Correlation Ventures: Taking the “Gut Feel” out of VC
Correlation Ventures leverages a comprehensive database on VC financings and an associated predictive model to make rapid VC investment decisions based on data.
Correlation Ventures is using data to take the instinct out of venture capital investing. Typical venture capitalists depend on “gut feel” built through years of experience to make investment decisions, but this can lead to unintentional adverse effects. Laura Huang, a professor at Wharton who researches how intuition can affect VC investments, believes that “gut feelings might just be a cover for our bias…driving most of our decisions.”
Founded in 2006, Correlation didn’t make a single investment for 4 years. Instead, the founders (HBS alums) David Coats and Trevor Kienzle, amassed the most complete database of VC financings in existence, by meeting with other VCs, entrepreneurs, and data providers. Using this data (consisting of financing details, investors, board members, management, industry segment, business stage, exit details, and more), Correlation built a predictive model that enables them to make many investments very rapidly using disciplined analytics.
While typical VCs may take months to make an investment decision, Correlation leverages their model to make decisions in as little as two days. For entrepreneurs and VCs seeking co-investors, the firm offers quick and easy access to VC capital without ever taking a board seat. The decision-making speed enables Correlation to make many more investments than a typical VC; in September 2015, they had a portfolio of 112 companies for their first fund. The model is built to achieve smaller successes. Co-founder Coats described the model as “kind of like counting card in blackjack” in that “it’s really designed to tilt the odds in our favor a little bit. If you play enough hands you should win.” The use of data to drive investment decisions seems to be paying off; Correlation has significant investments in Casper and invested in Virsto, which exited for $200M, resulting in a substantial windfall for Correlation and evidence of their ability to capture value they create.
Correlation is not alone in using analytics to identify good investment opportunities. Several other firms have employed similar approaches and amassed databases of information they use to create models. Correlation believes their database to be the “most complete, comprehensive database on venture investments and their outcomes.” Assuming no other companies can amass a comparably complete database, Correlation should be able to maintain a competitive advantage relative to competitors in making quick and profitable VC decisions. The data ownership also presents several opportunities for the firm. For example, the database could potentially be sold as a product, enabling Correlation to become the data provider to other VCs, a potentially lucrative opportunity.
If Correlation were to productize and sell their data, other VC firms would face an interesting organizational challenge like the one seen in Moneyball. Venture capitalists have historically relied on instinct and gut feel to make investment decisions. It could be challenging to convince these individuals to trust a computer model, particularly if they fear being replaced by the machine. It will be crucial to get senior leaders to believe that data analytics are a benefit to their role rather than a threat, and to recognize that their intuition is likely introducing bias that could be impacting results. Leveraging a predictive model in VC could help firms eliminate much of the bias that exists in today’s investment environment (for example, investing in male entrepreneurs over female entrepreneurs) while also increasing returns for firms.
Time will tell if predictive analytics lead to investment success for Correlation and other data-driven VC funds. With $350 AUM, Correlation is worth keeping an eye on in the coming years to evaluate the role data can play in VC.
Sources:
https://www.alumni.hbs.edu/stories/Pages/story-bulletin.aspx?num=4835
I find the analogy you used to compared this method to counting cards in blackjack very interesting, and a reason why I am skeptical of this model. Given that odds are marginally shifted in your favor seems not good enough in an industry where each bet should return at least your whole fund size.
Interesting post, Jessica! Although it is possible that Correlation Ventures is able to identify strong investments using this model, I wonder what the value prop is for entrepreneurs who might be looking for investors who can facilitate networking and provide operational know-how. I’m also very curious to understand how robust this model would be for seed/early stage investments where data is incredibly limited, as well as for new technologies for which historical and comparable competitive information does not exist.
I think this was such a great idea, and in a competitive field with hundreds of upstart VC firms fighting for LP funds and deals, I think the team at Correlation was really smart to find a way to differentiate themselves. While it sounds like they are still in the early stages of establishing a track record, $350+ AUM is no joke for a VC firm. Question is, as deal data becomes more easily accessible online (and through sources like Crunchbase and Pitchbook), and other VCs start building their own databases, how long until Correlation’s quant-driven edge gets competed away?
