Cumming (2008) shows instead that the probability of being acquired is positively correlated with having obtained financing through convertible securities rather than equity deals, while the opposite is true for IPOs.

Zarutskie (2010) also proved that if the partners of a venture capital fund have prior experience either in VC or startups, they are more likely to outperform their competitors (in terms of portfolio companies exited), and the same is also true for partners with prior industry experience (and, interestingly enough, having an MBAs is somehow negatively correlated with the fund performance). Gompers et al. (2009) also found that VC firms that are specialized in a few industries perform better than generalist VCs, while Ewens and Rhodes-Kropf (2015) showed that there exists a sort of persistence in the performance of a partner of a VC firm. If the partner has brought companies to an IPO in the past, she will keep doing it with future startups. If she instead preferred the acquisition path, more acquisitions will come later for her own portfolio companies — and if she failed multiple times, she will keep failing.

Tian (2011) found instead that syndicate deals are more likely to produce an exit and to do it at a higher valuation. Miloud et al. (2012) also showed that a higher valuation can be reached through a higher product differentiation, industry growth rate, completeness of the management team, and based on whether the founders had previous industry and management experience.

Finally, it is important to consider the investment critical mass, if any — i.e., a certain threshold of funding that increases the likelihood of being successful. Lasch et al. (2007) and Groenewegen and de Langen (2012) showed indeed that raising more than €75,000 results in a greater chance to outperform your peers.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Business features

Image Credit: Philadelphia University Online

This group includes instead other factors that explain the differential performance of a company and that are related to intrinsic characteristics of the business (e.g., technology, intellectual property, etc.).

Different studies (Cockburn and MacGarvie, 2009; Mann and Sager, 2007; Hsu and Ziedonis, 2011) prove empirically that the higher the number of patents a company has, the more likely it is to obtain venture financing and to exit through either an IPO or an acquisition. This is especially true for early-stage companies, where the financial information is either missing or simply forecasted.

Lindsey (2008) showed that strategic alliances are associated with a higher exit rate, and similar results are presented also in Hoenig and Henkel (2015).

Halabi and Lussier (2014) found that having clear financial and accounting information, as well as a certain degree of entrepreneurial attitude and an adequate working capital are positively correlated with a higher likelihood of success, while in a later study Marom and Lussier (2014) proved this likelihood to be positively associated with having professional advice. The study is then run across different countries (Lussier and Halabi, 2010) and extended to 26 independent variables in Teng et al. (2011).

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Bonus Paragraph: Industry Knowledge

All the information presented so far comes directly from academic studies or research-oriented projects. There are tough important signals that the industry has capture over the years which are worth to be mentioned.

In particular, a research made by First Round some time ago shows that having at least one female co-founder increases the probability of success and performance of a company.

David Coats from Correlation Ventures instead recently released an analysis where he shows that having more than two VCs on the board is counter-productive, as well as having none.

Although those are not peer-reviewed results, given the data-driven approach and the brand of the two firms, I found them quite plausible and I am then including them in this list.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

An outsider study: Hobos and Highfliers

Image credit

Even though is clearly a competitive advantage for a VC to know whether a company will be successful or not, this does not guarantee that the entrepreneur will want your money. Timing the market and establishing a strong relationship with a founder earlier on is as much important as knowing whether a company is more likely to succeed than not.

In this light, one study caught my eyes (and very likely it is a unique research of its kind), which was a joint study between Haas Business School and Bloomberg Beta (a Silicon Valley VC). The researchers tried to predict ex-ante who will start an interesting venture and reshaped the idea we have on founders’ stereotypes (Ng and Stuart, 2016).

In fact, potential future founders have at least 8 years of experience after college and start a company mainly during market booms. Furthermore, the higher the education degree, the lower probability to transition to self-employment but the higher the probability to become an entrepreneur (having a Master’s degree, or a PhD in a lighter way, represents the average for US founders). Even though a technology background is often preferred, is not strictly required (an MBA drastically increases the likelihood of starting a company, for instance). Finally, if you hold a position that spans both technical and managerial responsibilities it is more likely that you will want to start a new venture someday soon.


