Extraordinary potential, serious concerns

Editor's note: Steven Millman is global head of research and data science at Dynata. He can be reached at steven.millman@dynata.com.

If there’s one thing about the 2024 Consumer Electronics Show (CES) in January upon which most everyone can agree, it’s that 2024 was the year of AI. Carried on the back of the generative AI wave, AI was in just about everything this year – in toys, mobility assistance devices, software, financial tech, health care and literally every other category to be found.

The marketing world was no less affected by the AI fever circulating at CES and I had the opportunity to hear from and speak to a variety of chief marketing officers there and later about how their thinking on the subject has been evolving. Hopefully AI was the only fever circulating but when you put 130,000 people together in one place, you take your chances.

So what are CMOs thinking about the new AI revolution in the marketing space? Here are five major themes I heard from them over the week at CES.

Privacy concerns are expanding

CMOs are used to worrying about privacy, whether it’s protecting their own customers, following relevant data privacy laws such as the California Privacy Rights Act (CPRA), the European Union’s General Data Protection Regulation (GDPR) or handling sensitive health information.

The large language models (LLMs) that underlie generative AI models present new privacy problems that CMOs – along with the rest of the advertising ecosystem – need to manage. The privacy challenge from LLMs centers around the training data used to create these models. LLMs that are not very careful with what they choose to include in their training sets may easily ingest personally identifiable information (PII) that can resurface later in response to a prompt.

On the other side of the equation, PII may be inadvertently ingested into a LLM when included in a prompt – a serious violation of privacy rules – as many LLMs add user prompts directly to their training sets to improve the models. There are already easy-to-find examples of both kinds of violations. 

Finally, there is a new kind of hack frustrating LLMs. It’s called injection prompts, which are hidden instructions placed on a webpage that a bad actor understands will be pulled into the training set for one or more of the larger models. Researchers have created injection prompts that have made major LLM chatbots start to speak like pirates or ask the person using them to provide bank account details.1 Any of these kinds of privacy violations could create major havoc for a brand and result in potentially crippling fines.

How authentic does the creative feel?

In content marketing, brands are eager to engage with their audiences at a deeper level, to make their audiences feel heard and included and to build connections with people that feel personal and real. Authenticity creates both trust and credibility but it’s very difficult to establish in practice. Consumers are discriminating and quick to discern fake or overly polished attempts by brands to connect. 

While generative AI makes it easier to create new content at scale as you’ll see below, there is a concern that overuse of these models in the ecosystem will make all content feel, more or less, like all other content, eliminating the opportunity to create the connection with the consumer to which they aspire. This lack of perceived authenticity could actually turn consumers off to these brands, decreasing engagement with their most coveted audiences.

Speeding up the creative process

All the CMOs I heard from appeared to agree that generative AI as it exists today cannot replicate the kind of creativity that humans are able to produce; therefore, the central role of the content creator in the marketing process is not at imminent risk of replacement as many fear. That said, these tools do present those creators with valuable tools which can speed up the creative process and greatly enhance the scale of production. 

We have all faced at some point the “blank-page problem” – staring at a empty screen or canvas and not knowing where to begin. Generative AI can be an effective assistant in kicking off the creative process by simply writing a prompt that asks the LLM to provide a list of options. 

Consider this hypothetical example I posed to Google’s Bard: 

I am building a new ad campaign for a new electric car called the Whizzer. This electric car differentiates itself from its competitors in that it is sportier, has faster acceleration and better range. We want the ad to drive consideration of our brand for people thinking about buying their first electric car. The audience target is ecologically minded adults with no children aged 25-45 who own their own homes. Please provide two different messages for this ad campaign.

