You’ve been studying the adoption of AI by venture capitalists. Can you give us some background on your decision to investigate this topic?
In my research paper selected for the job market, I study how the adoption of artificial intelligence by venture capitalists (VCs) to screen startups affects the funding of early-stage innovative companies. The motivation relates to the fact that the past two decades have witnessed rapid growth in data availability and processing thanks to statistical techniques such as machine learning and AI. The adoption of these technologies by financial intermediaries has however raised concerns regarding their effects on investment decisions and, more broadly, on the allocation of capital. In my research paper, I focus on a key class of financial intermediaries— VCs, which are private equity investors in startups with high growth potential and play a crucial role in the financing of innovation.
Can you explain how venture capitalists use AI tools?
In recent years, dozens of VCs have adopted AI technologies for screening startups, i.e., sourcing, evaluating and selecting startups to fund. These VCs have developed their own proprietary platform which automatically tracks and scores startups in terms of future return prospects. Put simply these VCs employ algorithms to detect quantitative patterns in historical data from previous startups and extrapolate them to predict a new startup’s outcome.
How do you identify the venture capitalists who use AI?
I develop a classification of VCs to determine whether and when they adopt AI. Using job and employee data, I identify VC firms that hire data scientists who develop machine learning algorithms for investment screening, and I call these employees AI-related employees. Using job starting dates, I classify a VC as becoming AI-empowered from the date it hires one AI-related employee.
So, how is the use of AI by VCs changing their funding strategy?
In a nutshell, I show that VCs that adopt AI become better at identifying good quality startups, i.e., those that survive and receive follow-on funding, but only within the pool of startups whose business is similar to that developed by past companies. At the same time, VCs that adopt AI become less likely to fund breakthrough companies, i.e., startups that achieve an IPO or obtain highly cited patents. This finding is associated with an increase in the share of their investments being oriented toward startups developing businesses closer to those already tested. These results are consistent with AI exploiting past data informative about companies similar to past ones but not informative about breakthrough companies. Overall, my paper shows that AI adoption by investors might hinder the allocation of capital to breakthrough innovations.
VCs that adopt AI become better at identifying good quality startups, but only within the pool of startups whose business is similar to that developed by past companies.
How do you define and identify innovative startups vs. startups similar to past ones?
I construct a measure of “backward-similarity”. Specifically, I measure the similarity of the text we find in a startup’s business description and compare it to those of previous VC-funded startups in the same industry. So high backward-similarity startups run businesses similar to those that have been already tested by past startups. In contrast, low backward-similarity startups are more likely to be innovative and to develop novel products.
A few words on the method now: how can you be sure that your results are causation and not correlation?
To provide causal evidence that AI adoption leads to changes in VCs’ investments, I use a plausibly exogenous shock to one often cited determinant of a VC’s decision to adopt AI: the number of potential investment opportunities it faces. Indeed, given the large, fixed costs of evaluating investments and the limited scalability of VC firms, more investment opportunities make screening more onerous, creating incentives for VCs to adopt AI technology to automate screening with a view to saving time and costs. Specifically, my empirical strategy uses a quasi-natural experiment: the introduction of Amazon Web Services (AWS), i.e., cloud computing services by Amazon. This shock lowered the cost of starting new software- and web-related businesses, leading to more startup creations in specific industries and thus an increase in investment opportunities faced by VCs.
What do you think are the big implications of AI adoption for the VC industry and the funding of innovations?
Taken together, my results show that AI adoption by VCs affects how they select their investments and, more broadly, how capital is allocated among young innovative companies. This suggests that AI adoption by investors can shape the nature of innovation and thus can have a significant impact on future growth trajectory.