One of the greatest challenges in all of sports doesn't occur on the field of play. Building a sports season schedule!
Sports Industry Data and Analytics – 2023 Article Series - Article #2: The use of optimization and what role it plays in the scheduling process.
Intro:
Building a professional sports league or college athletics conference event schedule is one of the most complex and daunting challenges in all of sports.
It typically occurs far from the playing field, is normally managed by a very small number of analytically driven staff, and comes with a set of objectives and constraints, that in my opinion, far outweighs the number of decision-making factors across the respective playing environments. I’m sure some coaches and GM’s will disagree with me on this one.
To some it’s an exciting (and at the same time stressful) problem to solve. When I refer to the “problem” I’m specifically speaking to a mathematical problem in which schedule makers must consider hundreds and even thousands of constraints to build a schedule that meets the needs of several stake holder groups including the league/federation (professional sports), the athletics conferences (college sports), the individual teams, venues, fans, and all the requirements that set the athletes up for success both on and off the field of competition. All the while providing transparency and credibility to all stakeholders.
In my opinion creating an optimized schedule is much like an ongoing negotiation between several stakeholder groups. When all stakeholders feel a similar level of unhappiness, we know an equitable schedule has most likely been created.
How do schedule makers solve one of the most complex mathematical problems in all of sports? With use of data, analytics, and most importantly complex optimization techniques. Optimization as it relates to data and analytics is not a household name unless you work within the world of operations research (OR), or within industries like manufacturing, supply chain, healthcare, logistics, etc. But it’s one of the most important and sometimes underrated parts of analytics. And it’s no surprise it can play a major role in helping create the best possible schedules across the different sports markets.?
To understand Optimization's place in the analytics ecosystem, lets quickly discuss these types of analytics: Descriptive -vs- Predictive -vs- Prescriptive (and Optimization)
To better understand optimization’s place in the world of data and analytics, it’s important to understand the difference between these three types of analysis.
I’ve come across many definitions of Operations Research and Optimization over the past several years. I would best define it as a method of mathematical problem solving in which the use of advanced analytics techniques and algorithms are applied to help find the optimal solution for a given problem, or challenge.?
Now, back to sports! First, let’s talk about the “playing field”:?
When most of us think about our favorite team(s) sports schedule, we think to the specific opponents our team will play in the upcoming season, what games we can attend with our friends and colleagues, and the menu for our tailgate party. Fewer of us think about the total number of games/events in each season, and even less of us would consider how many distinct events actually take place in the respective league across all teams, or the total mileage the teams will travel in a given season.
In analyzing some of the most popular sporting leagues in the United States we can see that the number of events each league has to consider is much larger than most would think. This is more pronounced in the NBA, MLB and college sports, but regardless of the number of events, they each come with many similar challenges when creating the optimal schedule.
Next, let’s talk about the objectives and constraints:?
Objectives (Goals):
Building professional sports league, and or college athletics conference schedules starts with specific objectives which can include the respective league commitments (for example how many divisional opponents must a team play in a season), or several other factors noted in the below list spanning a broad range of considerations.
To those working in Operations Research, “objectives” may imply a more technical problem-solving definition in which numerical function(s) are used to try and maximize and or minimize something. For this section “Objectives” is meant to focus on the business objectives, or goals in creating an optimized schedule.
Like many things in business, we might assume revenues are always top of mind, and for sports scheduling it is very important. However, as seen below there are other factors just as important, and all of which are considered when creating a sports schedule.
Constraints (Penalty based scoring system):
The process used by most sports leagues and conferences to create their schedules rely upon constraints (see below examples) and the accompanying data points using a penalty-based scoring system. The number of constraints will typically increase in volume each year and can seem endless, can be variable, and the cause for much of the stress in creating an optimal schedule. For some of the sports leagues there could be upwards of over 1,000 constraints to consider.
