Blog Post

Venture capital in an age of algorithms

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

We live in an age of algorithms. Cheap computing and abundant software have brought huge advances in all kinds of fields. Data scientists are accurately making algorithmic predictions about weather, sports, disease, political elections, romantic partners and even box office hits.

Meanwhile, venture capital remains pretty old school. It is more art than science. So why aren’t more VCs using data science to guide their investments? As chief technical officer of Ironstone Group, a data science-based venture investment company, this is a question I’m especially interested in.

Venture capitalists (VCs) hear lots of pitches from startups, and usually use their personal intuitions to separate the good from the bad. For an elite few, this intuition-driven process has delivered fantastic returns. Yet most of the roughly 500 US-based VC funds have performed rather poorly, with the national venture capital index showing negative industry returns for the 2000 – 2010 decade. This isn’t meant to give VCs a hard time; picking winners is tougher than it looks. But why not branch out into a more data-driven approach?

geo data

Objections to the idea of data science in venture investing usually fall into two categories:

  1. First, some say it’s impossible. They argue data science can’t predict the fate of startups better than seasoned intuition.
  2. Secondly, even if it were possible, others argue it still wouldn’t be desirable. These folks believe data science-based venture capital could have negative social or moral implications.

And yet, it’s already being done. One unnamed data science-driven VC fund, with $165 million, oversees around 40 startups in its portfolio. Drawing on a database with 20 years of VC funding data, this firm promises funding decisions within two weeks; is never the sole investor in a startup (co-investments only) and does not take board seats. The firm uses inputs from startups, scores them, then subjects the highest scoring candidates to human review before funding decisions are made.

Similarly, my company, Ironstone gives investment candidates an answer in two weeks and uses a “hybrid” strategy combining human and mechanical processes, rather than a 100 percent algorithmic approach. If the data science says “no,” there’s no deal. If the algorithms say “yes” there’s a second layer of human screening. Ironstone uses data from a startup questionnaire, along with inputs harvested from a variety of other sources and is willing to be the sole investor or to lead a funding round.

Google Ventures, with over 170 investments behind it, also utilizes a hybrid system where analytics are deployed before startups are funded. Bill Maris of Google Ventures has said, “We have access to the world’s largest data sets you can imagine, our cloud computer infrastructure is the biggest ever. It would be foolish to just go out and make gut investments.” More traditional firms are also rumored to be using more data science nowadays.

Beyond the obvious benefits of picking winners and increasing returns, what are the other benefits of using data science to fund startups?

Speed. Data science has the virtue of speed. Inputs can be gathered and crunched, often producing fast, definitive results. This is why some firms like Ironstone are capable of funding decisions in two weeks. In contrast, traditional venture investors can be quick to say “no,” but slow to say “yes.” If they like a startup, VCs can spend months debating and turning over every rock.

Consistency. Traditional VC decisions are based on gut feelings. As a result, they’re subject to a litany of cognitive biases and human inconsistencies. People can be profoundly influenced by room temperature, what they ate, how much they slept or gobs of other random factors.

Algorithms may seem dull, but they can be consistent. They don’t get tired, sleepy, hungry, moody, or seduced by charming founders with impressive pedigrees.

Evidence-based. Entrepreneurs love to complain that VCs seem to judge their businesses by random, irrelevant or flawed criteria. Cognitive biases and intuition can distract VCs with noise, rather than focusing them on signal. In contrast, data science is the literal discipline of separating signal from noise. While algorithms may vary in terms of quality, they’re at least based on empirical hypotheses and data as opposed to whimsical feelings in the proverbial gut.

Incremental improvement. The underlying logic of an algorithm is explicit and transparent. You can literally see how decisions are made. This lends itself to improvement because it provides a stable baseline upon which to refine things over time.

Intuition isn’t as readily transparent, methodical or explicit. Gut hunches aren’t routinely subjected to anything close to the level of empirical validation a well-made algorithm endures as a matter of course.

It’s important to note the double standard. For an algorithm to be considered in venture capital decisions, VCs and their investors would almost certainly demand reams of data, statistical robustness and empirical validation. They’d insist on a meaningful sample size, tight correlations and explicit definitions. Compare this with the gut intuitions of VCs that usually have zero statistical testing, empirical validation or explicitness.

Scalability. Due to the potential for speed, consistency, accuracy and improvement, algorithms may drastically reduce the time and labor required for quality deals. This could allow VCs to grow fund sizes without requiring ever-larger deals.

On average, active VC funds do around four deals per year. This number isn’t only a function of how much money the fund has to invest, but is also constrained by the time and labor required to do each deal without sacrificing quality. The resource-intensive nature of traditional venture capital becomes a challenge as funds grow. As a result, bigger funds want to do bigger deals, even when it doesn’t otherwise make much sense.

VC resistance to data science may come from lacking will, skill, or a combination thereof. Some VCs are simply opposed, others are waiting on the sidelines to see how it all plays out and a handful are taking the lead by using data science to actively do deals today. Regardless of one’s predisposition, there’s no denying it’s the dawn of a new era. The age of algorithms is here. Is venture capital ready?

Thomas Thurston is Chief Technology Officer and Fund Manager at Ironstone Group, a venture capital investment company, and CEO of Growth Science, a data science firm that predicts business behavior using algorithms and technology.

11 Responses to “Venture capital in an age of algorithms”

  1. In some cases this will work. However, data back in 1962 would have told everyone not to make an investment with Arthur Rock into the Fairchild boys. Success in startups is about people, and the network early investors can bring to bear. That often is random. Even if I don’t have a key person in my current network, I can go and try to meet them.

    I agree that it’s great to look at data to inform you. But it cannot be the primary driver of VC decisions because so much of it is randomized.

  2. Hi Thomas we would be interested in having you run your algorithms against our startup which is a social network for Tech Pros. We give products social presence and familiar social tools for our users to find the best products based on crowd sourced reviews and recommendations of their peers.


  3. Thomas: I couldn’t agree with your approach more. But, I would not ever argue that the purpose of “data science” is to “predict the fate of startups.” And, I have absolutely NO idea as to what “negative social or moral implications” could possibly arise. I run a company called The Venture Alliance and we’ve been using proven scientific methods to rate companies for over 12 years. We’ve analyzed >50,000 entrepreneurs from start-up stage to liquidity stages and we’d never want to suggest that our methodology would predict future performance. All we use our process for is to help Angel and VC investors make better decisions. And, it works. In 2009, Forbes started a list called “America’s Most Promising Companies” and they used our process to filter >10,000 companies into a list of 20. Those 20 went on to raise >$92MM in the next 12 months. And yes, most of them are still doing well. We don’t take credit for that but we do admit that IF you choose well, your chances for continued success are better.

  4. mikedorsey

    Great article, especially about scalability. What once was a manual process done by well paid junior analysts can now be done automatically by algorithms – and MUCH more effectively. is a great example of tools being made available to individual investors that can level the playing field.

  5. I have to agree with most of your points. The VC industry needs to be more scientific than what it already is. However, from time to time, I’m thinking that there will be an “outlier” startup which will not conform to the metrics of a traditional successful company. Are cases like this possible? How should you proceed when the idea is really good but the data is off?

    Angelo from SeedAsia (

    • cdlingHQ

      Good point re: outliers. Which is why our platform use two different proven methods of analyzing expert reviews of startups. Prediction Markets to reflect an aggregate view. And a non-compensatory visualization that we call “Bean Grader” so that our customers can take a deep dive on exceptional views of why this particular startup may be the “Magic Bean” that is set for break out growth and stratospheric returns.