Available right after the final whistle: AI-generated match reports

27 January 2025 – Until recently, the match reports which appear in the Bundesliga app and on the Bundesliga website right after the final whistle of every Bundesliga and Bundesliga 2 match were composed in an entirely manual process: “During the match, our editors not only had to write the live ticker, which is updated roughly every three minutes, but also, in parallel, a match report,” says Karol Herrmann, Senior Channel Manager Web at DFL Digital Sports, describing the routine. “The final match report had to be published as soon as possible after the final whistle, which was an enormous challenge. The DFL has high quality standards in terms of form and content of these reports.”

As the DFL continues to expand the scope of its content services, the editorial team’s workload has increased steadily, says Herrmann. “With more detail tasks added to the to-do lists of our editors, such as storytelling, choosing and commenting a ‘Player of the Match’, and integrating images and statistics, it became clear that all this was no longer humanly possible.” In autumn 2023, DFL subsidiary DFL Digital Sports therefore decided to explore ways to employ artificial intelligence (AI) for certain routine tasks to free up the human capabilities for quality-enhancing, value-adding activities.

The human touch is key

Dirk Switalla, Senior Product Owner at DFL Digital Sports, describes the solution development process: “Our initial tests with existing AI solutions were somewhat disappointing. Then we realised that right here at the DFL, we already had an extremely valuable text resource we could use as a basis – our own match information.” Using text elements selected from the live ticker (see screenshot below) by the editor to build the backbone of the match report turned out to be the key to success. The result can then be enhanced by incorporating some of the official match data and photos generated by the DFL subsidiary Sportec Solutions (STS).

Armed with these assets, the DFL Digital Sports team began experimenting. “Our goal was to preserve as much of the human touch as possible in the final report,” says Switalla. The first step was to define a data model that instructs the AI solution what kind of information to use in the report.

Surprisingly quick results

After just a few weeks of development work, an initial, usable version of the AI solution was made available within the content management system (CMS). “We had decided never to release a text that has not been checked by a human editor,” stresses Herrmann. This means that the automatically generated match report is a draft version only that is not intended for instant publication. As a general rule, the editing team always has the last word on the content of any report.

Our goal was to preserve as much of the human touch as possible in the final report.

Over the course of several months, the editing team and the developers jointly refined and optimised the underlying model. The resulting solution has been in constant use since early November 2024 and has been proving its worth ever since. “Humans and technology work hand-in-hand in this approach,” explains Switalla.

A full development success

The new process has been extremely useful and effective. The AI-generated reports are directly fed into the CMS where they can be edited immediately. Herrmann comments: “The AI texts retelling the match are now so good they require little correction. Editing takes no more than five to seven minutes, then the article can be uploaded to the website and the app. The traditional approach would take us three quarters of an hour.” Images as well as hyperlinks to club and player webpages are inserted automatically. The editors have the option to give specific instructions to the algorithm regarding the main focus and headline of the report. Even the comment on the Player of the Match is generated by the AI solution. 

English translation included

As yet another benefit, all match reports are immediately available in English, as well, even for matches with no English live commentator present. All this reduces the editors’ workload while providing more opportunities for them to create high-value content. No staff members have to go – rather, the existing personnel resources can be reassigned to higher-value tasks. “The match reports continue to rely on human observation and assessments, but a very labour-intensive portion of the work is now being performed by the AI solution,” summarises Switalla. “The human-generated live ticker continues to provide the basis of our high-quality match reports,” adds Herrmann.

Whenever the need arises, the instructions for the model are optimised further to fine-tune the results. But these are merely subtle adjustments; the AI solution as such has already become an indispensable tool for generating match reports.