Vancouver Whitecaps: Boosting Performance Through Data-Driven Insights
The Vancouver Whitecaps FC has consistently demonstrated a commitment to excellence in the Major League Soccer (MLS) and other competitions in North America. With a rich history starting in 1973 and an engaged fanbase, the Whitecaps have solidified their position as a competitive force in North America. The success of the team extends beyond the players on the field. Coaches, support staff and management are equally focused on continuous improvement, seeking strategies and systems to boost the performance of the Whitecaps.
Their analytical approach to measuring player performance is one core aspect of their efforts to achieve top performance. As such, the Whitecaps have a dedicated Data Science team, generating insights from data collected every match to help the team make accurate decisions. These data scientists are responsible for creating reports that inform everything from player development and tactical strategy to scouting and recruitment.
The Challenge: Transforming Raw Data into Actionable Insights
In their search for competitive edges, the Vancouver Whitecaps’ Data Science team faced a significant challenge: efficiently converting a large array of player performance metrics into concise, actionable summaries. With over 150 individual metrics collected for each player, the manual analysis required to generate summaries for player scouting was incredibly time-consuming. Coaches and management rely on these summaries to quickly grasp a player’s strengths and weaknesses, helping them to define strategies before a match or for finding new players to expand the Whitecaps roster.
This process of translating complex player data into clear, non-technical language, however, was a major bottleneck. It requires extensive knowledge of soccer and time to deliver, hindering the team’s ability to make rapid and informed decisions before matches taking place in quick succession in the MLS. On top of having to deal with short turn-around time between matches, there are thousands of players to scout for team recruitment. These demands fall heavy on the laps of a small team of data scientists.
ProCogia’s Approach: Streamlining Player Analysis with AI
The Whitecaps partnered with ProCogia to revolutionize their player analysis process. The project was centered around using Large Language Models (LLMs) to automate the creation of player summaries. In the following points, we have listed three the aspects of the solution:
- ProCogia developed a system to convert raw numeric metrics into meaningful textual descriptions. This helps to give context about the magnitude of metric values for the LLM. The process involved standardizing each metric, then creating distribution-based binnings to categorize performance levels by assigning descriptive labels to each bucket.
For instance, a player with a standardized value of 1.7 in possession percentage will be marked as “Good” in this metric if this approach is used.
- The system flags metrics that are most relevant to a player’s specific role, emphasizing attacking metrics for strikers and defensive metrics for defenders, for example. This ensures the LLM will focus on the right metrics considering the player’s position on the field.
- To ensure accuracy and relevance, ProCogia implemented an adaptive few-shot prompting strategy. Few-shot prompting has been shown to improve the quality of LLM generations by providing cases that illustrate its task. Experts have selected a set of standard player summaries that best reflect how each metric can be interpreted to create scouting reports. Including all standard summaries in the prompt is not optimal and can lead to the LLM losing its focus. To avoid this, our solution draws only a few examples from the curated set of standard summaries that closely match the profile of the player being analyzed.
- These were all combined and presented to the LLM, which then generated concise and insightful summaries, highlighting the key aspects of the player’s performance in a clear, non-technical language for coaches and players to be able to quickly draw conclusions from them.
The solution was seamlessly integrated into the existing Whitecaps systems, reading from, and writing to their data warehouse. This ensures that data is always up to date and that the LLM-generated summaries can be easily consumed by their teams.
The Results: A Game-Changing Efficiency Boost
The implementation of the LLM-based assistant has clearly shown the potential of using AI in sports analytics. The Vancouver Whitecaps experienced a 90% reduction in the time required to create player scouting summaries. The streamlined process enabled faster decision-making in scouting and opposition analysis and allowed the Whitecaps Data Science team to focus on other strategic initiatives.
While the LLM acts as a powerful tool, it’s understood that it’s not infallible. With an average of 80% agreement with human experts, the LLM plays a role as a copilot, providing reliable foundations for analysis which allows the data science team to focus on refining and validating the insights generated.
Data-Driven Sports: What is next?
By automating the conversion of complex player data into actionable insights, the Whitecaps are now able to significantly reduce the time required to evaluate and understand player performance, ultimately contributing to faster strategic decisions.
During our partnership we noticed that the benefits of sports analytics go beyond evaluating professional players. For a modern sports organization like the Vancouver Whitecaps, data-driven insights can transform operations in numerous ways. From injury prediction and prevention to fan engagement to strengthen fan loyalty and drive revenue growth, data will continue to play a crucial role in sports.
As the development of sports analytics unravels, how will teams further leverage AI and other novel data science tools to unlock new levels of performance? Connect with ProCogia today and find out how our team can help you stay competitive.
