The Data Scientist Spotlight is back! Meet Xavier Conort, Chief Data Scientist at DataRobot. Xavier joined DataRobot back in 2013, and has nearly 20 years of experience working within the insurance industry at organizations around the world. We sat down with Xavier to talk about his role as Chief Data Scientist, advice he has for future data scientists, where you can find him when he's not at work, and so much more.
Hi Xavier, tell us a bit about yourself.
I studied Statistics, Finance, and Actuarial Science at university in the nineties. Then, I worked as an actuary in the insurance industry. I worked in France, Brazil, China, and Singapore. I worked as an actuary, but also as a Chief Financial Officer (CFO) in China, and as a Secretary General & Risk Manager in Singapore. Most of my experience was in life insurance, and I started to work outside of that field in 2010. After leaving my work in life insurance, I took a year off to travel around and even went to Antarctica.
My first job outside of life insurance was when I discovered statistical modeling and the use of generalized linear models (GLM). I found that these were things I could enjoy doing. I started to feel a bit jealous of my colleagues who were working on these types of projects, so I did some research online and discovered that there were even more exciting things that my colleagues did! And that’s when I found machine learning. I fell in love with it and discovered all the keys to succeed with machine learning through online courses, tools, open source packages, and free books (The Elements of Statistical Learning).
I realized that I could apply all of those awesome techniques to many different problems, thanks to Kaggle. Very quickly, I started to compete on Kaggle because I saw that I could learn a lot, and I was also lucky enough to win competitions. I really enjoyed Kaggle competitions because they helped me to be exposed to different problems, not only insurance problems. It was a good diversity of problems, and I met a lot of data scientists from all over the world. This is how I met Jeremy Achin (CEO, DataRobot) and Tom de Godoy (CTO, DataRobot), the co-founders of DataRobot. I could see that they were passionate people just like me. It’s not so much the passion for coding, it is more about the passion for solving problems.
When did you join DataRobot?
I joined DataRobot five years ago, in September 2013. But actually, my contributions to DataRobot are even older than that. The first time I contributed to DataRobot was in April 2012 when we did a competition — Sergey Yurgenson (Dir. Advanced Data Science Services, DataRobot), Jeremy Achin, and I. The name of our team was DataRobot. And we competed again under the name DataRobot in 2012 when I visited Jeremy. Each time, we got good results.
Can you explain the role of a data scientist and share what a typical work day is like for you?
I guess that my typical work is quite different from most other data scientists because my role at DataRobot is to develop a tool for data scientists. I need to put my data science efforts into R&D at DataRobot. We develop the tool. I still observe the best practices in data science but my role is to automate those best practices in the tool.
We also create new features and new insights so that data science can be trusted by a large variety of users. So, we work a lot making machine learning more interpretable through visualizations and reports that can be understood by humans and not only by machines. And we work to make the solution as efficient as possible and to get the best accuracy automatically. We ensure that the machine does something smart and doesn’t get too naive.
How can you address the fears that the robots are taking over and that automated machine learning is replacing data scientists?
One thing that a machine cannot easily do is apply common sense, so it’s important that humans supervise the machine. That is why we work so hard to make sure that the machine gives interpretable results that can be screened by data scientists.
Also, there are so many things that companies want to do with machine learning that there actually aren’t enough data scientists work on it all. And I feel that we are just going to make the work for data scientists more exciting.
For myself, I fell in love with machine learning because it was automating things that I found a bit boring and because what I really enjoyed doing was solving problems. Here at DataRobot, we are adding a new layer of automation that will make it better. People who love to solve problems can solve more problems, and people will still be needed to make sure that common sense is applied.
What advice do you have for someone interested in pursuing data science?
Be sure that they’re choosing data science for a good reason. For me, the main reasons to pursue data science should be the ability to solve more problems and make things more efficient for people.
You need to suffer a bit to practice and work with multiple problems. Don’t work on only one problem. If you’re working at a company and you’re always focused on the same type of problems, try to do competitions (like those on Kaggle) because they will expose you to a larger range of problems. You will see that even if you are not a domain expert of the problem, you can learn to get intuition from the data about what will work.
Now that I work for a software company, I’ve also learned that writing tests is very, very important. It’s not only about writing code.
And, overall, it is very rewarding to write software that is used by other people because you can see that your work is useful.
Do you have a favorite DataRobot feature?
I think that my favorite feature is Prediction Explanations because it’s useful in different ways.
Prediction Explanations help people who are not familiar with data science to judge whether or not the machine does a good job of making predictions. If you give the reasons for why you predicted this value, the final user can better judge whether or not the machine did a good job. I think it’s a very good tool to build trust with non-technical users.
Even for technical users, it serves as a way to double check the work. It’s not enough to only look at the models, double checking explanations at the prediction level can reveal some issues.
Another reason for me is that this feature generates actionable insight.
For example, if you simply tell HR or a manager that someone is likely to leave a company, that’s not enough information for HR because they want to know why that person is likely to leave in order to prevent him/her from leaving. Or if you know that a customer is likely to buy your product and why this customer is likely to buy, you can adapt your approach to encourage more purchases.
So yes, I find that this feature makes predictions much more powerful.
When you’re not working, what do you like to do with your spare time?
I love hiking, but in Singapore, there aren’t many opportunities to hike. So when I travel outside of Singapore, I try to do more things outdoors. I used to travel quite a lot, and now a bit less. And I used to dive, but I don’t dive anymore.
Today, what I really enjoy is spending time with my family and my two daughters. We do things like bike riding and swimming.
I also enjoy good food, and I'm French so that helps. Since I'm living in Singapore, my food preferences have gotten more spicy, and I love it!
If you weren’t a data scientist, what do you think you’d want to do?
Maybe a baker. One thing I enjoy in France is getting up early in the morning, and having some fresh bread.
Interested in meeting another data scientist? Meet Jessica Lin, a Customer-Facing Data Scientist who primarily works on time series use cases with customers across all industries.
Learn more about how DataRobot works for Data Scientists here.