Judging DataFest: Insights from the Other Side of the Table

Michael Sarkis, 2nd Order Solutions

_{We caught up with Michael Sarkis, a data science manager at 2nd Order Solutions and returning DataFest judge, to hear what keeps him coming back year after year. He shares what judges look for, how they weigh technical rigor against storytelling, and what makes a team’s work stand out.}

Tell me a little bit about yourself. 

I’m Michael Sarkis, a data science manager at 2nd Order Solutions, a niche consulting firm in Richmond, Virginia, we refer to as 2OS. I currently do data science work in the consumer and small business credit area, working with a wide range of companies—from fintech start-ups to top-10 US banks. I’ve worked on a broad range of projects, but almost all of them have involved using data to find insights and create solutions through statistical modeling. Before 2OS, I got my master’s degree in statistical sciences from Duke University and an undergraduate degree in statistics and economics from Cornell University.

How did you get involved with being a judge at DataFest, and what keeps you coming back?

2OS is a sponsor of the Duke DataFest, and we always try to send a few employees down to Durham to help consult with the teams and be judges. As an alumnus myself, I was happy to go back down to Duke’s campus and take part in the event. I also used to be a teaching assistant for Professor Fisher (who has run the Duke DataFest for the past few years), so I had some familiarity with the sponsoring team. What keeps me coming back is that I find it fun and rewarding to judge the competition and provide an industry perspective that can help future data scientists on their career paths.

What are your top five criteria for judging a DataFest project?

1. Business Insight – Is there clear insight being offered that will actively help the “client” who provided the data?
2. Communication – Are the insights and results effectively communicated in both the slides and verbal presentation?
3. Validity of Analysis – Is the analysis correct, and does it support the points presented?
4. Visualizations – Are the visualizations used in the presentation useful and aesthetically pleasing?
5. Statistical Complexity – What statistical methods are being used and how ‘difficult’ is the analysis the team did?

How do you weigh the merit and depth of an analysis versus the clarity and interest of the data story?

This is a tough one because both are very important in the competition. However, I do weigh the “clarity and interest of the data story” a decent amount more than the “merit and depth of the analysis.” Coming from consulting, I am trying to evaluate the team’s work from the client’s perspective. A lot of the time in industry, the story and the why of the analysis matters a lot more than the actual process or complexity of the analysis. This is a shift from the academic perspective a lot of students are used to, where the merit and complexity of the method can take the forefront. In my experience, not many clients want to know the ins and outs of the statistics used (just that it is correct) and are more concerned about how they can use the results.

What would you tell DataFest participants is the most important thing to consider when they’re putting together their DataFest presentation?

Focus on the insights you are providing the business. The key insights and results are what really matter most from a company’s perspective. It is still important to show the approach used, but there is a very limited amount of time and space (3–4 slide limit) with which a team can present their work. A lot of the exploratory data analysis and specific details on methods will likely need to be cut from the final presentation to leave room for developing a compelling story with the few good nuggets of information your team discovered. 

Specifically on the data visuals, a couple of well-made and clear graphs/plots is a lot better than a bunch of quickly made ones. Also, try to make a visualization your audience will be able to understand without you explaining it.

What’s challenging for you about being a judge?

The biggest challenge for me as a judge is the limited amount of time and information we have to evaluate a team. All the student teams that have presented to me have clearly put in a lot of work, but only so many can be passed to the final round and eventually win some of the awards. As a judge, I hear a team present for four minutes, see their four slides, and then likely get to ask one or two questions. Due to the number of teams, judges, and the timelines, it all makes sense, but it is an added challenge, as each team could likely present for at least 30 minutes on their analyses.

As an industry professional, do you see DataFest as an important component of a student’s education in terms of supporting future data science career opportunities?

I think DataFest provides students with a unique opportunity that is hard to find in a classroom. The freeform nature of the analysis plan, the usually very messy real-world data, and the communication aspects of the event are things many data scientists meet in their everyday work. It challenges students to use the skills they have learned and apply them in a collaborative setting to reach a solution.

How is DataFest similar to or different from working as a professional data scientist? 

DataFest is very similar to my work as a professional data scientist. As a data science consultant, getting new data from clients, analyzing it, and then making a slide deck to share our insights is a very common project structure. The challenges student teams face with messy data, clarity of analysis, or story building are a lot of the same ones I see in my work. However, the most important similarity is probably the fact that the final output is a presentation. As a consultant, communication is incredibly important and that’s where I see the DataFest really shining.

In terms of things that are different, DataFest is a lot more condensed and rushed than most of my projects. It happens over only a weekend instead of a couple of weeks to several months. DataFest is also a freeform analysis problem and, while that is not uncommon, there are projects that have very structured analysis plans with well-defined goals. The last difference is that there isn’t a senior person on the project advising. A team of new data scientists wouldn’t be placed onto a project by themselves to solve an issue. There would be a manager or adviser working closely with the team to make sure everyone was on the right track and issues they encountered were solved. This difference, however, is a great learning opportunity for the student teams to be the main decision-makers and get experience working independently under a tight deadline.

What advice would you give to a DataFester about how to make the most of their experience?

Don’t leave your presentation until the last minute! Make sure you put some time and effort into making it a polished product so your team can put its best foot forward for the judges. You and your team will need to prioritize your time so you have a finished product by the end of the weekend. This means you may not be able to dive as deep as you want or you may be limited to only trying a few statistical methods. Prioritization is a very important skill in data science work, and this will be useful experience.

Don’t worry about having the most statistically complicated methodology. A linear regression that offers a useful insight is a lot more powerful than haphazardly throwing more complicated methods at the problem. The second piece to this is to make sure you really understand the methods/models your team used, both for your own knowledge and because it’s a lot easier to present a topic you know really well.

What’s one of the best team names you’ve ever heard?

There have been a lot of good ones, but a recent one I liked was the Standard Deviants because I love a good stats pun.