Over the past few years, ProSiebenSat.1 has been working intensively on how AI can be integrated into day-to-day business processes and help us achieve strategic milestones. The result: Automated solutions take over individual, clearly defined tasks and thus make the work of our employees easier. The primary goal is to give our colleagues more space and time for creative work and to enable responsible interaction between people and modern technologies. At the same time, the various solutions contribute to the Group's overall goals: We work every day to increase our reach through high-quality content. We want to market and monetize this in the best possible way.
In this context, the various application fields of AI are a prime example of how our tech teams work together with the various departments: Our AI experts from ProSiebenSat.1 Tech Solutions develop the solutions together with colleagues from the specialist departments, all of whom produce, distribute or monetize content in a wide variety of ways. To this end, the AI Products team has developed five of its own automation products, whose areas of application range from sales to automated video, image, and sound solutions. They all exemplify how we are currently using AI at ProSiebenSat.1.
1. Digital Buzzroom:
more output in less time
The Digital Buzzroom of Seven.One Entertainment Group is a digital video editorial department that produces daily news clips. There, we use AI to produce fully automated videos. Our goal: to increase output with the least possible human intervention.
When the COVID-19 pandemic dominated the news and the incidence levels of the federal states determined our daily lives in Germany, we used a data-based tool at ProSiebenSat.1 that allowed us to create an individual, audio-visual Corona update for each federal state. The starting point was a generally valid text created by an editor. This summarized the daily updated data from the Robert Koch Institute, which an algorithm adjusted every day. As soon as new data was available, the algorithm automatically overwrote it – thus generating 16 clips on infection events fully automatically every day. The automation algorithm can therefore adapt clips individually and also produce them independently.
We automatically play out the clips produced in this way on numerous publisher sites, such as the online offerings of major newspaper publishers. This allows ProSiebenSat.1 to produce a large number of clips and increase the reach of the videos. At the same time, our editors can use and organize their working time more efficiently.
2. Automated tagging:
keywording made easy
As a digital media company, ProSiebenSat.1 produces a huge amount of multimedia content every day – including countless hours of moving images and video footage. Finding this content in our media documentation center is a key task, for example, to access archive material. In the media documentation center, colleagues describe the contributions of our editorial teams and document what they have seen in written form. Up to now, our media documentarians have manually indexed the broadcast material on a daily basis: What can be seen? What happens in the clip and which information of a scene is relevant enough to be reused? The manual documentation of a one-hour TV report can quickly take several hours. Together, the media documentarians and their colleagues from the AI Products team have therefore developed a solution that supports the manual archiving process with the help of AI.
An automated process generates keywords – so-called tags – that we use to document our content. The keyword here is video mining – the automated extraction of information from video content. For this purpose, the AI Products team has connected the specially developed and customized ICON platform ("Intelligent Content") to the media documentation system. This platform bundles various services, such as object, celebrity and voice recognition.
To ensure minimum tag quality and relevance, the team also established an intelligent, automated post-editing process. It serves as an additional correction loop and filters out irrelevant or incorrect tags. In this way, we ensure that the automated tagging results can be successfully used and later automatically integrated into the media documentation database. The automated process thus saves the editors a lot of time.
3. Transcript integration:
turning speech into writing
In addition to keywording content, transcribing can also help to quickly find statements and topics. Creating transcripts is one way to make spoken word more easily searchable by machine. They help to quickly identify and further process statements on specific topics in existing raw material. This makes it possible to find sequences again, especially in editing – and facilitates coordination between cutter and editor.
Here, too, AI simplifies the process at ProSiebenSat.1: Until now, our editors have spent a lot of time manually creating transcripts. As part of a pilot project with the editorial team of "Abenteuer Leben am Sonntag" (The Adventure Called Life on Sunday), the AI Products team developed a solution that editors can now use to "order" transcripts of raw material at the touch of a button.
This automated transcription enables cutters and editors to view the material quickly and in its entirety, and to coordinate their work easily. The duration of the transcription corresponds approximately to that of the inserted video and runs completely in the background. For comparison: to transcribe one hour of spoken word, three to six hours are needed.
finding the right target group to address
Targeting with AI makes it possible to play advertising to exactly the right people. AI allows to precisely identify, segment and address users, both online and on TV. In this way, our advertising customers can reach the right target group even better, which further increases the attractiveness of our advertising spaces.
The AI Products team is working with the sales experts at Seven.One Media on precisely develop such targeting solutions. The most important prerequisite here: the target group must be determined in advance of the advertising campaigns and be reachable for the advertising playout itself. This is where our AI products help: With our profile-based targeting, we will be able to address users in Addressable TV much better in the future. The underlying technology enables us to target TV advertising on Internet-enabled devices in real time and thus reach target groups even more precisely with personalized spots. To classify viewers into user groups, a machine learning model analyzes the TV behavior of known target groups. At the same time, we can apply the model to unknown users and thus predict which existing target group they belong to. The resulting user data allows us to assign the different TV devices with a high degree of accuracy to address viewers with suitable advertising campaigns. In this way, we reduce scattering losses and use our advertising customers' media budgets all the more efficiently.
Measuring the impact of advertising
Watch, click and maybe even shop: That's how TV advertising can work. TV campaigns anchor brands in the minds of consumers for the long-term. But they also have a direct effect on the number of visitors to websites and online stores. To measure this effect, the tool MediaLytics was developed by AI Products team in cooperation with the Performance Marketing & Attribution unit.
MediaLytics can be used to measure short-term website views triggered by TV. The traffic on the advertisers' websites is first collected automatically on a continuous basis. In this way, it is possible to document how many users usually visit the website. If the traffic on the website then increases after the TV spot has been broadcast, media planners and customers can attribute this to the TV spot. An algorithm links the reach of TV commercials, i.e. reach data from the TV world, with the individual traffic data of the advertised app or website, i.e. performance metrics from the digital world.
In this way, customers with digital business receive a transparent evaluation of how effective their advertising placement was. It also allows us to optimize schedules so that our marketing experts can better evaluate campaigns and plan future advertising times.
Five different functions, five different areas of application, five different effects – but what all the examples have in common is the collective benefit for us as a company. Because they all help ProSiebenSat.1 to increase our reach every day with high-quality content and to drive the monetization of our content via our sales units. Last but not least, they make the day-to-day work of ProSiebenSat.1 employees easier and thus create more room for creativity.