Automation and investment: Channeling the flow

It’s hard to argue against the idea that artificial intelligence (AI) will play a large role in financial decision making in years to come. For private equity fund managers, however, it has proven difficult to plug the trading advantages of superhuman pattern recognition into an investment judgment business. 

Even in the data-driven hedge fund space, where highly systematized techniques have been common since the 1980s, the widespread use of AI has yet to translate into confidence about consistency. An index of 18 AI hedge funds tracked by industry researcher Eurekahedge punctuated this sentiment at the end of June by posting its worst half-year return to date at negative 2.5%. 

Still, many PE investors are convinced that the efficiencies realized by AI in scaling operations around quantitative analysis will streamline deal screening, due diligence, and ultimately final asset selection and investment processes. The question is to what extent data-driven automation can be achieved in an industry where information is traditionally scarce, and many investments are based on relationships and circumstantial evidence. 

“In the hedge fund world, computers have essentially replaced human beings in terms of managing trades, they are obviously much more efficient at finding arbitrage opportunities,” says Veronica Wu, a managing partner at Hone Capital. “For us, gathering the data is more challenging and we don’t make decisions purely based on that. But technology does allow us to do some interesting modeling and get more data points than traditional GPs.”

Bulk processing

Hone scrapes up its data from a range of publicly available sources including PitchBook, LinkedIn, and Crunchbase, as well as a close relationship with AngelList, which provides 90% of deal flow. The firm’s AI system focuses on the probability of companies receiving follow-on rounds within 24 months and has achieved an 84% accuracy rate in its predictions. 

That score is not considered sufficient to eliminate the human element of the investment process, but it has had a noticeable impact on economics. Wu estimates that at least 50% of Hone’s efficiency gains are driven by leveraging technology and network data. 

The GP has built a portfolio of 300 seed companies, more than half of which have had valuation mark-ups or completed follow-on rounds. By comparison, less than 20% of fund managers are said to be able to achieve a 40% follow-on rate, even with portfolios of only 10-20 companies.  

Generalist early-stage investors are most in need of this kind of bulk processing strategy due to the larger pools of potential deals they have to analyze. But at the same time, they face the biggest challenges in collecting the required information since their target companies are still in the concept stage, typically with no assessable metrics whatsoever. 

The dissonance has resulted in deal screening remaining the safest investment application of big data in a private equity context and VC taking hold as a perhaps unlikely starting point for AI proliferation across the asset class. 

Other early-movers include US-based Correlation Ventures, which claims to be able to make data-based investment decisions within two weeks, and Sweden’s EQT Partners, which says that its Motherbrain machine contributes to as much as 30% of VC deal flow. Meanwhile, Deep Knowledge Ventures, a Hong Kong GP with a clear flair for publicity, has appointed an algorithm named Vital to its board of directors.

Jay Eum, co-founder of TransLink Capital, has personally invested in robo-advisors for public markets but believes the relative incompleteness of data in the private sector means that meeting founders in person will always be a necessity. Furthermore, the idea that businesses of all sizes are ultimately guided by human nature as opposed to computer protocols suggests that a one-on-one approach to deal assessment may remain indispensable across the spectrum of company development stages. 

“There’s no question that AI is going to help with efficiency, but it’s never going to be 100% machine driven because there’s just no alternative for getting a feel for whether or not you can work with a person or trust them,” says Eum. “There are many background cases of capable founder-CEOs who have derailed a company and the investors lost everything. That’s why that final step of direct interaction is irreplaceable.”

Innovations to come

The natural counterargument here is that human intuition is essentially pattern recognition ability developed over the course of a single lifetime – a skillset easily outmatched by AI. In this light, some investors suggest that the machine-assisted deal sourcing now being used in venture capital will make its way to private equity in about five years. 

