Early stage venture investing has little data to draw from to make good investing decisions. So how has Connetic Ventures successfully developed a data system to inform their investment decisions? We chat with Chris Hjelm about the process they’ve used to develop something that does just that.
Chris Hjelm: We’ve back tested the data on a large number of companies and so have a data asset that’s 400 to 600 companies with some returns data . . . the companies with the highest success fit a small number of these profiles.
Ginette Methot: I’m Ginette,
Curtis Seare: and I’m Curtis,
Ginette: and you are listening to Data Crunch,
Curtis: a podcast about how applied data science, machine learning, and artificial intelligence are changing the world.
Ginette: Data Crunch is produced by the Data Crunch Corporation, an analytics training and consulting company.
Ginette: Today, Chris Hjelm of Connetic Ventures chats with us about how Connetic Ventures uses data to guide their investment process with early-stage startups. Let’s jump right in to our conversation.
Chris: I’m a junior partner at Connetic Ventures, and we are an early stage data focused venture capital fund. And my background, out of school, I worked at Dunnhumby, which at the time was the largest data analytics company in the world. We analyze transactional data for over a billion households across the world. And through that I worked with retailers, CPGs, customer product goods companies, and we used data to change pricing, assortment and try to influence people’s behavior in retail stores. And so that was, I worked there for eight years, and then after that I ran my own proprietary trading fund and did that for three years and kind of built on kind of my data analytics background; it was much mostly algorithmic trading. And, uh, at the time I was renting space from Connetic, and they were starting to build this data platform, or at least explore how we could use data to make better investment decisions in venture capital, and a position opened and, uh, joined the team and now I run the Chicago office and oversee all data.
Curtis: Got it. Okay. So kinetic ventures then is about three years old., Is that right?
Chris: Yeah. Early 2015, so I think, yeah, three and a half coming on four years now.
Curtis: Okay. And it’s an interesting problem, right? I mean this is kind of the holy grail, right? If you can predict the companies that will do well and you can invest in those companies, that’s, that’s a pretty powerful thing. So tell me, there’s lots of people talking about this and trying to do it. Tell me, you know, have you guys cracked this, and what’s your approach? Why is it maybe different from what other people are doing?
Chris: Yeah, so in venture I think there are two different almost industries. Right? There’s early stage venture capital, which is investing in very early businesses. Um, you know, maybe not even generating revenue, uh, haven’t found product market fit and really investing in a team and, and an idea. And then there’s growth stage venture capital, which those are companies you’re reading about that are, you know, soon IPO. We Work, Airbnb, ones that have become household names and are taking on capital to continue growth and eventually exit, whether that’s IPO or a private sale. And so we, we’re an early stage, which in growth it really depends on the sector and how much capital is put into the business. At We Work, they make $2 billion, they lose 2 billion, and their business is built on really how much capital they can take in. So I don’t know how useful data is in that world, but in early stage there’s really nothing on these companies at all. If you’re a public company, you have quarterly reportings, you have all of the other stuff that comes with that, that you can build algorithms and models off of. But in early stage there isn’t any thing. You, you meet the team and you kind of hear the idea. So we really think early stage is where you can start to use data to create a picture and ideally an algorithm that predicts success.
Curtis: Interesting. Now you said it’s, it’s difficult, right? Because in early stage, you said, there’s not really much to go on. So, so what data are you looking at? How are you, you know, what do you, what are your sources? What are you looking at to try and make this prediction?
Chris: That was something we struggled with. So we started on Wendal. So Wendal’s our technology platform where companies apply and we collect our own proprietary data through. And because early stage, uh, you can’t do financial modeling, well you can on some companies, but it’s not accurate. You maybe have a year’s worth of financials to go off of. So we ended up building, uh, over half of the data points we collect are behavioral on the team and the founder, everything from ethics to personality type to experience—pretty much anything that’s ever been published or research as maybe being correlated to entrepreneurial success. We collect through Wendal.
Curtis: So you’re collecting this, this personality data or data about the people that are involved in the company. So maybe baked into that is the assumption that the people really are, that’s the predictor, right? It’s not necessarily or maybe the biggest predictor, right? It’s not necessarily, you know, product or, or, or fit or timing these other things that people talk about. Is that, is that the assumption that you’re running off of?
