top of page
Thomas Thurston

How is AI changing the job of Venture Capitalists (whether they like it or not)?

Updated: Aug 25, 2023


Recently I listened to a panel of venture capitalists discussing AI. Some of their comments were insightful, but an eye-opening realization dawned on me. The VCs were happy to pontificate about AI, but it became increasingly evident that none of them were actual users of AI themselves. They might invest in startups that use AI, but they were a step removed. In other words, we’re still in a historical period when the norm in venture capital is to manually meet with startups and invest based on personal acumen and gut instincts. No AI required.


This probably jumped out at me because, unlike the panelists, my career has been about using AI in venture capital. We’ve been building and deploying AI to analyze startups and guide investments for nearly two decades. Obviously this level of commitment to AI in venture capital is abnormal, but it was still a surprise for me to hear other VCs talk at length about AI without any faint notion that they might be, or ought to be, actual users themselves. It felt a little like watching vegetarians judge a barbequed ribs contest.


This begs the question, how is AI changing the job of venture capitalists?


Here are four ways we're seeing AI change the job of venture capitalists:


1. Geography becomes almost irrelevant.


Not long ago, to attract venture capital it was quasi-mandatory for a startup to be based in (or move to) the Bay area, the Boston area or New York. Today geography matters less, but still plays a major role.


One reason geography still matters is due to venture capital’s labor-based business model. The whole process is artisanal – relying on personal judgment and relationships. VCs like to invest in startups that aren’t located too far from where they live, or where they want to regularly travel, because everything hinges on a personal touch.


While this offers some practical benefits, it’s also severely limiting. That’s because, today, most startups arise and compete globally. Spotify, SoundCloud and Apple Music aren’t fighting to win their respective streaming music markets in Stockholm, Berlin and Cupertino, they’re fighting for every market, everywhere.


It makes less sense every day for a VC to limit its scope to… say… Boston, just because they happen live there. If the next unicorn shows up 10,000 miles away and ends up winning 90% of the market, there’s little consolation for investing in the 5th runner-up just because it’s near your house.


This underscores a huge advantage we’re seeing from AI. One of the things we use AI for is to scour the planet for every startup it can find in whatever industry we care about, whenever we want, regardless of geography. This lets us square our scouting efforts with the global realities of competition to, as best we can, identify the likely market leaders wherever they happen to be. It’s the digital equivalent of having an office in every city and village.


2. Personal deal flow networks are becoming less relevant.


To a VC, personal reputation and relationships aren’t just a matter of ego, they’re critical business assets. In a manual, non-AI world, a fund’s deal flow is only as good as the startups it can get in a room with. Everyone wants an “inside track” on the “best deals” that are presumably being whispered about in the “right circles.” (Sorry for the air quotes.)


AI makes these concerns largely irrelevant. Instead of drumming up quality deal flow through reputation and relationships, AI lets us analyze markets at scale to identify and cherry-pick the highest performing startups in the markets we care about (whether they’re in our personal networks or not). Once identified using AI, we reach out to the startups and start a conversation.


In other words, we don’t need startups to come in the door and pitch us. AI has made this form of inbound deal flow irrelevant. Instead, our deal flow is mostly outbound, in that we use AI to find startups ourselves. This spares us conversations with startups we’d never invest in while connecting us directly to startups with the traits we’re looking for. It’s a huge time saver for both parties.


3. Personal gut feelings are less relevant.


Trusting the “gut” has led to high historical failure rates in venture capital for as long as anyone can remember. For example, as Harvard professor Shikhar Gosh pointed out, around 75% of venture-backed startups typically lose money.[i]



With that in mind, an obvious benefit of AI is where it can potentially lead to better identifying which startups to invest in. Of course, there’s a wide range of AI quality. Some AI is amazing, other AI is trash. That said, when it works, AI can create a host of advantages for venture capitalists willing to use it.


For example, AI can be more objective, consistent and accurate than traditional personality-based processes. There’s also the potential for AI to reduce bad biases or discrimination. Sure, AI can reflect the biases of its creators, but bad traits in AI can also be identified and improved in a way that people’s brains can’t. Flaws in a version of AI are often more tactically improve-able than the secret thoughts, abstract feelings and biases of a human brain on any given day.


In practice, we’ve found AI to be a great counterweight to the cognitive biases and human shortcomings that plague most venture capital decisions. I’m as guilty as the next person of falling in love with a startup, its team, its dream. I know this about myself. So having a statistically sound, vetted, empirical set of guardrails to improve my decisions is both a personal relief and a competitive advantage.


We use AI as a series of gates that a startup must pass through. If these requirements are met, we then evaluate the startup in more traditional human-based ways. In other words, if the AI says “no,” it’s a hard no. Yet if the AI says “yes,” it’s still up to the humans to decide. At least this way, even if our personal gut feelings run away with us, we’re still only choosing from a menu of startups with the objective qualities we’re looking for.