The penalty-based scoring methodology requires an assignment of a scoring “penalty” value for each data point. These values can be subjective at times and reflect the schedule maker’s opinion on the importance of each data point as it relates to coming up with the best schedule. An example would be making sure there are enough premium rival matchups on key dates (e.g., holidays). The schedule makers can influence certain schedules from not being returned by the optimization process by assigning very high penalty values to certain data points. This subjectivity is also a large part of the challenge as the “beauty is in the eye of the beholder”. At least to some degree.
Each constraint when not satisfied generates a penalty score. Ultimately, the schedule with the lowest total “penalty” score would be the optimal schedule. Examples of constraint categories are as follows:
The Future of Scheduling Optimization:
The original sports industry scheduling process (I’ll call it scheduling 1.0) has come a long way from the days of white boards, lotus (I’m dating myself), excel and other like programs. And with it came a mentality of “this is just a nuisance” and an attempt to finish the effort as quickly as possible with a good enough result.
We have now experienced schedule making 2.0 for several years in which use of penalty-based constraint scoring, analytical models, optimization algorithm’s, more compute power, and a move to the cloud all contribute to allow these very complex problems to be broken up into smaller jobs/processes allowing for a much more efficient solving process.
Now we look ahead and the early stages of schedule making 3.0 in which we want to start to move from optimizing schedules based solely on a constraint-based /penalty scoring system in which humans still have a lot of subjective input, to the use of machine learning (ML) and artificial intelligence (AI) that can complement optimization techniques by helping to better analyze thousands of schedule options more quickly, building in more predictive analytics capabilities, and to provide better results around the following areas (see below list). And in speaking to the subjectivity of assigning penalty values, ML and AI may be able to contribute to creating the penalty values more accurately in the future.
To dig a little deeper, I had SAS R&D / Optimization colleague Sertalp B. ?ay to weigh in. “The use of ML and AI and optimization is not an either / or thing. In scheduling optimization, ML will be used in the future to help make the optimization process more efficient. To locate a good initial solution or to detect unfavorable combinations, ML can be helpful for optimization to achieve the best results. By using ML, we can speed up the process with the ability to better make predictions in a quicker amount of time and achieving the best results.”
One additional comparison to make when speaking about optimization with Sertalp is “Heuristic” techniques -vs- “Exact” optimization. Each has its place in the optimization ecosystem, but to keep things simple, Heuristics will aim to find a good solution more quickly, but in many cases the result may not be the optimal solution. On the other hand, Exact Optimization looks to find the most optimal solution, but can have slower solve times and certainly can be more computationally expensive. With SAS Viya this would not be a problem using an in memory engine.
The choice of what type of optimization and analytical methods to implement will largely be contingent around cost, the specific problem you are trying to solve, and expected performance with a tradeoff needing to be evaluated between accuracy and what’s the most efficient way to implement the technology.
How SAS Can help professional sports leagues, college athletics conferences, and technology providers to accelerate sports scheduling goals:
For more information on how SAS has impacted the sports industry, please visit https://www.sas.com/en_us/industry/sports.geo.html
To learn about how SAS optimization is being used across industry, please visit please visit https://support.sas.com/en/software/optimization.html
Additional Sports Optimization Articles by SAS Colleagues Sertalp B. ?ay and Caslee Sims, Jr. :
And last but not least, for those interested in the Fantasy Premier League, please check out SAS Colleague Bas Belfi and Sertalp B. ?ay 's podcast series "FPL Optimized "
Enterprise Account Executive @ Bandwidth Inc. | Direct Sales
1 年This is really interesting stuff, great article! Cool to see the impact that Data & Analytics has in optimizing this. Our personal sports/activities calendar could use some of this optimization as well ?? ??
Driving EMEA and Global Marketing projects @SAS | Connecting dots | Combining Sports and Analytics when I can!
1 年Great article, thanks Dan Axman for writing it up! Indeed, there is a lot we (or, Analytics in general) can do to help with schedule optimization. It might not be something people know about, your article summarized the possibilities very well. ??