How Sevilla FC is discovering future soccer stars with Llama

Sevilla Fútbol Club (Sevilla FC), the seven-time Europa League champions, has long been a vanguard of innovation in professional sports. Whether it’s match analysis, player performance, or fan marketing, the club’s data department has pioneered machine learning and AI-powered tools to enhance performance both on and off the field.

Despite these advancements, one challenge remained—the team needed a way to efficiently analyze and leverage unstructured data from the more than 300,000 scouting reports in its database. To solve this, Sevilla FC’s data department partnered with IBM to create Scout Advisor—a generative AI-driven scouting tool designed and built on watsonx, with Llama 3.1 70B Instruct. IBM’s watsonx and Llama enables Sevilla FC to bridge the gap between traditional human-centric and data-driven scouting in the identification and characterization of potential recruits.

“Our in-house tools excelled at identifying and characterizing players based on structured numerical and categorical data, but they fell short with unstructured data—an invaluable scouting resource that encapsulates the human expert opinions that are crucial for comprehensive player evaluations,” says Elías Zamora, Chief Data Officer at Sevilla FC.

While structured data like goal counts and passing accuracy can be quantified, the quality assessments of scouts—characteristics like attitude, tenacity, and leadership—require extensive amounts of unstructured data. Prior to Scout Advisor, these insights were buried within the club’s massive database, which required recruiters to spend 200 to 300 hours searching and analyzing a single shortlist of players.

Scout Advisor uses natural language processing, semantic analysis, and AI to provide conversational search tools, curated results, and player summaries on an intuitive and user-friendly platform. With Llama 3.1, recruiters can parse thousands of scouting reports in seconds to enhance player evaluations and summarize scouting reports on the fly.

“With Llama’s advanced natural language processing, we were able to bridge the gap between qualitative human insights and quantitative data analysis,” Zamora says. “This fusion enhances the efficiency and effectiveness of our scouting operations, ensuring that our recruitment strategies are both data-driven and deeply informed by human expertise.”

Scout Advisor was built on watsonx, IBM’s portfolio of AI products. Using Llama-generated prompt enrichment, Scout Advisor is able to comprehend soccer-industry jargon and answer questions accurately. Prompt enrichment ensures that idiomatic user questions have enough context for a thorough semantic search of Sevilla FC’s hundreds of thousands of scouting reports.

The data department uses few-shot learning with Llama on Sevilla FC’s proprietary scouting data to perform two tasks in the Scout Advisor pipeline, according to Zamora.

“We selected Llama 3.1 70B for its text enrichment and summarization performance, particularly in the Spanish language,” he says.

The application’s enriched prompt will only retrieve relevant scouting reports with a comprehensive summary of individual player performance. For example, a simple query like “show me talented wings” is automatically refined with soccer-specific context: “A talented wing takes on defenders with dribbling, creating space, and penetrating the opposition.” In contrast, a general-purpose model’s response may include irrelevant results outside of soccer, such as a chicken wing recipe, for instance.

Since implementing Scout Advisor, Sevilla FC has transformed its scouting process with:

• Reduced evaluation time: Player assessments that once took hours now take seconds with Scout Advisor’s fast, accurate, and conversational experience.
• Enhanced talent identification: AI-powered semantic search enables the club to identify players based on a large dataset of human expert opinions.
• Stronger competitive edge: With the creation of Scout Advisor, Sevilla FC has positioned itself as a global AI leader in the sports industry.
• New business opportunities: Sevilla FC’s expertise in AI has created consulting opportunities and new revenue streams.

“This is a revolutionary tool for a football director. I don’t need to review 45 reports for a player to know what my scouting department thinks of them,” says Victor Orta, Sporting Director at Sevilla FC. “In perhaps two minutes, I can get all the information that I need to make a decision.”

Zamora adds that the club’s expertise in AI has opened a new revenue stream, and they’re now consulting with other institutions in the sports industry on their AI projects.

Unlike proprietary models that require API access to third-party platforms, Llama’s open source architecture enabled Sevilla FC to customize and run AI models entirely within their secure environment.

“By eliminating the risk of data leakage, Llama helps ensure that our valuable information remains exclusively within our organization,” Zamora says.

In collaboration with IBM, the club’s data department was able to rapidly test the performance of several open source models on watsonx. Llama 3.1 ultimately scored the winning goal for the team, offering more computer performance and accuracy at a smaller cost than competitors.

With nuanced scouting information in their back pockets, Sevilla FC has set a new standard for sports analytics, giving the club an AI-enabled edge against the competition. For those inspired by Sevilla FC’s AI initiatives, Zamora says it’s crucial to have three key elements: a solid data foundation, deep business understanding, and a well-educated team. In the future, Sevilla FC plans to evaluate new versions of Llama to further refine Scout Advisor’s capabilities, ensuring it remains a leader in sports AI.