VWV Volumes #15: VP of Operations Ben Piekarz '24 on AI for VC - A Relevant Complement or a Threatening Replacement?

In the traditional VC fund, associates source deals and conduct due diligence processes in order to make investment decisions. This model has always remained the status quo, but with the rise of technological power, specifically AI and machine learning, is it still the most efficient system for a VC firm? Hypothetically, an AI investment decision model could eliminate the inevitable biases that cloud the judgments of human investors and lead to superior and fairer results. 

Image Source: Techopedia 

A Study from St. Gallen 

To determine whether AI-powered investment decisions lead​​ to greater returns than traditionally-made decisions, researchers at the University of St. Gallen “built an investment algorithm and compared its results with the returns of 255 angel investors.” The model was trained using 623 deals from a European angel investing network, and the data it was given mimicked that available to the angel investors at the time of sourcing. The model was then tested using other investments previously made by members of the angel network, and their past results were compared to those simulated by the algorithm. The algorithm achieved an average internal return rate (IRR) of 7.26%, beating out the average for the 255 angel investors, at 2.56%, by a significant margin. The authors of the study cited five different biases that cloud human investors’ judgements when dissecting their results: local bias, loss aversion, overconfidence, gender bias, and racial bias. 

While the algorithm achieved a greater rate of return overall, it is important to note that there are still shortcomings to an AI-only approach to deal evaluation. Indeed, the cohort of the most “elite”, experienced investors, as classified by the study, registered an average IRR of 22.75%, demonstrating that there are factors in human-led due diligence that likely cannot (at least at our current technological frontier) be replicated by AI. St. Gallen's research team specifically noted that “elite” investors who outperformed the algorithm were the investors who “showed far fewer signs of cognitive biases.” Personally, I believe a primary reason that the most experienced investors can still outperform AI is due to their ability to evaluate the character of the team. VCs typically cite the team as one of the most important factors in evaluating the potential of a startup. How can AI’s effectively determine whether a team member would make a strong leader or an incredible marketer? Sure, AI can look at past experience of team members and use that as evaluation criteria, but what about young, naturally gifted founders who just haven’t had the time to amass the experience? I believe that these questions about the team -- many of which AI can’t comprehensively answer -- are one of the primary reasons that human investors are necessary for optimal investment decisions. For now at least. 

Image Source: iStock 

How Firms Use AI Today 

With the benefits of AI deal sourcing becoming increasingly evident as AI technology develops, VC firms around the world have begun implementing AI to aid in their investment cycles. 

Exhibit A: Veronica Wu, managing partner at Hone Capital, a Silicon Valley-based investment fund, built a machine learning model to aid in their fund’s investment decisions. This model was created from a database of over 30,000 deals from a variety of sources, and today, the model comes up with a recommendation for every deal sourced by the fund. Wu attests that the machine learning model often causes their fund to reconsider their intuitive positions on any particular deal, signaling the trust her team puts into the algorithm. She notes, however, that deal evaluation remains a firmly hybrid process for her fund. 

Meanwhile, Signalfire is a San Francisco-based VC firm, founded by a former Google software engineer, spends $10 million a year to develop and maintain their AI platform. They use this platform for all four of their investment phases: deal sourcing, due diligence, investing, and adding value to the company post-investment. EQT Ventures, a Stockholm-based VC fund, also built their own AI platform, which partner Alastair Micthcell estimates “...has played a role in more than $100 million of the firm’s approximately $900 million total invested since its first fund opened in 2016”. Clearly, early VC adopters of AI​​ technologies use their models as complements and productivity enhancers, but not as total replacements to their human investors. 

Food for Thought

VCs are always eager to disrupt industries with new startups, yet they have been remarkably slow to adopt AI and disrupt their own space. Still, it seems the time has finally come for the VC industry to adopt AI models, with the data supporting AI/ML use and more and more VC firms spending money on their own models. It remains to be seen who can build the best model to enhance the human investor. Here are some questions I’m watching closely as we enter this exciting new realm of investing: 

• What makes a perfect model? 

• Will we see the models put out by different VC funds converge to near perfection after years of new data and information flow?

• Will AI ultimately close the gap between the highest-returning VC firms and the rest of the pack?  

• Will firms end up trusting their AI models more than they do the human intuition of their employees (shifting away from a hybrid AI-human model to primarily AI)?

• How much will firms be willing to spend on developing these models? 

Let the hybrid human-software investment model take over, and let’s see how the VC world evolves.