If 2023 was the year marketers experimented with generative AI for ad creative, 2024 is the year most brands will really dive in.
Today, marketing leaders are either implementing (29%) or piloting (7%) generative AI in their organizations, according to Nicole Greene, a VP analyst at Gartner.
In the next six to 12 months, however, 88% of marketing leaders will either have already implemented generative AI, piloted the technology or have an active plan in place to test related use cases.
And as marketers grow increasingly comfortable with using generative AI in their daily workflows, “we’ll see some brands that push the boundaries of what [it] can do,” said Jenny Kelly, head of content at digital creative consultancy Deloitte Digital.
Inventing tomorrow
When Deloitte Digital works with brands to introduce generative AI tools into their creative process, it starts by looking at their existing tech stack. For marketing organizations that already have Adobe Creative Suite, for instance, Firefly is an easy add.
An organization that uses Teams and Microsoft Office, meanwhile, might rely on an OpenAI-based set of tools, whereas a company wedded to Gmail and Google Drive is more likely to latch onto Google Bard and Gemini.
Eventually, generative AI text and image tools will become “embedded” in the day-to-day functions of marketing teams, Kelly said.
Presumably, generative AI audio and video tools will someday become omnipresent as well, though in the near term, AI video generation “isn’t quite there,” said Cristina Lawrence, EVP of consumer and content experience at Publicis-owned digital agency Razorfish.
But as video rendering and creation capabilities improve, marketers will start using generative AI to create both short- and long-form video, said Nikhil Lai, a senior analyst at Forrester.
In the meantime, we’ll likely be treated to gimmicky applications of generative AI technology.
An October Forrester report predicts that some advertisers will try a generative AI “stunt” during events like the Super Bowl and the 2024 Summer Olympics – and that at least one brand will issue a public apology “for not fully thinking it through.”
Creative overload
Gimmicks aside, brands in 2023 were primarily excited – “exuberant,” even – about using generative AI to exponentially increase the amount of ad creative they can produce, Forrester’s Lai said.
But although there is value in being able to do multivariate testing on hundreds of assets with slightly different variables and personalize ads and website copy to a far greater degree, marketers can also have too much of a good thing.
For instance, Lai pointed to a data company called AdCreative.ai that promises “millions of new ad creatives every day.” Brands don’t need that volume, however, and from a data engineering perspective, they often can’t cope with so much creative performance data.
Take the CPG brand Lai has been working with recently on a creative optimization project. The brand is “swimming in data” about the performance of its ad creative, Lai said, but “they’re not sure if the tests are conclusive.”
“Audiences are fickle,” Lai said. “What they responded to yesterday may not signal what they’re going to respond to five months from now.”
Let’s not be too hasty
Changeable as audiences are, they generally don’t like feeling lied to. And generative AI tools, with their tendency to confabulate as confidently as Baron Munchausen, pose a risk to brands. Which is why marketers will prioritize generative AI governance in 2024.
Generative AI legislation has made footfall in Europe, where the EU ironed out the terms of the AI Act in December, and in China, which released interim generative AI rules in July.
But in the absence of US regulation, both AI vendors and the companies that use the technologies will have to govern themselves. And since lawmaking is often a ponderous process – particularly in a US election year – marketers may wait years to find out whether they made the right calls.
The IAB Tech Lab’s AI subcommittee, which formed in September to examine the possibilities (and pitfalls) of generative AI, is working on a guide about AI applications in digital advertising, said CEO Anthony Katsur. The target release date is mid-2024 and will address topics such as how to evaluate AI models, bias, copyright, privacy and data accountability.
“The black-box era of AI will have to come to a close,” said Rebecca Sykes, a partner and emerging technology lead at The Brandtech Group. “It’s no longer a legitimate answer to say we don’t really know what’s happening inside the models or it’s unclear where the training data comes from.”
Marketers must understand both the legal landscape and the technology they’re using, she added.
Mitigating misgivings
To reduce their risk, many brands have created AI committees, consulted with legal and compliance folks and drawn up their own internal guidelines on generative AI.
“Responsible use falls on brands because it’s so vital to their brand reputation and trust,” Gartner’s Greene said.
The Brandtech Group shares its legal review and approval frameworks with clients, including a green list of tools and best practices. Companies that want to use generative AI for enterprise applications have to be mindful of data ownership and security, licenses and usage of foundation models, according to Sykes.
Bias and inaccuracy are the top concerns that Forrester’s Lai hears about from marketers when it comes to incorporating generative AI into their marketing workflows.
Many Forrester clients are therefore building brand language models customized for the brand’s own data and desired use cases, he said, “rather than relying on the more open models.”
Think small
But although LLMs will continue to loom large in 2024, we’ll likely see a shift to smaller, multi-agent language models.
These small language models can communicate with each other, learn from one another and conduct smaller, more specific tasks. They’ll be easier to use, less costly and less resource intensive, Greene said, which is also a benefit from a sustainability perspective.
Guess you don’t have to boil the ocean – or input vast amounts of public data – to get value out of an AI model.
Small models do one task very well and don’t require frequent training, according to Paul Pallath, VP of applied AI at cloud consulting company Searce.
“You don’t need to train on billions of parameters,” he said, “to capture the essence” of a task like writing emails.
Yet even as the models get smaller, marketer ambitions for generative AI will only get bigger in 2024.
“We don’t know what we don’t know yet,” said Natalie Silverstein, chief innovation officer at social influence agency Collectively. “There are innovations around every corner.”
