$1.5 million bet he can beat the markets

+ Add to

Arnaud Castillo got $1.5 million in prepayments from customers who think he found a way to make accurate predictions about financial markets…the weather…venture capital…and anything.

Arnaud created CrunchDAO, which holds online tournaments in which data analysts compete to see who can create the best predictive models. Skeptical? He might remind you of the Netflix Prize, the $1 million tournament in which a team of outsiders improved Netflix’s recommendation algorithm by 10.05%, or his entrepreneurial track of proving skeptics wrong.

Arnaud Castillo

Arnaud Castillo


Arnaud Castillo is the founder of CrunchDAO, which is building the future of asset management.


Full Interview Transcript

Andrew: Hey there freedom fighters. I’m Andrew Warner, founder of Mixergy, where I interview entrepreneurs about how they built their businesses. Before I started recording this interview, the founder you’re about to meet told me something interesting about the Netflix prize. See back in 2006, Netflix wanted to improve its recommendation engine.

So it created a million dollar tournament. Interestingly, the people who won weren’t just outsiders, but they were a team of many competitors who joined up and won the Netflix prize of a million dollars. And that idea of a tournament where outsiders can compete to improve on a prediction model, that idea Is the basis of the organization you’re about to find out about.

Arnold Castillo created CrunchDAO to create competitions like the Netflix Prize, where outsiders could compete to create better prediction models. This interview is sponsored by Origami, which creates DAOs for ambitious communities. To hear interviews like this, check out my DAO series on Origami’s site.

Go to join origami. com slash podcast. You’ll see how the team behind origami got over 1000 Y Combinator founders together to form a venture Dow that made headlines for raising 80 million, as well as other stories of people who are creating new organizations like the one you’re about to hear. Again, go to joinorigami.

comslashpodcast. Here’s the interview. Give me an example of the type of analysis that you expect the CrunchDao team to do in the future. The

Arnaud: obvious one, because this is our DNA. It’s in finance to make a market prediction. Tomorrow, it could be any kind of problem. You could think of a weather forecast.

You can think of a medical, a cancer. You could think of even of a VC data set. And there’s a lot of AI in all these subjects. But today the main focus is on finance. And we believe that because finance is a complex system, because when you look at the stock market, there’s a lot of noise and finding the signal within the noise is really complex.

And this is where. Our work can be really useful because we can access predictions from 2, 000 data scientists and basically make an ensemble of the best solutions. And the ensemble is in AI much better than any good solution.

Andrew: Why did you need to form a DAO to do this? Why couldn’t you just organize a regular company and start handing people rewards based on their performance?

Arnaud: DAO is that you can reward. Everybody based on their contribution. That’s very important that you can have everybody working in the community and having your share of the performance of the, of the community. But you can have a lot of different subjects that you can work on. This is why we have several Boonties from time to time, and it’s much flexible and easier to do it in the DAO format.

For us, when we talk about DAO and Web3, we really see it as a tool. It’s a tool to incentivize and align the performance of the data scientists who work with us. And the business in general, if

Andrew: it was one of the more traditional organizational structures, you might pay a data scientist for enhancing the data for sharing their models.

They would get paid, and that’s the end of it. With the DAO, they get paid in your token, which is the crunch token, which if they hold on to, allows them to benefit from the upside of the overall organization and other people’s contributions to the research. Am I right? In

Arnaud: today’s environment, in the crypto environment, you cannot really say what you just said, so I’m not going to say you’re right.

People receive the token, they can exchange it for USD, like if it’s just a standard payment token, that’s what they can, you can see it as a payment token. But the thing is that you can do much more with a token that you cannot do with USDC, which is just a, just normal bill. Basically today we have the tournament and people are rewarded.

You have to think that maybe tomorrow we can ask people to not only give a predictions, but put stake in the game. You’re not going to do it. It’s not going to be flexible to do it with dollar or whatever invoice or whatever, but you’re doing it with a token is very easy. Like you do a proof of stake on Ethereum.

You can just take Ethereum tomorrow. You can maybe stake Crensh and if your prediction is good, you get a higher reward and so on. And there is also a way to incentivize. People who hold the crunch, maybe they get a better share of the profits or, uh, contribution of the Dow. They can get a higher reward.

