Quantitative Research in Pro Sports – From Stanford to the Front Office

John Sears, MS&E MS ‘15, is a Special Advisor for the LA Dodgers. This workshop covered his career in front office sports, advice on how to get onto this career path and what to think about before/once you’re on it, and thoughts about the next, most exciting areas of exploration for sports quantitative analysis. 

The basics of overall sports team management:

• The team President or General Manager (GM) is the front office equivalent of CEO. They make the decisions re: on-field/pitch talent.
• Coaching is optimization of deployment of team, pitch calls, player development, injury prevention.

The Front Office specifically (where John works):

• This is the team making personnel decisions – draft picks, who to trade and for how much etc. It’s headed by the team president/GM, and has two main umbrella talent evaluation departments:
  • Scouting. 
    • Amateur - looking at the high school/university level and determining what those players will look like in 5-10 years.
    • International - looking at prospects outside of the US
    • Pro - looking at people already in the leagues, and determining what they’ll look like tomorrow.
    • Both are paying attention to trajectory and character/personality - who a player is on and off the field.
    • Scouts are expert evaluators often with history of playing the game at a high level. Have deep insight into things that can be harder to capture with data, and can pattern match to other players with similar abilities.
  • Tech group. This is where the engineering, product, data science, analytics, and design teams all fall.
    • Video is now in use for this too, SQL queries, stats. 
    • Quants – how can we forecast the future? Should we sign/trade? This team will build models to forecast for all the different plates, project their key attributes that predict their talents.
    • Sport scientists who understand the biomechanics, heart rate data, etc.
    • Analytics group is strategic, eg, we have this game on this date against this team, how should we sequence pitchers against their team. 
    • Data and software engineers to build custom tools, programs are often not bought off the shelf but built internally.
    • At smaller sports teams, the same people do everything.
    • All of these groups work together on a daily basis, though scouts are on the road so much that they’re not in-office as often.
    • Baseball is most advanced use of analytics and data, then basketball. Much bigger teams of analysts and engineers.

How to get into this career path:

John’s story:

• Super into sports as a kid, collecting player cards, started sports analytics clubs in college, but didn’t consider it as a career because he figured it was an old-boys network.
• Instead, went to work at the Federal Reserve, and was being encouraged to do a PhD.
• Decided to apply to one out-there job for every PhD application. One of those was to the Houston Rockets, and he got a call within days!
• Interviewed several rounds with Houston Rockets, and decided it wasn’t the right time to do it - didn’t have the skills to do the analytics the way he’d want to, wanted to learn those skills and do it properly.
• Applied to MS&E MS program as the best place to learn these skills – fundamental treatment of decision making under uncertainty, stochastics etc.
• Started Stanford sports analytics club while here.
• Started working with the Philadelphia 76ers part time while working on MS, then joined full-time after graduation. Left after shakeups at executive level.
• Went to Uber, where the work utilized his skills in decision making under uncertainty: eg, how to spend the money they were raising? Turned out there was lots of overlap with sports. Joined a startup, Motive, where he knew strong leaders from Uber.
• Went back to sports with Timberwolves in 2021 to lead Basketball Analytics.
• Now with LA Dodgers. Works as Senior Advisor, building quantitative models for decision making.

John’s advice and thoughts:

• There is not a lot of networking to get in the door. If you can do the work, you’ll get an interview. So send in resumes! 
  • Teamworks is where jobs will get posted, follow your target teams on Linkedin, follow people in the industry to find out about where jobs are being listed – send resumes to those people.
• In contrast, at the high level eg team president, it is a lot of who you know and what your involvement has been. 
• Make sure you work for a team who will listen to you - ask questions to gauge this during your interview process.
• Internships - many teams do have them, especially in baseball. For basketball it’s less so. The confidential nature of the work means it can be tough to bring people in for such a short time.
• Sport specific knowledge can help you be more comfortable and gives a headstart. It’s not enough to just know how to build models, you’ve got to get your hands dirty and think about other aspects, eg, doing a lot of back testing on updating long term models, dealing with the uncertainty of small data (cf, eg, Uber, where there are millions of rides per month, in sport, there are minimal plays in an even smaller number of games). It’s easy to overfit your models, and assume that the past can 100% inform the future (eg, how Steph Curry totally changed how basketball is played).

What John would work on if he was in school today?

• Machine learning on skeletal tracking data. There is a flood of this class of data coming online in every sport -- how can we build end-to-end ML models that learn interesting structure from the raw data.
• The interesting ways in which statistical models break down (ie. team misses 27 straight threes, data are not IID Bernoulli trials), and can we systematically learn anything from this beyond just explaining it away, "the model can't capture everything."
• Induction and the assumption of the future being like the past -- this is at the heart of all predictive modeling, but in an era when team strategies are constantly being optimized, how can you avoid the unpleasant outcomes that come when .