Hilary Mason is the chief scientist at Bitly, a web-based “curation” tool that allows users to save, share, and discover new things on the web. To do so, Bitly tracks social media data with the focus and intensity of a scientist trying to unravel the human genome. In other words, they’re data scientists. But Hilary Mason, Bitly’s chief scientist, calls them “awesome nerds”—tech-savvy science types who not only know how to mine the gold from an endlessly expanding information universe, but also know how to talk to normal humans about what they’ve discovered. Credit Suisse sat down with Mason to find out who these people are and what they actually do.
Credit Suisse: The Harvard Business Review called data scientist the sexiest job of the 21st century. You call them awesome nerds. So what exactly is it that they do?
Hilary Mason: A data scientist is somebody who can understand business problems, who can actually do an analysis that informs a solution to a problem and then communicate it successfully. But they do it by using a skill set that has never before been combined into one profession.
The basic skills are the technical abilities to get data out of a system and process it, and perhaps build infrastructure on top of it—that’s engineering and hacking. Then you need to do an analysis—that’s statistics, linear algebra and probability theory on the math side. And then the last piece is the combination of social science and curiosity and understanding business—asking the right questions, translating them into your mathematical and engineering analysis, then translating that into something you can actually talk to other human beings about.
CS: Where do you find these people? Are universities teaching data science?
HM: I haven’t hired anyone with a data science master’s degree because the programs are just starting. Data scientists come from all different fields, including a lot of academic scientists who are leaving academia and who can be trained to communicate. I’m a computer scientist, and I work with an astrophysicist, a physicist, another computer scientist and a mathematician. But I have peers in other companies and universities who come to it from political science and psychology. It’s such a young field that people are arriving in it from many different directions.
CS: It has fast become an elite position in a healthy IT job market. There are nearly 5 million IT jobs in the U.S. alone. Is supply meeting demand?
HM: It’s very hard to find people who have any useful experience. A lot of the companies hiring data scientists are hiring their first one, and that means they don’t have an infrastructure for mentoring or cultivating them internally. And people with even a little bit of experience are very hard to come by.
CS: But I’ve read about data neophytes taking a course or two, then going on to solve complex problems posted online. Some might say data science may be easier than you’d think.
HM: I am a little skeptical of that. Large companies can package algorithmic problems they need to solve and put a challenge online. But that’s a very defined error metric you’re trying to optimize. It’s like that little wrench you get from IKEA in your toolbox. When you need it, it is the perfect tool. But most of the time it’s not the best tool for solving your data problem. The job of the data scientist is to know what the problem is in the first place.
CS: So who needs a data scientist?
HM: You need a data scientist when you think you could be making better decisions than you are based on available data. Take what we do at Bitly. In the past, a marketer would just look at their own data and work with their own materials. But now we know what everyone on the Internet is paying attention to, and you can use that to inform your practice. And you can build products that just weren’t possible before. The way I usually describe the whole thing is: Level 1 is using data to make better decisions about the business you have now. Level 2 is taking your business in a direction that was never possible without the data.
CS: Where does the data chief sit on the company totem pole?
HM: My friend D. J. Patil, who actually co-wrote the Harvard article you referred to at the beginning, has the best way of describing this: You should think about your chief data scientist as your “Spock on the bridge.” He’s not going to be issuing the orders, but when Captain Kirk has to make a decision, that’s the person he will turn to.
CS: How do you structure a data team?
HM: You have a bunch of pieces—the data warehouse, the analysis infrastructure and the interface with other groups that need to use the data. So data teams tend to do a number of different things. They do business analytics: How healthy is my business? If I do X, do I make more money? They do product development: What is this cohort of users who came into my product at this time doing in the product, and what does that mean for the design? They build things like recommendation algorithms, search engines and spam filters. And they do a lot of research.
CS: Why can’t you just bring someone in when you need them?
HM: Hiring contractors can work, but you really want somebody who can internalize your data and systems, and just knows how to answer a question.
CS: You keep coming back to that—knowing the right questions.
HM: It’s not the kind of question as much as how you ask it. And that really goes back to following the scientific method: have a theory, look at the data and see if it confirms your theory, and then make a decision. So I can show you a graph, but we might have very different stories that describe what we’re seeing in that graph. And we might not be able to agree on an interpretation unless we’ve agreed on what we’re studying in the first place.
CS: How do you convince your CEO to go along with your proposed solutions to business problems?
HM: It’s very hard to argue with the person with the data.
No comments:
Post a Comment