Media agencies move beyond buzz: Practical AI takes center stage at Media Buying Summit

As a Digiday+ member, you were able to access this article early through the Digiday+ Story Preview email. See other exclusives or manage your account.This article was provided as an exclusive preview for Digiday+ members, who were able to access it early. Check out the other features included with Digiday+ to help you stay ahead

When developing generative AI strategies, media agencies are thinking not just about content generation, but also about what the technology means for overall data strategies. 

At the Digiday Media Buying Summit in Nashville this week, media agencies addressed how they’re using AI in practical ways like optimizing existing media. Speaking onstage at the summit, iProspect chief growth officer Amanda Moore mentioned the Dentsu-owned agency is using publicly available information like weather data. That way, messages are more personalized without using personally identifiable information.

“We really see this year as sort of being more of practical AI coming on board vs. just generative AI,” Moore said. “It can generate things, but there’s a lot of bias within there and the ability to make some mistakes…This year we’re going to start to see very specific instances of practicality around really testing and learning into what it means to actually act on the data.”

One of the first things IPG did was develop a platform for the agency’s 13,000 employees to have access to a custom version of ChatGPT, which also allows them to understand generative AI even before guiding clients. 

Getting a brand’s documentation in order is a key step in creating an AI strategy, according to Graham Wilkinson, chief innovation officer at IPG-owned Kinesso. Ahead of thinking about AI’s outputs, he said a lot of brands need to make sure their brand strategy is consistent and data architecture is in place. To create the right framework from developing an AI strategy, Wilkinson said it’s important companies understand their tone of voice, brand hierarchy, the target audience and brand safety standards.

“It’s kind of like the 23andme of a brand,” he said. “How do you take the blood sample and be able to describe exactly what your brand is? Once you’ve done that, then you can start to think about the technical aspects of emulating it.”

Media agencies are also testing how generative AI can help them understand their audiences. For example, one built a tool that ingests first-party data to build audiences, which then allows them to ask the synthetic audience various questions. However, others say developing their own in-house AI tools in-house has been resource-intensive, with one agency taking three months and dozens of engineers to build a their own privacy-compliant AI environment.

Other agency executives say generative AI applications for audio ads could also lead content to feel less authentic. Maria Tullin, Horizon Media’s svp, managing director for performance audio & podcast strategy, said AI-generated audio in ads might work for generic narration but not for host reads resembling real people. She also noted there should be different pricing tiers in place for human voices and AI-generated voices.

“I hate the idea of hosts’ voices being AI because the whole point of podcasting is authenticity and taking advantage of the audience that loves this host,” Tullin said. “I don’t like the door that it opens of people endorsing things without really meaning it.”

Avoiding AI-generated audio isn’t something everyone is doing. Some say they work. According to a recent study by Veritonic, generic AI audio ads for Intel led to brand favorability increasing 10% while personalized AI audio ads led to as high as 22 percentage points. According to Veritonic, personalizing audio ads with AI also led to higher purchase intent. The ads were developed in collaboration with Instreamatic and bought through a campaign delivered by Dentsu. 

Agencies are also still aware of AI’s problems with diversity and representation. Alvin Glay, chief strategy officer at Response Media, has been experimenting with various AI photo apps and said they’re still “grossly off” when it comes to accurately representing people. Glay, who is Black, recalled uploading a photo of himself into one of the apps and having the AI photo show him “looking like an Indian guy.” When he tried another app, his AI-generated image made him look like he was Polynesian.

“There’s confusion in the data,” Glay said. “So from a representation standpoint, it’s very important to have the people who are coding understand the data that they’re capturing.”

Generative AI won’t be a way for marketers to use as a “cheat code for culture,” said Albert Thompson, managing director of digital innovation at Walton Isaacson. Like Glay, he also noted the inadequacies and inaccuracies of multicultural data sets used for AI models.

There’s also still a need to improve algorithmic bias with programmatic advertising, said Sherine Patrick media strategy lead advisor at Ops Shop. She gave the example of programmatic advertising block lists including the word “bomb” that can also be used culturally as a positive word. 

“The truth is, culturally we use the word to represent something good,” she said. “Like my bomb hair. You know? So if that’s on a page, that publisher automatically gets no funding, because they’re blocked.”

https://digiday.com/?p=537103

More in Media Buying

‘More distrust in the marketplace’: Agency execs press pause on Forbes spend after domain spoofing report

Forbes’s subdomain scandal has the buy-side asking a lot of questions when it comes to programmatic reporting.

Puzzles pieces forming a dollar sign

CG Life doubles in size by buying a digital shop to expand its health and sciences targeting

Integrated agency CG Life acquired digital agency Toolhouse to expand its technology and omnichannel capabilities in the health care and biopharma space.

Interpublic Group duo sued in a case alleging ‘misappropriation’ of ID data

Identity vendor Adstra claims Kinesso broke data-licensing terms to help IPG-stablemate Acxiom compete unfairly.