Sramana Mitra: How do you make money? What is Kaggle’s business model?
Jeremy Howard: We get paid for running these competitions. Generally, we get paid by organizations that also access this amazing community of data scientists. A lot of people invested a lot of money in data collection and data management, and a lot of CEOs and CIOs are looking at all the money they have spent and the huge strategic opportunity there is in their data. Then they are wondering how to unlock it. So, they need smart people with access to various tools. Those people are hard to find. The McKinsey Global Institute, which did a study a couple of years ago, stated that the number one challenge for organizations in unlocking their data is the shortage of data scientists. We sell the access to that community, whether through competitions or direct work relationships.
SM: When we met at the MIT Center a few weeks ago, you mentioned that you are funded by the Khosla venture fund.
JH: We are funded by Khosla Ventures, Index Ventures and Max Levchin, the founder of PayPal.
SM: Have any of these investors funded any of the companies that are incubated in your community?
JH: The one I mentioned – Jetpac – I believe was funded by Khosla ventures.
SM: Is that a trend? With so much activity in your community, there have to be interesting businesses that are forming.
JH: The bulk of our business at the moment is more in the Fortune 500 companies. The reason we created the startup program recently is because we think there is a lack of understanding in the startup community about the ability to use data science to improve one’s value proposition. I think the issue is that most founders of startups tend to either have software engineering or service business development. They don’t generally know much about data science. They are usually unaware that some of their toughest problems can actually be solved by algorithms. I would say at the moment, there is underutilization of Kaggle amongst startups.
SM: Are the 75,000 data scientists working on interesting data sets and Fortune 500 problems becoming startups themselves?
JH: They are. Some of our top participants have started consulting firms in machine learning. Our top data scientists on the platform have their Kaggle profiles, and they get a lot of page views and people who ring them up. They are not exactly creating startups in the sense that we would think of them. They are not startups that are building leverageable products. They are generally building one-man-shop consulting companies because this is what they are good at and what they like doing. These guys just love analyzing data, and they don’t want to spend time building a startup.
SM: What about intellectual property? When somebody is on a competition and a number of data scientists work on it and they submit the solution, who owns the intellectual property for it?
JH: The intellectual property rolled around a particular competition can be set by the competition sponsor. Generally, for public competitions, we recommend that the [competitor] gives a non-exclusive license to the algorithm developed. For private competitions, there is usually some kind of ownership around the IP. But that always depends on what the sponsor wants. In the case that you lock it down more, you might get less participation, because obviously people do not want to lose access to algorithms that they came up with. Generally, we try to suggest to the company what they need to be commercially successful.
SM: This is one of the coolest stories I have encountered so far.
JH: That means a lot coming from you. Thank you.
SM: It is amazing the way you guys have pulled this together. I really think that the psychology part is very important, and you are one of those people who instead of playing computer games you like solving complex problems. I think that is the real psychology behind this.
JH: I used to play computer games, but not anymore. I think this is much more fun.
SM: Congratulations. I love this story. Thank you for your time, Jeremy.
JH: Thank you.