AI is AI and remains AI

In the end, the important question is not how an AI solves a problem, namely by means of deep learning, but which problems GenAI solves, especially those that could not be solved (satisfactorily) before.

Take, for example, a platform that offers targeting based on the content of a website or only wants to place an advertiser's ads in a "safe environment" (brand safety). You don't need GenAI for this, other existing processes can already solve this well and - surprise - these have also been based on deep learning and neural networks for years. There is therefore a functioning system that is highly specialised in contextual targeting and brand safety, long before the current hype surrounding GenAI. Automated production of texts, images, graphics or other output is not necessary here, nor is interaction via prompt, so the real unique selling points of GenAI simply do not apply here.

The question here is not one of disruption, but of gradual evolution. Can we achieve similar results more efficiently? Can workflows in media planning be optimised or is previous specialist knowledge about LLMs now more readily available?

Back to the field of advertising technology: a highly automated industry like ours, where AI and all conceivable types of machine learning are already widely used, will not be transformed by generative AI from one day to the next. After all, GenAI cannot do everything better that has already been done well. What it can do is obvious: communicate, appear human and conjure up creative results seemingly out of thin air. We will see many services that build on something like this or speed up things that are still very slow today. In the end, however, even these will have to be integrated into large, automated workflows. And that is always an issue and usually takes time.

In the beginning was the bot

And precisely for the above reasons, the first application that everyone looks at is a chatbot. Why? Because the old chatbots were no good, but they already existed. Everyone has probably despaired at them at some point. So it makes sense to replace these questionable interlocutors with real, high-quality interactions and help for the customer. The effort involved in creating these new bots is absolutely manageable, as "generic" language models exist, the data for training the old bots or simply the classic help documentation already exists and it is not rocket science to extend this now. That's great, and I'm grateful for these better bots several times a month, but it's not a disruption. Nevertheless, it triggers the users, in a positive sense, because there is now finally a natural language conversation with a machine that can even help in many cases.

However, it still remains difficult, especially with generic analyses such as research, perhaps even on the topic of AI. The more critical the analysis, the more important correctness and traceability, naming of sources, reliability of the bots' statements. Can I really trust the result? This is exactly the problem we humans have had for a long time with every type of AI, not just GenAI. As a result, GenAI or AI per se becomes another filter stage in many cases, but humans remain "in the loop", at least for a while longer.

AI also still has problems with repeatability and iterations, for example in the area of creation. Sure, AI technologies are good as asset generators and for creative input. And if we now think about the topic of corporate identity etc., then AI is (still) out or only plays a small, well-defined role. Human-AI collaboration has to be learned. A lot is still unclear until it works properly and efficiently.

What is ultimately needed is "augmented intelligence", an extension of human intelligence, not a replacement for humans. But we're not there yet. At the moment, AI is still often used as a text or image generator or to generate ideas. And yes, much more is possible, but that takes time and trust.

It will still take time

Nobody can tell me that they simply press a button and trust the AI when creating their own content or in areas where sensitive data is often used. It will take some time before such processes are up and running. But we are clearly moving in that direction - no question about it.

But even then, the question remains: How much better is a super granular deep learning model in campaign optimisation compared to a classic machine learning model? How much does a classic model provide? How much cheaper is it and how much more sustainable (as it requires less computing power)? How often do you want to retrain, how often are new signals available for training? Once a week or 1,000,000 times a day? The cost-benefit ratio is an important issue! And no amount of hype will help if one model yields 80 percent prediction quality and the other 80.4 percent, but at the same time costs 20 times as much, or is only available after the campaign has ended because the data situation changes too frequently. We are talking about very dynamic systems in advertising.

Generative AI is here to stay. Clearly. And the technology will sweep over certain industries. But certainly not over the advertising industry and not over the way many things are already being solved today. A lot more will happen before GenAI unfolds its full potential on our playing field.