Here is Bard’s response:

Unleash Your Green Speed
Headline: "Eco-Warrior with a Need for Speed: Introducing the Whizzer."
Visual: Show a sleek Whizzer tearing through a winding coastal road, leaving a trail of leaves instead of exhaust.
Message: "Reduce your carbon footprint without sacrificing the thrill of the drive. The Whizzer's sleek design, lightning-fast acceleration and extended range let you zoom past expectations – and gas stations. Leave the boring hybrids behind and rediscover the joy of driving, guilt-free."

Charge Up Your Lifestyle
Headline: "The Whizzer: Power Your Life, Not Your Emissions."
Visual: Showcase a montage of a young professional using the Whizzer for various errands, charging at home and enjoying leisure activities.
Message: "Live sustainably without compromise. The Whizzer seamlessly integrates into your dynamic lifestyle. Power your morning commute, fuel your workday errands and charge up for weekend escapes – all from the comfort of your own home. The Whizzer isn't just an electric car, it’s an extension of your conscious, connected life."

While neither of these message campaigns are nearly polished enough to be ready to use in this form – although I’m personally quite taken with the idea of leaves coming out of a non-existent tailpipe – they do have the ability to inspire a creator to start thinking about versions and variations that may resonate with their audience, greatly reducing the time necessary to generate the new copy. 

If these did not inspire, the creator could simply ask the LLM for five more ideas. Or 10. Or 20. It is similarly simple to iterate on an idea by creating the copy you want and then asking the LLM to give you a version that was directed to men, or to the older age range, or to certain races or geographic residences. LLMs can help creators craft variations on their ideas at speeds that would have been unthinkable just two years ago.

Hyper-personalization of ads

With generative AI tools, it may become possible to not just create new versions of ads for different segments quickly but to deliver unique ads to individual consumers based on demographic and other data already known about them. This is something that has many of the CMOs I spoke with very excited about the future. 

In today’s digital and connected television advertising, the ad delivery systems already go through a process of selecting the most appropriate ad for a given device or household based on what is known about them in their identity graphs. For example, a lipstick brand may not want to waste its ads on men or an arthritis drug brand on an 18-year-old. 

Several large advertisers are currently experimenting with the idea that an ad could be modified in real time to suit the characteristics known or believed to be true about the consumer it is being targeted at. A base ad creative could be modified on the fly as it is delivered by placing the generative AI tool in the same workflow as the ad selection process. The technology isn’t there yet, and speed is still an issue, but this capability could be coming sooner than one might think.

Organizational governance of generative AI

Another area of broad agreement among CMOs was that a new organizational model needs to be developed to oversee gen AI initiatives. It is no longer supportable in their minds for the leadership of gen AI to be verticalized into single organizations like technology, product or data science. 

In order to be successful, there was a great deal of support for the creation of cross-functional teams that combine not only the expertise from the groups above but also marketing, operations, legal and potentially more. This collaborative approach fosters the sharing of ownership and accountability across different functions and ensures all stakeholders are heard. Several CMOs took it a step further and encouraged the creation of centers of excellence to serve as centralized knowledge hubs for expertise, best practices and resources. Teams across the organization would be able to then tap into this expertise for their specific needs.

Regardless of how the governing organization is formed, each of the CMOs indicated that a generative AI lead should be selected to serve as a dedicated specialist, spearheading prioritization and implementation across various functions. This requires finding the right individual with a diverse set of skills and experience who may already exist in the organization or who may need to be hired from outside.

Ignited the marketing world

The generative AI revolution over the last few years has ignited the marketing world, sparking both extraordinary potential and very serious concerns. From lightning-fast creative sprints to hyper-personalized ads, AI promises to transform how CMOs engage with their audiences while at the same time forcing them to navigate a complex and sometimes opaque series of new risks. 

A balance needs to be developed to amplify human creativity and productivity with these new AI tools while at the same time safeguarding privacy and preserving brand authenticity. AI-savvy CMOs who can walk this line are going to do extraordinary things for their brands in 2024 and potentially change the way we think about marketing into the future. 

References

1 https://www.vice.com/en/article/7kxzzz/hackers-bing-ai-scammer.