Very interesting. Although I still have some reservations on the model, should it work, I believe this can break the very last limitation entrepreneurs face in the wake of raising money : getting geographically closer to where the source of capital is.
Provided this system delivers on its promise, I foresee entrepreneurs across the country (the world?) being able to jump-start their businesses and focus on growing it instead of having to move to the Valley area when it isn’t always necessary.
Very interesting model! One thing I took away from Founders’ Journey is the importance of the founding team (when it comes to a successful venture), and I am wondering the extent to which this model works to assess important attributes that are less data-driven like the chemistry between co-founders?
Thanks for your post! Similar to EIO’s comment, I guess this data model won’t be able to capture the qualitative insights gained from the conversations with founders and observing dynamics at the office. (For example, the investors are often betting on the founder’s abilities and potential).
Since the traditional VC model is based on “spray and pay” model, the historical VC investment data inherently has a majority of deals that won’t be successful, so it feels like Correlation Ventures will be working with flawed data to begin with. If the company only uses the “successful” investments as data, I believe there may also be a lot of noise and non-documentable reasons for those successes that will be hard for a data model to capture.
Very cool post! We may need to change the way that we discuss VC success when evaluating this type of model. You mention that the company points to investments in Casper and Virsto as signs that their model is working. But in a model in which they’re making so many smaller bets, is investing in a couple of successful companies really that meaningful? For a typical VC that is making fewer, bigger bets two great picks could be significant. But if Correlation is investing in over 100 companies per fund it seems almost inevitable that they’ll pick a couple of winners and their payoff from each of those investments seems like it can’t be that huge.
Great post, Jessica, thank you. This is so interesting, as the traditional VC model seems a bit under attack. As you mentioned, if this model can remove many of the biases inherent within investors, that could be a huge value-add for the VC team in assessing an investment decision objectively. I like the idea that a mission for machine learning, data analytics, and AI is to process enormous amounts of data, and freeing up the humans to use judgement and analysis — as we’ve discussed in class, and an area that AI still struggles with — to ultimately make the investing decision. Therefore, this could be a tool both for Correlation Ventures and, as you mentioned, a tool they can license — that’s an interesting thought too.
I wonder too how the entrepreneurs feel about receiving money from Correlation’s AI. If I were an entrepreneur, I would hope my investor is passionate about serving on my board and helping our company succeed – not investing because the algorithm thought so. I doubt Correlation lacks passion for the investments, in fact, I would imagine they would be more excited once the data supports their intuition. But as an entrepreneur, I want to be lock step with my VCs, utilizing their understanding of an industry and harnessing their guidance, to grow the company and achieve profitability quickly. I wonder how Correlation provides those sorts of services to its many investments.
Very interesting idea! Thanks for sharing! I agree with you that the biggest barrier here will be getting VCs to buy into the technology (ironically). I think many VCs already have anxiety about how much value they’re adding to the business. This idea really needs to be pitched as one where the VC can better organize and evaluate the deals based on multiple factors and then analyze the information/recommendations presented to them and make a final decision.
It will fundamentally change the VC industry though — it is certain that these algorithms also have some bias in them and only certain types of businesses might succeed. It might be necessary to give VCs autonomy in some of the input factors as well.
Thanks for a great post Jessica! I’m interested to learn a bit more about the factors included in the model – specifically, do they go to the detail of characteristics of the founders? Beyond the team chemistry mentioned above, I worry that if the model includes data about demographics of the founders, and is too backward looking, it may eliminate more diverse founding teams because the algorithm doesn’t have historical data or the ability to be forward looking.. given lack of data perhaps those factors are not included, but I worry about companies being able to hide behind machines, or use algorithms to justify unfair decision making.
I don’t understand how they can gather enough high quality data to make this anymore than a marketing ploy to differentiate themselves to LPs. Start-ups are by definition new and unproven, data is scarce and often incorrect. A lot of these start-ups are making bets that the future will not be the same as the last, so is historical data really a strong indicator of success?