III. Who is playing with f-(AI)-re?

Image Credit: archy13/Shutterstock

Assuming all the material presented above is groundbreaking would be short-sighted and thinking to be the first one to propose a data-driven approach to investing would be, if not only pretentious, also completely false.

In this respect, there are a bunch of funds and individuals out there who are actively looking at the same space. Clearly, knowing exactly what they do is quite impossible without having inside information, so I am simply reporting what I read and heard:

  • Correlation Ventures: probably the first real data-driven investor, it reaches a decision on whether to invest or not in 2 weeks, plus other 2 for extra due diligence. They only do co-investments in the US and don’t take board seats;
  • EQT Ventures: with more than €500M AuM and equipped with an AI system called “the Motherbrain”, they have done more than 20 investments in less than two years;
  • SignalFire: the firm run by Chris Farmer does not only use analytics to pick the right companies but also to help them grow by providing market intelligence and talent matching services;
  • WR Hambrecht Ventures: Thomas Thurston is the key man behind WRH (and Growth Science, its sister tech company) who is advocating for the use of data science to guide growth investments;
  • venture/science: a quant-driven VC led by Matt Oguz, it uses AI and decision theory to compute the risk associated with different attributes such as team completeness, vision, etc. ;
  • 645 Ventures: Series A investor, it follows a metrics-driven approach to Growth Seed investing in a bunch of different sectors;
  • Hone Capital: the Palo Alto-based US arm of CSC Group, it partnered with AngelList to create their proprietary model;
  • InReach Ventures: led by Roberto Bonanzinga, InReach has quickly built a name as the software powered house able to scout early-stage European startups even before others VCs realized they need funding (see what happened with Oberlo, for example). Apparently, it took them two years and £5m in investment to build their proprietary software;
  • Fly Ventures: the recently closed a first €41M fund to do small ticket size investments (up to €1M) and have invested in companies like Bloomsbury AI, recently acquired by Facebook;
  • Social Capital: led by Chamath Palihapitiya, the firm is better known to have started the Capital as a Service (CaaS) concept, and more recently they created an analytics due diligence tool (which is hosted on their webpage) to help them invest in early-stage companies;
  • GV: everyone seems to know GV (formerly Google Ventures) is using AI and machine learning to inform their investment process, although almost no one knows exactly what they are doing and how;
  • Follow[the]seed: a post-seed global algorithmic VC, they have developed two data-driven methodologies (one B2B and one B2C) to simply the investment process;
  • Right Side Capital Management: with more than 800 pre-seed investments done so far, they make small investments ($100k-$500k at valuations of less than $3M).

I have also heard Sunstone , e.ventures, and Nauta Capital use machine learning and analytics to some extent, but I could not find much info so if you know anything about them I would love to learn more.

It is also worth to mention there are a few platforms (not VCs) that are looking at either helping investors or simply democratizing the investors’ skills as much as possible. The first one is Aingel.ai, which has recently filed a patent for a machine learning system that scores startups and founders on the basis of a set of different variables. PreSeries is another fully automated solution to discover and evaluate startups, which it also has a voice interface (through Alexa).


IV. Conclusions

The idea of de-risk and unbias the selection of early-stage companies is certainly attractive, and I believe it deserves further studies and attention since it has the potential to increase the quality of companies that get funded (and founded as well).

Of course, having a better due diligence and decision process can favor investors but it does not solve all their problems. Whether the entrepreneur gets an investor’s money does not depend, in fact, by the ability of the VC to do her due diligence but is rather driven by establishing a personal relationship and providing some additional value to the mere monetary contribution.

Moreover, it is hard to understand ex-ante what effects this class of models might have on the creation of new startups (maybe only a few clusters or group will be founded in the future to reflect the “success factors” identified by the AI models, in a sort of adverse selection phenomenon) and it is also quite cumbersome to disentangle and consider the effect of the VC additional value in computing the likelihood of success of a company.

Is being data-driven a value for investors? Certainly. Whether anyway a fully automatic VC can be created is still to be seen.

References

The reference list includes about 50 articles, and to avoid for all those words to raise excessively the reading time of this article, I have posted it on a different blog post, that you can find here.

This post is an excerpt from my forthcoming book “An Introduction to Data“ edited by Springer.