So it’s much more flexible than a normal system like the old system. And the reason that

Andrew: you say that you can’t repeat and agree with what I said earlier is You don’t want your token to be considered a security, you don’t want it to be considered a way of giving people a profit for other people’s work.


Arnaud: exactly right. That’s any project today, uh, you have to be very careful. So our token is… seen in Europe as a utility token in the US as a payment token. And we’re fine with that. And basically we’re not selling the token to any US person. So we know we don’t get any problem as whether it’s a US, uh, investment contract.

That’s the fact today that things are not clear. And we’re waiting for the SEC to make things very clear so that we know what we can say, do. Like any project,

Andrew: just give us some clarity and we’ll work within the clarity. Give us lack of clarity and we won’t be able to work. Um, your team told me that you have revenue of 1.

5 million. How could you have revenue? Of that size at this point, it’s a pretty early

Arnaud: project. Who told you that is not, I think the team, it’s probably PR agency. Okay. This is token that we sold. So it’s future

Andrew: turnover. Isn’t it more like an investment where you got 1. 5 million? Oh, I see. You can’t use the word investment, but it’s, you sold those tokens.

You ended up with 1. 5 million in your

Arnaud: treasury. It’s turnover, it’s future turnover. It’s like you’d make an invoice for future turnover and you receive dollar or euro or whatever or Ethereum for future turnover. Who paid you that and for what? It’s investors are mainly a lot of people from the financial industry, a lot of family offices, a few VCs because they want to support the project and they believe they can use the service in due time.

Andrew: Because their vision, some of them is… They want to have an advantage in making predictions on investments. If they could tap into CrunchDAO and have your team, your community, give them better analysis than others, they’ll have an outsized performance.

Arnaud: Exactly. If they want, they can use tomorrow. The community to get a prediction on their own problem, whether it’s a market neutral or long only portfolio, they can give us their data set.

We can play it and they can pay with the crunch token. They bought in advance the token because they bought it probably cheaper than what they would buy today if they had to do it today.

Andrew: I see. They’re buying the tokens for the ability to use those tokens in the future to get analysis. Or they could sell it, presumably, if they decide that they don’t want to use it themselves, let someone else use it.

Arnaud: If there is a market, but today we don’t have any market for the token. .

Andrew: That’s what I’ve seen. I think I have to buy through your site and sell through your site, right? Through App dot

Arnaud: Investor A U US .

Andrew: I am in the us Yes. So you, so I can’t do it. So how is it for you working in this kind of environment, you’re a person who’s been in finance four years, you’re an entrepreneur who started three companies, two of them at least have sold.

How is it for you to operate an environment where you keep having to justify your legitimacy before you can even do the work you started out to do?

Arnaud: So what we focus on first is the community. So we’re building the community. That’s the key, important thing for us. Second, I love finance because as you said, I’ve been in finance.

I’ve been investment banking. I worked at Lehman brothers as a financial analyst. I worked at Dutch bank. I was professor in finance in Paris and I’ve been an entrepreneur for 20 years. I wanted to come back to finance. So for me, it’s being able to be an entrepreneur in finance is really what I like the most.

And we’re just building. And we really believe in this DAO approach where we have a community, we can work together and you have different stakeholders in the community. Obviously, you have people who bring money so that we can live because as any startup, you need to have a run, you need a runway. So basically you need financing, but you also need the community and you need a team.

So this is what we do and the DAO is really great to do this. How did you get the

Andrew: original early DAO

Arnaud: members? We started doing a hackathon. That’s what we did. Uh, because we wanted to test, actually we wanted to test our own data set. We, we started to do, uh, our own data set. And we said, okay, let’s try to, to see what we get from the community.

And we did a hackathon. And we actually were impressed by the quality of what we got. It was better than what we did internally. Um, but what is interesting, for example, you have a lot of AutoML projects, like H2O, another one, I don’t recall the name, but they are doing, they are selling their product.

Actually, they are valued like billions of dollars, and they are doing great doing machine learning and predictions. But if you rank these guys in our tournaments, they are below the top 100. And if you make an ensemble of the best solution that we have on our end, it’s even better than that. So for us, this is the, the big advantage is that having a community is, is much more powerful.

