AI Can Help Venture Capitalists Fund More Diverse Startups
4 min read · Jan 2025
Lyonnet and Stern (2022)
“...an algorithm with no embedded gender or other ‘in-group’ preferences, but simply tasked with predicting venture success, would almost double the proportion of VC-backed female founders in the study’s test set [to 18%].”
Summary
This quantitative study examines venture capitalists' decision-making processes and compares them to predictions of a machine learning algorithm. The authors find that in selecting investments, venture capitalists prioritize founder characteristics, such as age, gender, and education, over firm characteristics. As a result, this bias leads to the underfunding of potentially valuable ventures. The authors also show that a machine learning algorithm outperforms venture capitalists in predicting startup success and improves the diversity in the businesses funded, particularly regarding gender.
Method
The authors collect data on all firms established in France between 1998 and 2010, totalling 123,511 new firms, including information on founder demographics. They use machine learning techniques to identify the firm characteristics that correlate with strong performance, measured by total sales and value-added figures from tax files. These firm characteristics are used to build an algorithm to select firms for investment. The algorithm's selections are then compared to those made by venture capitalists.
Key Findings
VC-backed firms outperform the average firm, but VCs make costly mistakes.
VCs invest in firms that turn out to be poor performers.
Eliminating the bottom half of their portfolio based on performance would increase returns by 48%.
Algorithm-selected firms perform better than VC-backed firms.
The algorithm achieves this by avoiding firms that fail within 5 years and identifying a higher number of “super performers” when compared to VC-backed firms.
Algorithms designed to predict firm success would improve the diversity of VC-backed firms.
VC investment decisions are heavily influenced by founder demographics such as age, gender, and education. In contrast, the algorithm places less weight on these factors, leading to a more diverse representation of entrepreneurs.
The algorithm could almost double the proportion of female-founded VC-backed firms to approximately 18%.
This improved diversity is estimated to enhance overall VC performance by about 6%.
Takeaways
There is growing evidence that venture capitalists underfund women-led ventures. This study explores one potential solution: the use of machine learning-trained algorithms. The results are promising, suggesting that such tools could nearly double the proportion of VC-backed firms founded by women.
However, more data and testing across diverse contexts are needed to confirm that an appropriately tuned algorithm can improve representation without introducing other biases. While AI has some potential to mitigate some biases, it has also been shown in other scenarios to amplify them.
References
Lyonnet, Victor, and Léa H. Stern. 2022. “Venture Capital (Mis)Allocation in the Age of AI.” https://papers.ssrn.com/abstract=4035930 (September 19, 2023).
About WIN-VC Canada:
New Power Labs is the research lead of the Women and Nonbinary (W) Impact (I) Network (N) for Venture Capital (VC), a national collaborative of organizations working to provide services, programming, events, and dedicated resources to women and non-binary entrepreneurs and gender lens investors across Canada who are working towards becoming investment ready and increasing the pool of investors driven to invest in these ventures.
This research is part of WIN-VC Canada, supported by the Government of Canada. WIN-VC acknowledges the support of Innovation, Science and Economic Development (ISED). ISED has awarded funding for WIN-VC that will make the venture capital environment more inclusive for women by transforming traditional investment processes, processes and knowledge into respectful and meaningful approaches that value equity and impact with a focus on diverse women and non-binary entrepreneurs and SMEs including Black communities, Indigenous peoples, racialized populations, persons with a disability, 2SLGBTQ2+ and new Canadians.
