Gen AI and advertising

CreativeAds created by Gen AI already exist. Meta’s Gufeng Zhou highlighted a recent campaign for Coca-Cola with a storyboard and visuals created by AI. However, the key question when it comes to generating a core creative for a campaign is whether Gen AI can take a genuinely creative step. Some early indications from creative testing with consumers are that ads generated by AI tend to perform at around the average of all ads.It makes sense that Gen AI trained on absorbing all ads is producing something that reflects an average, but do advertisers want to invest in a technology that, to use a statistical term, ‘regresses to the mean’? Can it actually move beyond the average? Creativity is essential and there is a danger that advertising may not only regress to the mean but also reflect and amplify current societal biases: surely advertisers will want more than that?The point was also made that there is a difference between performance ads that are designed to promote activation and those that are aimed at brand building and values. It may be that Gen AI has a greater role to play in the former, particularly when it comes to personalization and activation.

Personalization and adaptation

Where Gen AI may have the most initial impact is in adapting advertising campaigns to make them more targeted and relevant for specific groups. Rather than taking the initial creative step, Gen AI can create adapted versions of a campaign to aid hyper-targeting. This concept already is used in radio advertising for example, where national campaigns can be adapted to reference local names or even the current weather in that region.

So, Gen AI may play a powerful role in executing content production based on an original creative, with AI working to create variants for imagery, taglines, formats, colors and other relevant factors. Those may well be analyzed and isolated by AI analysis of past campaigns and other inputs. As we discuss later, AI can also be helpful in analyzing the effectiveness of those various versions after the campaign has run.

Concern was expressed about the degree to which, if taken too far, such adaptation could actually undermine the identity of the brand itself. If Gen AI allows a brand to become effectively a chameleon, with the ability to be ‘all things to all people’, to be both transgressive to teenagers and traditional and reassuring to the elderly then is there any brand left as a result?

“If each person continually sees some different image of the brand based on their interest and needs, and if each person has one personal idea of my brand and if we continuously run this type of precision marketing activity that artificial intelligence is pushing us towards, the question is: in the long term, how does the brand have a unique proposition? How about the brand, the future of the brand? What about the brand equity?”

Marco Robbiati, OMG

There is definitely a fear of a loss of control, that the strategy or communications objective will be lost if Gen AI becomes more than just an enabler.

“Our clients are saying we're trying to use Gen AI to drive personalization and align different messaging to different segments that drive a differential outcome. Well, how do you do that in a way that still maintains brand integrity? And I don't think a lot of people think through that issue. super well, right? We want to govern content production around the brand guidelines and use Generative AI in a way to manage personalization, but personalization with the sensitivity to the brand.”

MacDonald, McKinsey

Ad testing and campaign evaluation

As with the role of Gen AI in program generation, vital existential issues are being discussed in the advertising industry when it comes to creating and adapting advertising. However, it’s clear that Gen AI is already playing a role in the evaluation of advertising creative and the success of campaigns.

Isolating the impact of elements of content has arguably been harder to evaluate until now, compared to more quantitative metrics like placement and frequency. AI-driven object recognition techniques provide an opportunity to better quantify the impact of a campaign’s creative elements.

“What was definitely undervalued so far is the creative itself, the content itself. It was very difficult to take creativity into account in ads effectiveness measurement. With the new AI capabilities, especially image & object recognition, it's becoming possible to predict ads response by decomposed creative elements in a scalable way.”

Gufeng Zhou, Meta

As a result, Gen AI could greatly enhance media measurement and marketing mix modeling, providing analysis to assist the optimization of campaigns and media planning with better measurement. We may be able to cast greater light on the relative contributions of placement and creative as AI-powered creative testing can facilitate variation of imagery and video content that would have been cost-prohibitive in the past. Significant innovation is anticipated in the field of media and advertising measurement. Gen AI also has the potential to democratize media buying, widening opportunities to new customers who are less familiar with how to buy media by allowing them to analyze creatives, give advice and predict campaign outcomes. As we saw earlier, some feel that Gen AI may have the most impact on smaller producers of content, allowing them to do things that they previously could not afford or have the relevant skills to do. In the same way, it may remove barriers to entry to buying advertising, opening the floodgates to allow new users to create their own audio or video campaigns in just a few clicks.

Is AI about regressing to the mean?

However, the potential limitations of AI do need to be borne in mind. Once again, it's the issue that Gen AI can only be as good as the data it has been trained on. For example, Dutch company Ster - responsible for advertising on NPO - has developed a tool called AdScan. Ster has accumulated a very large data set of respondent-based research on advertising creatives. This data has been used to train models that can give advice. The system is highly scalable, with answers available within minutes as opposed to putting a project into the field. The tool is proving useful, but one of its acknowledged limitations is that it can tend to lead to recommendations that could be seen as stereotypical or obvious.

“It gravitates towards the average of proven concepts, so it's giving advice everyone can think of, like ‘put animals into your commercial, put children into your commercial, use cheerful colors.’ We've been thinking about using that evaluation to maybe use Generative AI to update those creative flows commercials, but we’d risk seeing a lot of commercials starting to look the same. Because if commercials that did well in the past have a dog or a cat in it, the model will naturally reproduce those commercials as best it can and start putting dogs and cats in every commercial.”

Rick Hoving, Ster

What would a Gen AI creative appraisal have made of some of the most popular past campaigns that have caught the imagination precisely by being different from the norm? Can AI be intuitive, or even counter-intuitive? Concern was raised that research that doesn’t involve human appraisal may not reflect that there is not just one common idea of what good looks like. Success is not always rational.

“The idea that AI does a perfectly average job? Yeah, that really worries me. Something can be excellent because it breaks the mold - because it does something that's never been seen before. But then we write it off because the automated script gives it a five out of 10 when, if we had the time and resources to have people rank it, it might have got a nine out of 10 from the audience it was intended for.”

Scott Thompson, Publicis

Is Gen AI sexist?

A key element of Gen AI is that it is trained on available data. As we have just discussed, this means that it is effectively trained on the existing biases and stereotypes in society. This isn’t AI’s fault, but it does mean that it is reflecting society back on itself: if you ask Gen AI for a picture of a doctor, you get a man. If you ask it for a nurse, you get a woman. There is an implicit danger that unless this is addressed, Gen AI may create one massive societal filter bubble that reinforces negative as well as positive traits in society.

"One of the things we were noticing is that AI can be heavily biased against women, unfortunately. We have been seeing that it actually advised customers to not use women in their commercials. Reason is, it picked up a correlation between women and the boring household product commercials they were often featured in and those just aren't very fun to watch. It's like one of these things we have to be very attentive to, just not to reproduce those biases we have in society because they get picked up very fast."

Rick Hoving, Ster

There was some debate about whether the use of Gen AI should be limited to the performance of tasks that can be more quantitatively defined, as a tool for a creative as opposed to being used to take that creative step itself. This assumes that AI is indeed the sum of what has come before as opposed to being capable of making true creative steps. However, this assumption was contested by some as being too generalized. It may well be that in the future AI will move beyond competence to become capable of excellence and innovation. It is in our nature to overestimate the short-term consequences of new technology and underestimate the long-term impact.