Andrew: So how did you get the original people who are in the hackathon? I have

Arnaud: three team members from the founders who are from school 42, which is an IT school in Paris, very famous in Paris. And we did the hackathon in, in, in this school. So the first guys came from this


Andrew: And then once you had that hackathon, you were able to build and grow the community.

How? Mainly

Arnaud: word of mouth. You have a lot of data scientists who like to do tournaments. Whether it’s on Kaggle, whether it’s MRI, whether it’s a touch challenge, it’s another one, and when they start to learn and to hear about these tournaments, then they, if they like what you do, they start to, to join and it will grow like that.

It’s really

Andrew: organic. I didn’t know what school 42 was. I just actually asked Chatupiti about it. It was started in 2013 by a French billionaire. It’s a tuition free computer programming school in Paris designed for based on peer to peer collaboration, project based learning, and I see that in a short time, it’s gained a lot of prestige.

Arnaud: Yeah, I think they are, they are the same probably in 10 or 10 countries. They have seven school 42 in 10 countries. And yes, it’s really, it’s really funny because it’s peer to peer review. It means, first of all, you don’t have to have any diploma to get into this school. You have people who started, like, people in our team, some started coding like at 12.

And what they like to do is coding. And they probably, maybe they fail in their studies. They did nothing. But they can still go to the school. You have a, actually, the first exam you do is online. You have three questions that you have to write, solve online. And then, if you are selected, you go to something they call the swimming pool.

You have one month where you can work 24 hours if you want, and they maybe take the 10% best at the end of this one month. I see. So you, so they just, you’re just being selected like that. And then you go, and then the school, you have to do different exercise and you’re reviewed by your peers. There’s no professors.

Okay. So you just code different languages, different projects. And you have a peer review and at the end you get your diploma. I could

Andrew: see then why that community would be such a good community for what you’re envisioning for CrunchDAO. They understand results, they understand working through this type of peer reviewed experience.

Okay, how did you get the investors? We

Arnaud: basically had one guy who was really interested in AI, who came to us and said that he was really interested in the project and basically… He connected with many, many investors. That’s the way it happened. So it’s basically business angel who helped us. I

Andrew: noticed that you say you have how many data scientists in the community?

Today it’s 2,

Arnaud: 500.

Andrew: But when I went to the discord, it was fewer than 2000 people in the discord who had signed up ever. Where’s the community

Arnaud: then? You have 2000 people in the discord?

Andrew: I thought it was fewer than 2000 people. Maybe I’m misreading it.

Arnaud: I think it’s 2200 today. Okay. Uh, maybe not all the data scientists are in the Discord channel.

Andrew: So then how else do you keep the community

Arnaud: aligned? They come to the, to the website and they can play the tournaments. They download the data, they play the tournament, they push the predictions and they, that’s where they go.

Andrew: So it’s 2000 people who are not necessarily talking to each other, but participating together through these tournaments.

And I think they get a notebook, which is a digital notebook, right? With the data that they need and so on. Exactly. You had just announced a prediction company. So

Arnaud: let’s, yeah. So tell me about the tournament is Adyalab market predictions tournament. Adyalab is an independent organization, obviously doing research, and it is sponsored by Abu Dhabi Investment Authority from the Emirates.

And so what is very interesting is that Adyalab is trying, is testing different solutions and Adyalab believes. Using a community of data scientists, like using crunch tower, basically, they can get better predictions. That’s what they want. They have, they can access anybody they want. They can have, if you look at the, the board of, of idea lab, you have PhDs, even Nobel prize, so they can have any data scientists they want.

They know that if they get predictions from their own. Everybody will follow the best prediction, so you will end up having one prediction. Using a community, you have different predictions, different people giving different skills, and that’s a better, uh, that’s a better approach to ensemble and to get the

Andrew: best prediction.

And I asked you before we got started, what are they looking for predictions for? What data are you getting? And you said, We don’t know. We don’t know even what the data is, right?

Arnaud: Exactly. We have no clue what the data is. It’s totally obfuscated. So we know it’s, we believe actually, we don’t even know. We believe it’s about the stock market, that it’s equities.

But we have no clue. We don’t know what the ratios are. They have, we have features in the data set, because in the data set you have the X, which are the features and the target. We don’t know what the features are. We don’t know what the target is. We don’t know what they’re looking for in the target.

