AI’s strength lies in its predictive prowess. Fed enough data, the conventional thinking goes, a machine learning algorithm can predict just about anything — for example, which word will appear next in a sentence. Given that potential, it’s not surprising that enterprising investment firms have looked to leverage AI to inform their decision-making.
There’s certainly plenty of data that one might use to train an AI-powered due diligence or investment recommendation tool, including sources like LinkedIn, PitchBook, Crunchbase, Owler and other third-party data marketplaces. With it, AI-driven financial research platforms claim to be able to predict the ability of a startup to attract investments, and there might be some truth to this. One study of hedge fund performance found that AI-driven funds generated higher average monthly returns over a 15-year period than their human-guided counterparts.
VC firm Signalfire uses software called Beacon to project the performance of investment targets, drawing on data sources including academic publications, pitch decks, open source contributions, regulatory filings, sales and even raw credit card data. It isn’t alone in its efforts. Hone Capital created a model based on a database of over 30,000 deals, considering aspects like whether a team made it to a Series A round and investors’ historical conversion rates. Meanwhile, biotech-focused Deep Knowledge Ventures claims to have “appointed” an AI system to its board of investors, giving it a vote informed by prospective companies’ financing, intellectual property, previous funding rounds and clinical trial data.
Gartner predicts that by 2025, 75% of VCs will use AI to figure out when to invest, where to invest and how much to invest, and determine which leadership teams are most likely to succeed based on factors like their employment history and previous business success. But we’re a far cry from that number today.
In an informal survey to determine how widespread the use of AI at VCs might be, TechCrunch reached out to major firms including New Enterprise Associates, Sequoia, Kleiner Perkins, Accel, Index Ventures, Lightspeed Venture Partners, Greylock, Matrix Partners, and Plug and Play Ventures. The vast majority didn’t respond or declined to comment, suggesting that AI-powered investing tools are a closely guarded secret or simply aren’t the norm today.
Considering the VCs using AI and data analytics tend to shout it from the rooftops (see: Correlation Ventures, EQT Ventures, Jolt Capital, InReach Ventures, etc.) it’s a relatively safe assumption that AI isn’t a massive component of prospecting today. Vendors like Raized.ai may want to change that — Raized sells software that selects startups for VCs based on investment goals and past decisions — but the anecdotal evidence suggests otherwise.
AI’s applications in VC range from discovering companies seeking funding to identifying early growth signs and tracking the growth of existing portfolio companies. So why the slowness to embrace it? It could be that AI hasn’t proven to perform better than the tools — and expertise — VCs already have in place.
AI can also have pretty significant downsides. In an experiment last November, Harvard Business Review built an investment algorithm and compared its performance with the returns of 255 angel investors. The model outperformed novice investors with the same data. But it underperformed compared to experienced investors, and — worse — exhibited biases that the investors did not. For example, the algorithm tended to pick white entrepreneurs rather than entrepreneurs of color and preferred investing in startups with male founders, perhaps because white men were overrepresented in its underlying dataset.
Imbalanced or inadequate data often proves to be the Achilles’ heel of AI systems, especially investment-analyzing ones. Systems are prone to overfitting (i.e., becoming too attuned to the data on which they were trained) when dealing with rare events, times in which extrapolating from the past to the future isn’t necessarily appropriate. Consider the current economic downturn: There haven’t been many recorded U.S. recessions in history — much less with an actively tightening central bank, a war in Ukraine and the tail end of a pandemic that’s battered global supply chains.
The pandemic alone may have sent some algorithms for a loop. Computer-driven funds Voleon, Renaissance Technologies and Winton Group suffered large losses in 2020 as COVID-19 rattled the markets.
That’s not to suggest AI can’t be useful to VCs in other ways. For example, using an algorithm to analyze satellite imagery of a retailer’s parking lots can give an idea of crowdedness and, by extension, merchandise flow. Elsewhere, preliminary research suggests AI might be able to predict corporate decisions from linguistic signals.
Feng Li, an associate professor at the University of Michigan, found that signs of self-serving attribution bias (SAB) — when people take credit for successes and blame others for failures — in the management discussion and analysis sections of annual reports tend to correlate with cash flow volatility. Firms with managers who had higher SAB performed worse than those whose managers presented as less egocentric.
A human-hybrid strategy is perhaps the most sensible — and the approach even the most AI-forward of firms appear to be taking. AI as it exists today has the potential to save labor, a godsend for investors who spend upward of 118 hours on screening and due diligence per startup.
But it’s far from a silver bullet. Far into the future are the days AI begins to make investment decisions autonomously — assuming that those days ever come.
Comment