On June 6th 2024, Media Distillery were joined by our customer Swisscom and media analysts Omdia for a fascinating webinar about advertising in pay TV, moderated by The Media Leader’s Omar Oakes.
In this extract from the webinar transcript, Philipp Wuethrich from Swisscom explains how Switzerland’s leading telco has implemented Ad Break Distillery from Media Distillery to power their premium ad-skipping option, Replay Comfort.
You can find a recording of the full webinar here, and at the bottom of this page, you can access a more complete transcript, both of which include detailed insights from Maria Rua-Aguete, Senior Research Director, Media and Entertainment at Omdia.
Speakers are identified throughout this transcript using their initials:
OO: Phillip, I understand that regulation has forced you [Swisscom] into taking action. Explain to the audience what changed, what did you have to do?
PW: The TV market in Europe and the US and especially Switzerland and the US is indeed quite different. And I have to start with a brief overview of the industry agreement GT12, which is known as common tariff and is basically a regulation implemented in Switzerland to address copyright and advertising issues related to time shift television viewing. It was jointly negotiated between broadcasters, distributors and collecting societies who represent the interests of rights holders. The agreement took effect in 2022. It ensures broadcasters receive compensation for financial losses incurred from viewers skipping ads during replay TV. Moreover, GT12 allows distributors to offer seven days of replay, providing extended access to time shifted content.
So, with GT12, new forms of dynamic advertising have been introduced, which include short advertisements of approximately seven seconds shown at the beginning of a replay program. Ads shown when a commercial break is skipped during time shifted viewing. And last but not least, static ads displayed when the pause button is pressed during live or time shifted viewing. Most major TV providers in Switzerland are part of this agreement. They now show non-skippable ads during replay TV, replacing the original content’s ad breaks to comply with GT12. The agreement includes a special premium tariff for customers who wish to fast forward or skip commercials entirely. Swisscom offers a replay comfort option that utilizes the premium tariff of GT12, enabling customers to skip commercial breaks on replay TV with pinpoint accuracy at the touch of a button.
So, why do we partner with Media Distillery then? Only 26 of our numerous channels participate in the industry agreement. It was a challenge to relay to our customers that the Replay Comfort feature would only impact specific channels. Consequently, our goal was to provide an ad-free experience on our most viewed channels. Originally, we developed an in-house solution for commercial detection, which yielded fairly good results on a few channels. However, we soon recognized the complexities in maintaining an AI model with adequate generalization capacity for all designated channels. Recognizing these challenges, we partnered with Media Distillery to focus our internal resources in other areas. The rollout took longer than expected due to the complex technological requirements and need for precise tuning. Despite this, the collaboration has been a success.
We have seen a significantly increased net promoter score among customers using the replay comfort option, providing an ad-free experience on our most viewed channels. Additionally, we observed similar scrubbing behavior as seen with GT12 channels, where we receive accurate ad markers from broadcasters. This consistency highlights the effectiveness of our collaboration with Media Distillery, enabling us to deliver a seamless and enhanced viewing experience.
OO: Phillip, why not, just build a solution yourself?
PW: It takes quite a few resources to build up an AI model and to maintain it and to make it generalized for all different channels out there. And we really wanted to focus our resources on other topics where we can generate even more revenue or customer satisfaction and leave this task up to the specialists, in this case Media Distillery.
OO: And before we bring in Martin to talk about that in more detail, what advice would you have for other Pay TV operators? You were in this very particular situation where you've had to do something, but as Maria began by speaking about, Pay TV operators all over the world are thinking about how to improve user experience whilst being mindful of trying to further monetize the operations. What advice would you have for them in trying to embark on a project like this?
PW: I mean it's not just that, it's also like Maria stated to prevent churn, right? We have to differentiate ourselves from the competition and that's one way of doing it. And maybe as a piece of advice from, since I am an engineering manager, so from a technological point of view, it certainly helped to start off developing the feature ourselves because it gave us some good insights, we knew the problem very well and I think it helped us in collaborating with Media Distillery.
OO: Martin from Media Distillery, explain generally what this solution does?
MP: As already said by Maria and also by Philip, it's about making sure your viewers have a really compelling user experience because they have so many services to choose from, you do not want to give them a reason to churn, which has already been a big theme. The product we used to help Swisscom deploy Replay Comfort is called Ad Break Distillery, and basically what we try to do is identify where in the video, where in a broadcast do the ad breaks appear and end. We do this by analyzing the video signal in real-time and then have trained the system to identify when the ad break would start or when an ad break would stop and give that information as metadata to our clients, which then can use it for several use cases. One of them was already mentioned by Philip, it's about enabling ad skipping. So, as a premium option to some viewers, this will not be for all viewers.
Another option would be to improve trick play restrictions. Some of you may know that sometimes broadcasters will not allow you to search forward or seek in the content because they do not know where the ads are and they are forced to show the ads. And we enable them to make sure that people can interact with the content in the other places, which makes for a way better compelling user experience for catch-up replay.
But we're also looking at identifying the ad breaks in order to support ad replacements. This is something we're working towards currently, and I think that fits in well with the developments in a world where, as Maria already told us, ads are an important part of video. Video services would not exist without advertising, but it's also making sure that you offer a compelling viewing experience to your audience where the audience also changes. There's not one single viewer where one product meets all needs.
OO: Okay, and what issues did you run into when doing this? Is this a completely bespoke solution, or are people watching this as something off the shelf? And what issues do you have in creating it?
MP: We build AI-based solutions and what people know with AI is there's always some uncertainty. I think the last few years were really great for people experiencing ChatGPT, that it works quite well, but not always is perfect. And it's also part of what we saw with developing this solution. Philip already elaborated on it a bit. We tried to really train an AI system to identify what is content, or what is part of the programming, and what is part of the ad break. So, we're not trying to identify if this is the McDonald's ad, or this is a beer commercial. We're really trying to teach a system, can you tell me what is content, what's part of programming, and what's part of the ad break? And we saw that initially it worked quite well on a few channels, but we saw that it was becoming difficult to generalize it to all the different channels that, for instance, Swisscom has, but also some of our other customers have.
But by giving enough training materials, I think we trained on about 30,000 programs, which amounts to 8,000 hours of annotated label data – so, we really had people annotating the videos, what is ad break, what is content - we were able to train the models and make sure that on channels we hadn’t seen before, the model started to perform. So, it was initially, there's a startup phase where it was really challenging to make sure that the model performed well, but now once we got that on the road, for us, we now enable it on a channel that we haven't seen before, with ads we don't know, and it still is able to detect them. And then we still might need to do some tuning, but it's just minimal. And this is really great because it allows us to scale to all types of channels.
June 26, 2024