Could be total return, could be sharp, could be some kind of residual, we have no clue. But the good thing is that they gave us this data set. The data scientists are taking this data set and doing their skills, which is machine learning. And whether it’s a financial problem or a medical problem, doesn’t make any difference.

The data is totally obfuscated. It’s standardized. And so they can do whatever, they can do whatever machine learning they want on this data set. Data

Andrew: scientists can take data that they don’t even know what it’s about, and somehow crunch it, and end up making predictions about what other data will be added onto it in the future, without knowing anything about it.

Arnaud: Exactly. They have many skills. They will look at correlation between features. Try to eliminate some features, try to create new features that can make synthetic features, whatever. There are many, I don’t even know what they can do, but. They have many skills they can apply, and then they get the

Andrew: better prediction.

So you’re saying they might even have a hunch that the weather somehow relates to the data they’re given. They could add that weather data in, try to see if there’s a correlation, if there is, no.

Arnaud: Actually, no. Because there would be a leak in the data. So they don’t know, they have dates, but dates are not just, they don’t have the dates.

It’s just, it’s just telling you this is coming after this. It’s just that you don’t know which year, which month it is. So you can’t even say the weather or whatever. The GDP at this period, or the inflation at this period was X, and you add it in the data set, you cannot. So unless the inflation is already in the data set, you cannot add any information into your data set, into

Andrew: the data set.

Only what’s there and that based on that, without any outside knowledge, you have to make a prediction. Exactly. Which

Arnaud: is great because it makes the competition really a machine learning competition. You cannot come with any, with like financial skills or your own data. Whatever you can have to make your prediction is in the data set already.

You got

Andrew: started as an entrepreneur three companies ago. What’s the first business that you started? So the

Arnaud: first business year 2000, it’s an internet business. I did photo printing business. Yeah. Take me back to

Andrew: that time. Netflix started. Netflix started with DVDs in the mail.

Arnaud: Exactly. So we did the

Andrew: rolls in the mail.

So they would send you rolls in the mail, you would print them out and mail ’em to their house?

Arnaud: We would, they would send rolls by mail. We would print them out. We would give this the, actually we, you could view your photos on the web, on the website. Okay. You could pick the photos you wanted to get to.

And then you would receive only the one picture that you would like on the 24 roll picture. It’s a great idea. The idea was that, obviously, we knew we were going to move to digital. That’s what, that was the goal. But the, maybe the good or bad idea, depends how you see it, was to say, okay, since 98% of the market is still on rolls, just like Netflix wanted to do DVDs, we said, we’re going to do the rolls to start with, but we will get clients on board.

And then when they move to digital camera, then we’ll have the clients already. But the, to tell you the end of the story, it was a good idea from a strategical standpoint, but from an economical standpoint, it was a very bad idea because the cost of receiving a role, scanning a role, showing you the pictures, and then having you to pick which pictures you wanted to get was not really, basically the acquisition cost of a client was very expensive.


Andrew: servicing was expensive. The acquisition came from where, how did you get new customers?

Arnaud: It could come directly or we were doing partnerships with, we did a partnership with like, what’s called, for example, TISCALI, which was Internet Service Provider. We made partnership with ClubMed and then we would get a huge amount of, of roles.

But we were happy to go to move into the digital world because then the model became much more profitable. That

Andrew: was a model where people took photos using digital cameras. Yeah,

Arnaud: they just send by, by the, by internet, uploaded their photos and received prints. And then you moved at the end, in year 2000, probably six, we moved to photo books printing.

You just put photos into a photo book and have them print. I saw on

Andrew: your LinkedIn that you had reached 25 million in sales before you sold the business and it came through acquisitions. Acquisitions of what? Yeah,

Arnaud: we met several acquisitions in the process of building this business. Obviously, Kodak was selling a lot of businesses, divesting everything.

We bought a photo finishing lab, a digital photo finishing lab from Kodak, because we needed the labs. It’s very complicated to set up a lab in France because you have regulations. chemicals, so it’s really regulated. So we were able to build the business through acquisitions.

Andrew: It stayed the same over time.

Yeah. Once it went from printing rolls of film to printing digital, that was where you were.

Arnaud: Yeah, mainly, yes. The main What’s the big

Andrew: takeaway you took from the business, from building it?

Arnaud: Exactly what we started to talk about is that you should really, maybe it’s better to take, when there’s a new trend, a new business, like the digital camera was coming in, maybe it’s better to Only focus on the 2%.

