Latest from academia, plus Sir Martin

Highlighting some interesting academic studies relating to marketing.

IAB Tech Lab Summit: TikTok

Jeremy Bloom interviews Aashish Takkala of TikTok from the IAB Tech Summit.

Marketecture Podcast: Sir Martin Sorrell

Sorrell is always brilliant to listen to. He tells us about his mantra for AI in advertising: Wow, How, Now!

Insights from academia

I got my MBA at Columbia a lifetime ago, graduating in ‘99 just in time for the dot-com bubble burst. Now the school has a beautiful new building in Morningside Heights and regularly throws events relevant to advertising and marketing. I’ve attended and spoken several times this year and I thought I’d share some research that I found interesting.

Will AI-generated minority models turn off consumers?

Rhonda Hadi, Esther Uduehi, Andrew, Stephen, and Felipe Thomaz presented Synthetic Diversity: Racial Depictions Influence Consumer Responses to AI-Generated Models in Advertising

The team studied consumer reaction to AI-generated faces and models, with a focus on racial minorities. The hypothesis being that while consumers already cannot tell real models from AI-generated ones, brands risk a backlash when adding racial diversity without real actors.

You can buy AI-generated models for commercial purposes from companies like. botika.io. In some cases, like political ads on Facebook, you are required to disclose this fact, which can cause some backlash.

Participants in this study were shown ads with and without a notice the models were generated by AI, and then asked a series of questions about the companies and their intentions. Consumer reaction to ads with AI-generated black models were significantly more negative than those same ads with white models and panelists also showed concern about “woke washing” and a lack of authenticity.

Analyzing customer voices to infer certainty

Maximilian Gaerth, Hauke Roggenkamp, and Christian Hildebrand presented Using Voice Data to Predict and Understand Consumer Preferences. The study does not appear to be online but here’s a link to reference.

In this study they hired voice actors to record statements of consumer choice like “I want the one on the right”, in both certain and uncertain versions. They then used statistical software to train a model to understand when voices could be interpreted as certain or uncertain. The implication is that certainty among consumers could be used to upsell or change the decision processes in the purchasing funnel, and avoided among consumers with clear preferences or pre-made decisions.

Inferring personalities on Facebook

Professor Sandra Matz of Columbia published Large Language Models Can Infer Psychological Dispositions of Social Media Users.

Psychology research talks about the “Big five” personality traits of Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. These have predictive value and can be used for a bunch or purposes.

Professor Matz’s paper shows that LLMs can predict personality scores based on social media posts with as much accuracy as specially-trained tools.

In another paper Using The Big Five For Customised Advertising On Facebook, she was able to use the data in Facebook’s advertising system to predict users’ introversion or extraversion using interests like “parties” vs “reading.” Based on this targeting she then showed consumers creatives more suited to their personalities, resulting in higher click rates.

Hope you enjoyed this little change of pace.

Reading list

• Omnicom buying Interpublic (link)

• OpenAI releases Sora in the US (link)

• Lina Kahn out (link)

•  Krogers acquisition of Albertsons = blocked (link) 

• LoopMe acquires Chartboost from Zynga (link)

• Google CTV ad manager updates (link)