Chatbot delivers answers to complex questions on football data
20 August 2025 – The DFL has an extensive collection of historical and current football match and player data. By employing state-of-the-art AI technologies, accessing and understanding this constantly evolving treasure trove of data in a wide variety of applications has now become much easier. A current example: from autumn 2024 to summer 2025, the DFL and its subsidiary Sportec Solutions AG (STS) developed an Al-supported chatbot in collaboration with Amazon Web Services (AWS) that can provide answers to suitable, contextual questions from the database within a very short time.
Live queries of all official match data
Following the completion of the minimum viable product (MVP) phase in June 2025, the tool is now being made available to professional users in a phased approach during the first half of the 2025-26 season. Integrated into the DFL’s Match Analysis Hub (MAH), the chatbot is currently accessible to internal and external editors and commentators for editorial purposes, enabling live queries of all official match data from all Bundesliga and Bundesliga 2 matches since the 2023/24 season.
“Unlike generic AI tools, our chatbot relies on the official Bundesliga and Bundesliga 2 match data as the only, accurate data basis. The flexibility of Amazon Bedrock also helps us word understandable answers by selecting the right natural language models,” emphasises Luccas Roznowicz, Head of Strategic Cooperations at DFL GmbH. Josef Kaes, Senior Product Portfolio Lead at STS, adds: “Our chatbot lets users draw information from our full repository of match data – from match information, event and position data all the way to the Bundesliga match facts – in a live query. External tools cannot retrieve this data.”
Direct answers without requiring complex searches
The previous research and statistics tools for commentators and editors, which are tailored to the specific requirements of this user group, require familiarisation and knowledge of the dashboard navigation system. The new Bundesliga chatbot simplifies searches for information enormously, says Josef Kaes: “For example, if a player scores a brace, you can ask the chatbot when the same player last accomplished that feat, or when he last scored against the same opposing team – information that resides in our database but was not retrievable by users of our previous tools so easily in a matter of seconds.”
Anyone using the new tool can receive instant, to-the-point, contextualised, direct answers to specific questions.
Josef Kaes, Senior Product Portfolio Lead at Sportec Solutions
Feedback from initial user tests shows that the chatbot can make the work of field and studio reporters, commentators, analysts, broadcast managers or interviewers much easier, especially when preparing for and reporting on matches. “Anyone using the new tool can receive instant, to-the-point, contextualised, direct answers to specific questions,” explains Kaes. “The chatbot replaces dashboard navigation, which can be quite complex, with the simplicity of natural-language input and output, while also improving the direct availability of a wider range of information – including live data, as the database is updated every minute from the STS Data Hub. The tool processes German and English input, and more languages will be added in future.”
Trained by people for people
In an iterative process, the chatbot was trained on how to interpret questions and find the right answers. “We confronted the chatbot with many sample questions and answers and fed it with key football-specific knowledge. This means it knows which categories to look for when answering certain types of questions, or how to aggregate data in a meaningful way,” explains Josef Kaes. “Using these guidelines, it can even answer complex, open questions as well as follow-up questions.”
A user question is processed in three steps:
- First, the chatbot must “understand” the question, i.e. analyse the natural language and generate database queries from it, using a large language model (LLM).
- In the actual database query, the tool searches for the required information.
- Using another large language model (LLM), the search results are transformed into a human-language answer which not only delivers facts but also context.
As the market continues to evolve dynamically, the models used are updated regularly using Amazon Bedrock to achieve the best balance between accuracy and response speed. In addition, some initial tests were carried out to explore the use of Retrieval Augmented Generation (RAG) for the purpose of incorporating information from preview articles, match reports and other texts to complement the database results.
Future development potential
From information on matches, teams, players, coaches, referees or specific match events to Bundesliga Match Facts powered by AWS, the intuitive new chatbot greatly improves and accelerates the usability of comprehensive information from the world of German professional football. Work continues to further optimise the tool based on feedback from current users while incorporating data from earlier seasons. The next step will be to make the chatbot available to a wider range of users, both media partners and fans. From the outset of the project, the developers of the chatbot considered it their mission to open up the DFL’s vaults of football knowledge to many user groups.