And do it well, so that when the 2% becomes 100% basically at the end, you’re best, you’re the first, you’re the first on this market. Instead of doing rolls, in order to be able to get the digital. You see

Andrew: the people who are still living in the past knowing that eventually they’re going to live to the new future.

You say, who are the wackos who are living in the future, even though it doesn’t make sense today? They have digital cameras. They can’t really print it. They’re not a lot of them. Their friends might think they’re a little bit weird. Focus on them with the vision that they will become the future that everyone else is going to be like them.

That’s a big, that’s a big, important lesson.

Arnaud: So what makes sense on the strategical point of view, because you say, okay, 98% of my market is here today. So maybe I should focus on that. Well, I think it’s better to start little, but to focus on the 2% that will grow to eventually 100%.

Andrew: What about Assureback?

This was the insurance broker that you started in 2013. By the way, it looks like you took a few years off. Like, 3 years, you get to go take a vacation, do something fun?

Arnaud: Actually, when you sell a business, you have to work a little bit for the acquirer. I see. So, basically, I sold the first business in 2008, and I spent years.

Working with the, the, the people who acquired the

Andrew: company. And then the next business was the insurance broker, Assureback. Yes. What did you do there that worked? And what was something that you wish

Arnaud: you’d done differently? Actually, it worked, it worked very well because we created a system, a broker, where we could compare different offers, and then we would recommend basically the best offer to the client.

The difference between a typical comparator, where you take the clients and you give the comparator, to the broker or to the insurance company, We were able to be, we were actually the, the broker ourselves. So we were selling the contract to the client. So we are not just getting a lead and getting a client and selling a lead.

We’re getting a client and selling an insurance for several years, on average, five

Andrew: years. The standard model was I go in, fill out a form, do a search, and then I get hit with somebody else who I have no connection with, who your site says is a good match for me. You make a sale once and that’s it. But being a broker, you get an ongoing relationship, ongoing revenue, and there isn’t a third party that’s in there.

Arnaud: Exactly. We work like a world broker because we had, we were able to sell the contract on our own website. We are delivering the contract. It could be AXA, Allianz, generally this big European insurance company, and, and we were doing all the work. We were. Getting the premium in paying the claims. We’re doing the whole work for the insurance company.

We’re like an IT partner, but the good thing is that we’re able to give this comparison and give the best quotes to the clients because the clients, if you look at the Even a car insurance, you think they’re all the same, but they’re all different. So depending on your car, maybe you should better be on this, in this insurance company rather than this one.

If you have a Tesla, you probably have big difference between one contract and another. Why

Andrew: did you sell that business in 2019? I

Arnaud: wanted to come back to finance, actually. I wanted to do this, it was not crunched out at the beginning, not exactly the way we do, we’re doing it today, but I wanted to come back to finance.

I wanted to use AI in finance, basically. What was the original vision? We started with our own data set and trying to see if, if with, uh, AI, we could actually have a strategy. And from that, we moved to, with this hackathon to, to make tournaments. And then we decide recently, we decided that we can, because we have this great community, we can actually have other clients, power their hedge fund or whatever, use our community to get predictions on that,

Andrew: on their data set.

Now that you’ve got people, not computers doing the research. How do you keep your community going? What do you do to keep the, to keep people incentivized, to keep them caring, to keep them

Arnaud: connected? The idea is that when we onboard a new client, the process will stop as a step one will be that you make a tournament.

In this tournament there is a big prize. In the case of Adyalab Market Prediction Tournament, it’s a 100, 000 prize for the, which is shared between the top 10 winners, and that will bring people to the community, obviously. But then there will be another one. And ideally, if you have a few new clients every year, that’s a lot of money to be shared with the community.

And then once this first step one is done of the tournament, what is interesting is that the data scientist. They don’t only take, give predictions, they share the codes that they use. So we can get new predictions on the fly going forward. And they will be remunerated and rewarded for any usage of the code that they share.

So basically, once you have worked on a tournament, you leave your code there. And if the client wants to use your predictions once more, every week, whatever, you get a share of the profit

Andrew: again. The share comes in the form of Crunch, the token that you’ve created, right? Yes. I looked at some data scientists write ups of their experience, and one of the things they said was, when they join, they can’t just go and participate, there was somebody who would onboard them, a human being who talked to them, who walked them through, right?