However, AI is unlikely to be allowed to lead deal processes all the way through the execution phase anytime soon. This has more to do with fiduciary responsibility than an emotional reaction to human obsolescence. The broader ecosystem of regulation and commitments across PE and VC would need to be overhauled, probably during a decades-long process, to accommodate the fully machine-led investment of coffers as socially sensitive as pension funds.   

As a result, the near-term expansion of AI in deal activity could be mostly limited to shortlisting proposals through specific criteria such as the probability of follow-on funding or IPO success. But a number of firms, including Correlation, are pushing this approach into more nuanced diligence work by setting screening filters based on factors such as the track records of individual lead investors in various deal opportunities.  

Further developments in this vein will likely extend to AI-powered reference checking of entrepreneurs and founders, especially given industry observations that the most valuable input is sourced through indirect contacts. The grunt work in this area is time-consuming but could be automated to the point of sending out reference request emails via a generic template to machine-identified colleagues of the candidate investee.

“Talking to investors can be a waste of time for everyone,” says Andrey Shirben, a serial entrepreneur and a founding partner at VC firm FollowTheSeed. “They can drag you to meetings with no intention of investing, talk to you for six months, and then tell you at the end that they’re not investing when they pretty much could have told you no on the first or second meeting.” 

FollowTheSeed aims to disrupt the who-you-know networking approach to deal sourcing, which Shirben says is too subjective and results in too many missed opportunities. The firm, which operates across bases in Australia, China, the US, and Israel, uses two algorithms to automatically collect data over a three-week period on each company that fills out an online registration. Data is collected by a 200-question survey before a profile of the company is created and ranked.   

Part of the process is to screen out the most serious businesses by using AI to search survey responses for indications of customer addiction such as a tendency for app users to log on between 6-8 a.m. Shirben describes this process as reversing the order in which start-ups pitch their two main talking points: the story behind the business and the operational particulars. 

“The data driven approach goes straight to the questions about things like how many users they have and how much cash is in the bank,” he adds. “There’s no ‘maybe’ or ‘it’s a good story’ because the algorithm doesn’t care about the story. It only cares about user engagement, and what it looks for is products that create compulsive, habit-forming behavior.”

The expectation game

Competition from emerging alternative capital-raising models, including crowdfunding and virtual token sales, is expected to accelerate the uptake of such strategies, especially as a new crop of specialist suppliers drives diversification in data service availability. Third-party players in this space include BankerBay, which claims to operate the world’s first deal origination platform, and Tracxn, which counts some 500 corporates and VCs as clients.  

“Sooner or later, the whole investment process is going to be automated,” says Daleri Nasibi, CEO at Exit Factory, a Malaysian firm that offers services across deal sourcing, database access, document drafting, and decentralized investment via a blockchain-based platform. “But for another 5-10 years, until smart contracts are fully integrated into every step, there’s going to be a human factor in picking projects.”

Much of the expectation around the rapid expansion of tech-enabled deal-making is based on the idea that participating investors will reap disproportionate rewards. This will in turn cause industry players at various ends of the value chain to realign their interests with the winners, leaving the stragglers to catch up either by building systems in-house or collaborating with platform providers. 

The difficulty with this outlook is that for early movers, the secret to success will be less about the tools of the trade than finding the right, economically viable combinations in which to use them. The value created by automated investment practices is based on the articulation of the outcome being sought in the strategy and the kinds of raw data that are being fed into the algorithm. The technology, contrary to the hype, is hardly bleeding edge. 

“There are not that many investors doing it right now, so it’s still a point of differentiation in the industry, but it will eventually become a necessity,” says Hone’s Wu. “I personally think that’s going to take a long time, though, because it’s very much limited and determined by how much data you can collect – not the AI. The machine learning models themselves are nothing new in academia. Many of them were developed 20 years ago.”   

• Tweet  
• Facebook  
• LinkedIn  
• Google plus  
• Save this article  
• Send to  

• Topics
• GPs
• Technology
• Performance
• Investments
• Asia
• Translink Capital