What happens when AI takes over media planning?

Good-sounding answers are not yet good plans. This is exactly where Guido Trevisan comes in. He looks at AI not as a promise of salvation, but as a system that comes with its own blind spots. Where data volumes are impressive, knowledge does not automatically begin. And where machines radiate security, they can also conceal a distortion.

What initially sounds like a steep thesis becomes clearer on closer inspection. If AI-based media planning systematically pushes budgets towards those platforms that provide the loudest, densest and most machine-readable information about themselves, then more than just the balance of power between digital and traditional channels will be shaken. This raises the fundamental question of how advertising money can be planned responsibly, effectively and in line with the market in the future.

In this conversation, Trevisan is not sounding the alarm for the sake of it. Rather, he comes across as someone who is following a trail hidden behind plausible results. His suspicions are not aimed at the technology itself, but at its data basis, logic and blind self-evidence. This opens up a larger topic: it's not just about print, TV, audio or DOOH. It's about the architecture of media planning in the age of AI - and the question of whether the industry is currently encountering a new form of bias that sounds too slick to make people immediately suspicious.

Guido Trevisan, you talk about «AI bias» in media planning. Are we talking about a real market problem here - or a rather steep thesis with political benefits for traditional media?
If the thesis is correct, it is not just print, TV or OOH that has a problem, but the entire advertising industry. If you take the thesis further and follow what the current AIs recommend, ineffective media planning would also weaken the advertising impact as a whole. As a result, investments in commercial communication would be reduced and companies would invest elsewhere.

In your view, what exactly is the bias: that AI plans incorrectly - or that it ruthlessly reveals which channels simply dominate in terms of data today?
The bias clearly lies in the incorrect planning of the AI. Assuming that media agency employees and marketers know something about their work, it is very surprising that the planning recommended by AI - especially in the USA, where this bias has already been proven - differs greatly from human media planning. The fact that digital media «produce» more data, even if only partially measured independently, cannot be denied.

You say that large language models (LLMs) and AI systems favor digital platforms. Where does the bias begin here - and where does simple data-driven logic end?
When it comes to media planning, even more reference should be made to existing independent data. Properly prompted, LLMs are a fantastic tool. However, the same applies here: «garbage in, garbage out».

In the end, isn't AI bias simply the new, algorithmic version of the human bias that the industry discussed years ago?
Both are dangerous when it comes to using advertising money in trust. However, I believe that the structural damage potential is much greater with the algorithmic version, because you get a good-sounding answer to almost every question at the touch of a button.

Hand on heart: How much confidence do you have that a machine can actually make media plans worse today than a person with experience, gut feeling and Excel?
As already mentioned, correctly prompted, the machine can certainly produce satisfactory plans. Today, however, most results are inadequate. If an agency briefing is created with an AI and then forwarded directly to the agency one-to-one without experience, gut feeling and Excel - i.e. without any knowledge of its own - the result is rarely any good. I have been told that this is already happening with well-known clients. And then there is the problem that strategic media planning is taught less in further training courses. This is where we as an industry need to start, so that the «human in the loop» generates important added value.

Your argument is based heavily on observations from the USA. What is robust enough about this to shake up Switzerland right now?
It is possible that the deviations from booking behavior in Switzerland will be even greater, as traditional media are more relevant in this country. The data from the USA has at least made it possible to engage in discussions with renowned organizations such as Leading Swiss Agencies and the Swiss Advertisers Association. Our common goal is to implement a similar analysis for Switzerland.

Couldn't your thesis also be read like this? The tech platforms have done their homework when it comes to data - and traditional media have put it off for too long?
Sure. But in my opinion, it's irrelevant. And although the tech platforms have some of the smartest minds, I don't believe that this bias in media planning has been planned this way for a long time. We need to act now and, with relatively little effort, we can provide the Swiss advertising market with a tool to plan reach and impact more effectively based on existing Swiss studies.

If AI recommends traditional media too little: Is this really a bias of the machine - or a visibility problem for traditional providers?
One idea was to simply publish more Swiss data and thereby correct the outputs. The LLMs are based on global information. Our realization: Of course we can and should publish more, and the service providers downstream in the value chains should also make greater use of this. However, believing that we can significantly influence the system with a little additional Swiss information has proven to be a misconception. Quite apart from the fact that the research organizations' data has value and they want to continue to capitalize on it.

The classic genres in particular often insist on their impact, but struggle with comparable, open and machine-readable data. So is the AI bias also homemade?
I have a different view of things. It is the classic genres that can be measured independently wherever possible and financially viable. It's not just in Switzerland that the multinational platforms find this difficult, even though they would have the money and the technical possibilities. Various industry organizations have already tried to involve the platforms in various projects. However, the majority of them have rejected comprehensive cooperation.

You warn against unjustified budget shifts. But who says that these shifts are unfounded as long as the digital side simply provides more reliable signals?
The buddy who chats the most at the regulars' table is not always right. The amount of data alone has no value. The point is to clarify whether it serves the purpose, namely the most effective media planning possible.

In your view, how problematic is it that the large platforms of all things measure their own impact, display it themselves and are also the loudest source of data for many AI systems?
Agency employees and marketers must be aware that the player and referee are the same person in a figurative sense when they are planning the advertising money. Whenever possible, they should consult additional, independent data as a basis for their decisions.

When would your thesis be proven for you? What would a Swiss investigation have to show in concrete terms for the suspicion to become a reliable finding?
In my opinion, it depends on the overall picture. In the US, the model has recommended investing 7 percent in traditional linear TV and 18 percent in CTV. Although the market is of course very different, I would interpret this as a strong distortion of reality in Switzerland. The results must be interpreted by experts. I don't think it makes sense to define a fixed value in advance.

Your approach is called «Swiss Media Agent». Is this a neutral corrective - or ultimately an attempt to give traditional media more weight in the engine room again?
Even if we were to put all the Swiss data available to us online, we would not be able to trim the LLMs to Switzerland. The concept of the «Swiss Media Agent» is that we collect the media data that is freely available today - IGEM Digimonitor management summary, Media Focus advertising compass, WEMF circulation bulletin, data from the Statistical Office, etc. - in one data pool. Anyone using a standard LLM to create a briefing for a campaign or campaign planning can simply integrate the «Swiss Media Agent» into their query in order to enrich the global information with specific Swiss know-how. We create transparency by communicating which data has been taken into account. The data sources are to be defined by a committee - for example from academia, advertising clients, media agencies, marketers and publishers.

And perhaps the decisive question: Is your ultimate aim to make AI better - or to prevent traditional media from losing relevance in the age of AI?
Today, the results of AI simply sound good. We want to refine them with our Swiss data. The efficiency of AI is thus combined with in-depth market knowledge. In this way, we create the basis for campaigns to be played out even more precisely with critical classification and responsible budget management.

What concrete next steps are you planning?
On the one hand, the AI bias in media planning for Switzerland is to be demonstrated. To this end, we are in contact with data scientist and software architecture specialists. Secondly, we would like to bring together interested organizations and individuals who are willing to work on a joint solution or further development of the «Swiss Media Agent» with their ideas, expertise and financial resources.