Chris: Yeah. So we, um, over the last six months we’ve worked on and just recently launched our own behavioral assessment as part of Wendal, which is called Team DNA. And a person takes it, it takes three to four minutes to complete. And the end result is one of 22 profiles, and each profile is put into a role fit with a, an executive role or a team role at a startup. So let’s say there’s 22 profiles. We have X number of profiles that our CEO fits, and we’ve back tested the data on a large number of companies and so have a data asset that’s 400 to 600 companies with some returns data. And so we’ve built this knowing . . . it’s great, when we looked at it, the companies with the highest success fit a small number of these profiles. And so now we collect that and we pretty much only invest or well, a large part of algorithm is fed by that behavioral assessment.
Curtis: Got it. That’s interesting. And so, I mean the training data here is really interesting. Innovative, what you did, you looked at these companies that had already gone through the stages that you’re looking to take companies through, looked at their, you know, their leadership and found found with this could be, was there, um, are there psychologists on your team or things like this to, to develop this or how did you guys approach that?
Chris: There’s a number of open source behavioral assessments that, um, or at least the data is so, uh, predictive index culture index, you know, Myers Briggs, they’re all built on research that was done 50 to 60 years ago, and it’s publicly available. Um, and so our initial, kind of, the core of what we built was built on that data. And then we wanted this to be software. Most of them currently are consulting based assessments. So you got this output and you need an interpreter to tell you who you are pretty much. And we want it to be software. You take this, Hey, you’re a CEO or your technical specialist, you’re actually a CTO, CIO, but in a CEO role. So it really allows us to have good conversations with the companies. A lot of the times, you know, at first it was kind of scary. You don’t want to tell someone they’re not suitable to run a company and not that you know it’s are, not that that’s true or not, but that’s what we believe. I would say 90% of the time they already know this about themselves, and it lends itself to a great conversation. I’m like, okay, well for us to be an investor, we would really like someone on your team that fits this profile that’s really missing from your team, and it allows us to be a much better partner going forward. And it was something early on that was kind of scary to talk about and now it’s right almost a hundred percent of the time. And I think it helps us build a bond with companies as well.
Curtis: That’s interesting because that gets to a part of, you know, machine learning and data that, that I really like to think about or touch on in that is when you get the results w what do you do with them? How, how are they received, how do you take action on them? And it sounds like people are pretty receptive to these assessments that you’re doing ’cause they kind of ring true to what they know about themselves.
Chris: Yeah. And you know, that’s a, I think you asked earlier what the difficult part is, and I think the hardest part in venture and why most people aren’t trying to tackle this is the average exit timeline from investment is eight and a half years. And so for us to get a really good, like we’re in our infancy, Wendal’s probably in elementary school. We have, we’ve proven he’s two times better at predicting bankruptcies and series A than the average human. But we only have say two to maybe three years of returns data. And really that’s, that’s not exits, that’s just performance. And have they gone on the raise series C, series D, but to get really good data, we need, you know, at least six to maybe even 10 years. And that’s, I don’t think that’s a challenge or a problem that most people want to sign up for.
Curtis: Sure. That’s a long, that’s a huge lead time. And then of course you’re changing it as you go, right, and trying to improve it. But you said you’ve been able to, to distinguish that versus a human, it’s better, for example, at predicting bankruptcies. How did you guys come to that number?
Chris: Uh, so that was based on all of our prior investments as well as investments in our ecosystem that have gone through our database that, um, so pretty much when you take Wendal, you’re in the Connetic database, and we track performance and indefinitely. I can’t say how often we reweight, but we have various APIs and are connected to different data sources that measure performance, whether that’s exits or financing rounds. So everything, once something material happens, the algorithms reweighted. And so we really just, we found companies, we know that, we knew the outcome or we know are really successful or that have announced bankruptcy. And then we went back and, and we looked at, and we only started using Wendal four to five months ago. And so we just went back and said, what would Wendal have said about this company? Um, and that’s how we got to, I forget the exact sample size initially, but I think it was 80 to a 100 companies. So a pretty, pretty good sample size.
Curtis: Yeah, that’s pretty good. So people then, the way it works is, people just hit your website and they fill out the form and they fill out the assessment and that kind of tells them, you know, who they are, what they’re good at, these kinds of things. Then you guys do a call and make a decision, but also a recommendation on “Hey, you should get someone else as the CEO,” or maybe maybe not. Right? Is that kind of how you operate?
