As we discussed in this post, it is not easy to schedule a tournament with the help of a computer. Most (if not all) youth sports tournaments are manually scheduled in the US currently (as of January 2026), and presumably around the world as well.
Most Basic Logic. Most tournament hosting websites claim to have an auto-scheduler, but most of them apply a fairly rudimentary logic. Which is - take one game, put it on an available field. If the time doesn't work (say there is a conflict), then put it on the next time slot. Keep doing it until you have scheduled all the games. The problem is that this approach doesn't work. Typically after putting about 40% to 50% of the games on the schedule, you run out of options for the next game. At that point the partial schedule in front of you is quite useless. You have to start all over.
Gold Standard Technology. There is a mathematical technique that is considered a gold standard for solving such problems. It is called linear programming, or a variation: mixed integer linear programming (ILP). In fact, this very powerful technique is used in a variety of different industries and contexts where you need to create schedules while satisfying constraints (like airlines, industrial production, distribution logistics etc.) and even other business problems that are not exactly scheduling (like building an investment portfolio to maximize expected returns for a given risk budget, deciding product mix to manufacture to maximize profits, where to build your warehouses or stores, deciding the path your phone call or data packet should take on a telecom network, and so on). In the real world, you would typically solve these problems in two steps. First, a decision scientist (or a team) would create a mathematical model that represents the desirable outcomes and constraints specific to the problem you are trying to solve. Then, you would use a solver (such as Gurobi) to calculate a solution to the problem. In fact practically every major professional sports league or tournament that we have come across uses ILP at some stage - usually followed by a person manually editing the output. So the obvious question becomes, why doesn't everyone use this method for youth sports scheduling too? Well, actually some people do. For example Exposure Events uses this method (they call it SchedulePro with AI). But it is true that many youth sports platforms don't actually use it. I am speculating, but it might be partly because it requires special expertise beyond the ability to create a hosting platform. It could also be because people have tried it but it fell short.
GotSport is an interesting case study. They actually have a specialized offering for professional sports scheduling (GotSport Pro), which utilizes ILP models to create schedules. But they don't leverage the same tools for youth sports. That is because there are some limitations to extending the technique to youth sports. They primarily stem from the fact that youth tournaments have significantly more teams than a professional league or a tournament. To understand why that is such a big problem, you would need to understand how you solve ILP models in the first place. This technique belongs to a class of problems called combinatorial optimization. Which is to say that there are many combinations possible, and you're trying to pick the best ones. The solution complexity comes from the sheer number of different combinations possible theoretically. Like we showed in this post, if you evaluated every combination, it would take you longer than the age of the universe to create a schedule for a 100 team tournament. Solvers like Gurobi use a method called "Branch and Bound". So if you have a relatively simple 100 team setup (all games are the same duration, and all fields are at the same venue), then Gurobi can generate a good 100 team schedule in about 5 minutes. But if you introduce multiple venues (and time it takes to travel from one to the other), or different game lengths for different age groups, the problem becomes much more complex (the solver may take more than an hour). And even if all you do is add teams, the time that a solver takes increases exponentially. So if you go from 100 to 200 teams, even a simple tournament would take 20 minutes. 400 teams would be an hour and thirty. 800 teams would be five and a half hours. In the youth sports market, if a model takes more than a few minutes to run, it becomes relatively unusable. That is because almost no one can configure all the constraints accurately upfront. After you run the model once, you realize "I forgot to add a field" or "I have this field ending too late, it should be sooner" or "A team just registered, I need to rebuild the schedule". 15 minutes to run a model would feel like eternity.
Our solution - a little bit of cheating. Instead of using a mathematically rigorous technique like ILP, we use something called heuristics. Which is roughly like knowing which doors to knock at because we suspect that's where we might find a good solution. It doesn't guarantee a perfect solution like ILP, but it gives us a strong schedule in 60 seconds - even for a thousand team tournament with 4 different game lengths, 40 different venues up to 90 minutes apart, and all the coaching conflicts and special requests. That doesn't mean it takes us one minute to schedule a whole tournament. There's quite a bit of preparation required before we get to that point, and some manual massaging after the model has run. But it does make our lives a lot easier than doing it manually or running an ILP solver.
Is it AI? If you look at Exposure's website or GotSport Pro, they call ILP as AI. We get this question a lot. Do you use AI? The answer depends on who we are speaking with. Usually when we're speaking with a layperson (i.e. not a computer scientist), it's just easier to say yes it is AI. But when speaking with an engineer, the answer is more nuanced. These problem solving techniques have existed for as long as computers have existed (at least 50 years). ILP or heuristics can only be called AI, if you consider an old fashioned GPS device to be artificial intelligence. It's not exactly machine learning. It is not prone to hallucinating like a generative AI tool such as ChatGPT might. It doesn't automatically get better simply by "training" it on more tournaments. (A human programmer has to go in and make changes to make it smarter). But it is a smart system, and within the field of computer science there is a name for it. Good Old-Fashioned AI.
How do we generate a schedule?
April 20, 2026