Arnaud: we, that’s what we’ve done so far. We do one to one onboarding, because we want to know who, who are the people who are joining. There’s also a rule that you cannot create several, uh, accounts, you have one account, one person, one account. So yes, we’ve been making one on one onboarding so far. And it’s that one

Andrew: on one onboarding.

It is the notebook that I mentioned earlier, which is a collection of data in, not a file, but in an online experience, right? Where they could get it. Anything else that you do to keep people engaged? Yeah, there is

Arnaud: a Discord channel also, where people can come and ask questions and, and learn and, uh, so we also have this Discord,

Andrew: which is great.

I’m in the Discord. I wanted to go and see how people communicate, what they do. And then you also have governance, right? Token holders get to govern and guide the organization.

Arnaud: Yeah, exactly. Whenever we want to propose something, actually, even actually the communities often propose new things. We have votes on the different proposals.

We had, for example, big question on whether people should stake on their model. And how it should happen, I told you, for example, that we believe like in the Ethereum mechanism, you have a proof of stake. We thought that in our system, we could also have a proof of stake, so that data scientists can actually put stake in the game on their predictions.


Andrew: you make a prediction, and you give some tokens to say, I believe in it this much. Exactly. I’m really

Arnaud: confident I did a great job because there is a way, there is a way to do predictions and data scientists know it. It’s very easy to get wrong or to what we call over fit. And so the way you do your science is very important.

And you are, if you are very confident. And what you’ve done is good. You can stake tokens on your prediction.

Andrew: All that you’re saying makes so much sense. Community gets tokens. They get an ability to vote. They get an upside for helping each other and helping the organization grow. They get the governance.

All of this is what goes into DAOs. As a smart person who’s been in business, in finance and DAOs, maybe you can answer this. What’s the challenge that DAOs are having? Why aren’t DAOs more prolific considering how logical it is the way you’re describing it? Yeah, I

Arnaud: think in general DAOs will, will grow.

There are many models, business models, where DAOs are very, would be very powerful. Not all the business models, but I was talking to someone recently, he said, why Uber wouldn’t be a DAOs, DAO basically. You have members who are people who are driving their car and you have clients and that’s it. And the IT behind it is very small.

I don’t think it’s that big of a deal. So if everybody was organizing to a DAO and they decide what development they do, basically you know what you have to do. That would be more, much more powerful, I think, to have a DAO in this, in this example. But there are many businesses where, where you would see it.

And I think it’s going to come. It’s coming. There’s going to be more and more DAO. There’s going to be also, uh, better regulations to explain exactly how it should work and, and what you can do and cannot do. Since we’re working with, with Adyalab, who is in Abu Dhabi, we’ve looked into Abu Dhabi global market.

They just issued a first draft of the DAO foundation they want to launch. In the coming weeks and we’re talking, looking into this because we believe that’s the way to go. So I think there will be more and more DAOs, but the obstacle

Andrew: today to having more of them is, why aren’t there more now? It’s

Arnaud: a, it’s, I think it’s a matter of also a philosophy of how do you want to see the business?

People have come from a couple of centuries of shareholding and you have to think, okay, can I do a business without having a shareholder? Only stakeholders? And that’s the big, big challenge even in terms of raising. Maybe if you have an equity company, it’s easier than if you’re a Dow. Obviously, with the winter we’ve, we had, we’re having right now, probably people are thinking equity is easier.

Andrew: But you’re saying long term, this just makes sense. It’s going to take a while for people to figure out where it makes sense and how to work with stakeholders who are part of the business, not just investors in the business.

Arnaud: Exactly. I think we, I believe in a system where all the stakeholders have a stake in the game and can build together the business.

Andrew: All right. And that’s what you’ve got going on right now at CrunchDAO. One of the things that I love about CrunchDAO and other DAOs is you could see the inner workings of the organization without being a part of it. I did it. You just go into the site, which is crunchDAO. com. And then go into the discord community and see how people interact, see what they’re working on and, uh, and just be a part of it before you’re officially a part of it.

Thanks so much for being on here. I appreciate it. Thanks for having me, Andrew.

Who should we feature on Mixergy? Let us know who you think would make a great interviewee.