4. There’s still fear, anxiety and self-serving disbelief.


Most laypeople are surprised when I tell them VCs don’t typically use AI, or anything vaguely resembling statistical insights, to guide their investment decisions. There’s an assumption that modern VCs do “that kind of thing.”



This reaction is fairly new. Only a few years ago people still mostly recoiled at the thought of bringing AI, or even basic statistics, into venture capital decisions. It was often seen as a battle between art and science, which I’ve always felt was a sham dichotomy and a false choice.


It’s been interesting to see a shift in what non-VCs assume, but the most fear, anxiety and disbelief has come from VCs themselves.


Yes, some VCs like me went headfirst into AI, but that’s a small minority. Other funds have dabbled in some quant, which usually means getting a Pitchbook subscription and a Summer intern to poke around for correlations. Meanwhile, the vast majority VCs continue to do nothing. Zero.


It isn’t that they can’t do it, it’s that they don’t want to.


A lot of VCs have the same reaction to using AI that yellow taxi drivers had to Uber. It’s the reaction Blockbuster first had to Netflix. It’s the reaction the old baseball scouts had in “Moneyball”[ii], and the reaction traditional record labels are having to digital platforms that let musicians self-publish. Harumph.


Most VCs don’t have skills or expertise in AI. They also don’t perceive a problem in the first place, since they’re generally self-assured about their acumen and gut feelings (even if their performance has been poor).


At a deeper level, VCs who earnestly attempt to use AI can be surprised by how difficult it is to create reliable quant models, and how disruptive it can be to their organizations. The implications can be further-reaching and more existential than most funds expect.


For example, last year it was announced that Google Ventures (GV) stopped using an algorithm known as “The Machine” that, for around a decade, helped determine which startups GV invested in. [iii].


Why? Was Google’s algorithm ineffective? Did it stop working?


Apparently, the algorithm worked just fine. It isn’t that The Machine didn’t work, it’s that the humans overthrew it. There was a coup d'état. A mutiny. According to Axios:


“GV investors sometimes tried to game the algorithm by manipulating the inputs. In general, however, the firm abided by the machine’s red lights (plus greens and yellows)… it became GV’s de facto investment committee...

…There doesn’t appear to be a single incident that killed off the algo.

Instead, it appears to have been a gradual process born of growing self-confidence (GV now has nearly 40 investors managing around $8 billion in AUM) and growing frustration (particularly when the algo would rule against follow-on investments for existing portfolio companies, as the deal market deteriorated).

The bottom line: GV still relies heavily on data. After all, this is the corporate venture arm of Google. But data has been relegated to its original role as an aide, rather than arbiter.”[iv]

That pretty much sums it up. VCs say “no” to startups all day long, but they don’t like hearing it themselves. They want to get their deals done and hate the idea of being second-guessed by a machine, even if it works. Imagine trying to impose rules on 40 venture capitalists running around with Google business cards and their own feelings about what startups they like and you'll get a sense of what The Machine was up against.


These are just a few of the reasons it’s hard for VCs to use AI. Sometimes it’s about lacking skill, but mostly it's about lacking will.


What's ahead for AI in the job of Venture Capital?


Despite some industry resistance and a relatively slow start, venture capital has indeed been moving in a more quantitative direction - albeit gradually. I think there's a hint of inevitability in this continuing forward, but venture capital's quant journey won't have the same timeline or flavor that the stock market had (for example) in its quantitative journey.


At a minimum, we're already seeing first-hand how AI can push back against traditional boundaries imposed by geography, personal networks and subjective decision making.


After all, the goal is to better find great startups, to better help entrepreneurs succeed, to make investors more money, to solve more tough global problems and to deliver net-gains to the wellbeing of society as a whole.


At least... that's how AI should change the job of venture capitalists (whether they like it or not).



















[i] Shikhar ghosh, The Venture Capital Secret: 3 out of 4 Start-Ups Fail, Wall Street Journal, September 19, 2012. See: <https://www.hbs.edu/news/Pages/item.aspx?num=487> [ii] Lewis, M. (2003). Moneyball: The art of winning an unfair game. W. W. Norton & Company. See also the Movie based on the book (starring Brad Pitt and Jonah Hill): Miller, B. (Director). (2011). Moneyball [https://www.amazon.com/Moneyball-Brad-Pitt/dp/B006IMY5ZU]. Columbia Pictures. [iii] Dan Primack, Google Ventures shelves its algorithm, Axios, Economy & Business (September 28, 2022) < https://www.axios.com/2022/09/28/google-ventures-shelves-its-algorithm> [iv] Dan Primack, Google Ventures shelves its algorithm, Axios, Economy & Business (September 28, 2022) < https://www.axios.com/2022/09/28/google-ventures-shelves-its-algorithm>

bottom of page