Chris: Yeah, there . . . so the assessments one, we call them modules. There are five different modules. Each one encompasses something different. So there are a variety of chatbots, forms to fill out, and the assessment. So it kind of, they’re all pretty fun and interactive. And the first module say is a company overview, which includes financials, fundraising, history, experience of the founder, MRR, ARR, so actual revenues, and location. So, you know, for us it’s, is this company even a fit for us? We don’t invest in San Francisco or New York City, just part of our investment thesis. But really each module is like, it goes down, is this company a fit or what is some of the behavioral ethical components of the founder? We have a valuation module that creates over 200 calculations on the back end. Is this deal priced right for us? Is the team, right. So then it gets a final score and a ranking. So it takes 10, yeah, eight to 10 minutes to complete, and we get a really good picture of is this a company we’re interested in, how’s the team and as the price right.
Curtis: That’s cool. That’s cool that you can do that in just just 10 minutes. And this may be a side note, but you said you don’t invest in San Fran or New York. I’m curious why.
Chris: So another data decision, oh, and Boston, so we don’t do those three cities. So in any region, outside of those three cities, you get 68% more shares for your money. Meaning . . . I don’t think I’ve updated it in a month or two, but the average seed round in those three cities is 6.4 million. Anywhere else it’s closer to four. I think it’s 3.8 or 3.9, and the probability of know unicorn or exits or it’s actually almost exactly the same across every region in America. So just because you’re paying more, you’re not getting better talent and you’re not getting a better company or a higher probability of a successful return.
Curtis: That’s interesting. So, so maybe better to invest in the, in the Midwest or something then . . .
Chris: Yeah, well, you know, there’s adverse selection. If a company can’t raise money in San Francisco and is talking to Connetic Ventures, there’s gotta be something else going on . . . our brand just doesn’t carry enough weight yet where . . . but, yeah, it was a, a data decision.
Curtis: Okay. Interesting. Can we talk also a little bit about how you built this system? There’s obviously a lot going on under the hood here. You guys took a lot of time to think through the design of this, which is really great. I’m curious also about the technical side of it. Was it difficult or how did you guys approach building a system that could actually do these, do these calculations for you? Are you, I dunno, leveraging sort of more traditional machine learning models. Are you leveraging deep learning? What did you find works for this?
Chris: Yeah, so this was, as I mentioned, probably three years in the making of, and we knew we wanted to collect data on companies, and besides that we spent a year or two really just figuring out what is out there from research, and and the common thing is if I take the leading data indicator, if you’ve had a successful exit and your odds of success go up 60%, so you know, there’s all these little snippets of studies that have been done that have data backed in science. So we knew we wanted them to ask questions and collect some of this info. And so we’re using third party software, different survey platforms. We were doing calculations in Excel, and it was just unstructured and a complete mess when we tried to look at everything together. And so we tried, we started as a partnership with a machine learning company in Cincinnati, which is where our headquarters is actually in Covington, Kentucky, across the river. So they helped us kind of work through the framework, and we use Microsoft Azure platform for it. Uh, and then it was just taking too slow. And so we hired an engineer full time over the last four to six ,, and we’re at this point we’re running a venture fund and we’re very much product managers. And our managing partner, Brad Zapp, is kind of the head Wendal product manager. So we, we know everything that we want to do. We’re just not technical experts and so eventually we’ll think later this year we’ll have a full team, four to five people that just work on Wendal.
Curtis: Got it. Okay. It sounds like you were able to get something working with outsourcing a little bit and then just having one per one hire on your team. Cool. That’s awesome. And you said you’ve been using Wendal now for a few months. Have, ah, and again I understand your lead time is you know, 10 years in the future, but is there any sort of leading indicators that you watch or have seen, uh, get better as, as you’ve been using Wendal?
Chris: We’ve been so focused on the behavioral component and getting that right and making sure, uh, ’cause for us it’s all about throughput at this point. Um, we need enough companies in the funnel spitting out, let’s say we have 125–150 companies a month on average. We need at least 10 of those companies to be a pass in our system, which every company gets a star rating, one to five stars. We only move companies in the human diligence that are four-star and above. So for us it’s all about, recently has been about tweaking the algorithm to have enough throughput and working on the behavioral aspect. So of the data that we collect, that’s not part of the behavioral assessment, we’ve not really looked at which of those questions are leading indicators of success, which should we be tweaking? We kind of want to get at least a year or two of getting our model down and collecting the same data across every company that, that hasn’t been a priority yet.
Curtis: Yeah, that makes total sense. Now you mentioned something before we got going here that I wanted to touch on because I thought it was interesting. You said, um, that someone actually contacted you, a PhD student, that had been studying this problem and said that pretty much you guys were the most innovative or, or, or the only ones doing something that she thought would work is that those are my own words. Please fill, fill in what actually happened here.
Chris: Yeah, so probably not, um, it was a student doing her master’s thesis in Portugal. That was on venture funds using AI or machine learning to make investment decisions was the thesis, and she talked to Google Ventures and Correlation Ventures. A lot of, so a lot of European funds are, I would say Europe is more innovative in their approach to using data. And you know, we have a number of reasons why we believe that is, but our conversation was to date, everyone is using data for screening purposes or for like defined companies, and platforms like PitchBook and Crunchbase kind of make that easy because I think every fund, if you’re not in San Francisco or New York, has a deal flow problem. And so funds are starting to use data to how can we scrape off Twitter or LinkedIn for companies fundraising or how can we use PitchBook to find companies that are raising in our stage.
Chris: So people are using data to find companies, but no one is collecting proprietary data and using data to make an investment decision. That is still a majority vote of their general partners. And separate thing, we can talk about the bias that that creates in venture. You know, most of the general partners are white males, and less than 6% of venture funding goes to women and minority founders. And it’s been kind of long documented that that’s probably the main cause. And so, you know, for us personally, we love the data aspect and the ability to remove the human bias from the investment equation, is pretty fun.
Curtis: Got it. So, so I’m assuming then the way you designed the, uh, the algorithm, obviously you wouldn’t design any racial bias into that, but it sounds like you were particularly careful to make sure that what you were doing didn’t, didn’t have those biases inherent in it.
Chris: Uh huh, yup. Even on the, or talking to someone, Oh, you know, a university is correlated to a success in one paper. If you graduate from an Ivy league, your odds of success are, forget what the exact number was, but I think 1.6 1.8 times higher. And that’s something we’re not comfortable in using, just like race or gender. As a startup brought it up to me about six months ago, like this platform probably does a good job of removing bias. And we looked back and 42% of our investments are in women or minority founders. Venture average is 6%. So it led to an eight times higher rate of funding just by using data.
Curtis: That’s awesome. And you know, that’s such a huge problem nowadays, especially well just in general. But with machine learning and things, you know, a lot of these algorithms are tuned on data that has some sort of inherent bias in it. So, so that’s awesome that you guys have been able to, to create something that, that overcomes that and, ah, that, that’s really cool. Um, we’re coming on time here. I want to be respectful of your time, but you just made one comment that I, and maybe this is another side note, but I was curious about it. You said that, uh, firms in, in Europe you feel are more innovative, uh, with data, machine learning, things like this. Where’s that? Uh, wha . . . can you explain that a little bit?
Chris: Yeah. So Europe is essentially middle America, uh, in terms of, there’s all these countries with cities, but there’s not a great concentration of startup companies, uh, like there is in San Francisco or New York City or Boston. You know, in San Fran, you raise money, you hang a sign and companies come to you. Uh, but in Europe, if you’re in Berlin, you have a very limited pool of startups and talent to invest in. And so there’s a $200 million fund in Berlin. Uh, they can’t deploy all of that capital and companies in Germany. So they have to find ways to find and invest in companies, in Tel Aviv, in London. And so you have to create, essentially all of Europe, I think is still less than 25% of the deal flow in San Francisco. So you have 180 square miles in San Francisco, over 2 to 3 million, uh, in Europe. So they, they have to be scrappy and find ways to find good talent, find good companies, and invest their dollars. And I think that’s where we, I think because we started in Cincinnati, Ohio, we faced a very similar challenge. There’s not enough talent in Cincinnati to run a venture fund at this moment. And that’s kind of how the whole idea came about. Um, how we got to where we are now.
Curtis: I just want to leave the last word with you. If there’s anything that we didn’t cover that you think is important or interesting or anything you want to leave the audience with, ah, we’ll leave you with the last word here.
Chris: Yeah, I would, um, I’d like to use kind of the opportunity and the platform, uh, if anyone knows companies. Some of the, for us, this model only work if we have people going through it, and we’re not doing this to be mean and saying your company’s bad or good, you know, we’re, we’re just like a startup, and we’re trying to build something that we think will enhance venture investing. And so I would just say if you know anyone raising money and, or if you’re just interested in what we’re doing, please reach out to me. And companies go to either connetic.ventures or visit Wendal.io if you’re fundraising and you just want to see what the outcome is, and you know, we do our best to try to provide feedback timely, uh, to everyone if they’re not a good fit for us at the moment. And depending on amount of deals, that could take, you know, two hours or it could take a week. So just that would be great.
Ginette: Thank you, Chris Hjelm, for being on our podcast. If you’d like to see this episode’s transcript or our attributions, head to datacrunchcorp.com/podcast.
“Loopster” Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License