Rex Briggs, a noted AI technologist, was conducting research into boating enthusiasts to see what marketing messages resonated most with different types of buyers of high-end and low-end watercraft. Briggs turned to OpenAI’s ChatGPT to get answers from AI-generated customers, asking the digital copycat consumers what they thought about boats.

“It gave me goosebumps in a way,” said Briggs, who recently joined Claritas as its chief AI officer and also works with MMA Global, a marketing industry association. “The more profiling data [you give it], the more realistic these personas are becoming.”

The research was for a marine vehicle technology brand that Briggs was not allowed to identify publicly. By feeding OpenAI anonymized customer data from zip codes across the U.S., Briggs was able to interrogate different computer-generated personalities. One was a fisherman that called itself Earl. It said it wore a John Deere hat, Carhartt shirt, faded Levi jeans and Red Wing boots. What did Earl want most in a boat, for its frequent, yet completely non-existent, weekend fishing trips in rural America? “Durability,” Earl said.

“It accurately understood the difference of the types of boats that different collections of zip codes would very likely buy, and it matched up with sales data,” Briggs said in a recent phone interview, “and one of [the AI consumers], from the rural part, named themselves Earl and described himself.”

AI-generated consumer profiles can be used by brands to create digital focus groups that act like real people, or they can be used to hyper-personalize products and services.

Also read: Pereira O’Dell holds focus groups with AI-generated consumers

In the past year, as OpenAI’s ChatGPT rose to prominence, ad agencies and major brands started racing toward an AI future. With chat-enabled AI, brands can ask questions and get feedback from characters that mimic real consumers, and that’s just the start.

Some AI evangelists see the potential to make versions of consumers, based on highly personalized data, that will almost represent an exact replica, or digital twin, of true customers. Brands could use digital customers to gauge the potential for a new product, test marketing slogans and conduct product research. In the future, brands could even market to machines that are stand-ins for the people they mirror, and consumers can use their digital twins as assistants for shopping.

“You give it your own proprietary set of data and information, and you layer it in a way that builds up a persona that’s realistic,” Briggs said.

AI likes it

Coca-Cola, Mondelēz, Estee Lauder, Marriott and many other brands are using AI tools to guide their strategies and tailor offerings. Publicis, Omnicom, WPP, Dentsu and other large advertising holding companies are working with OpenAI, Microsoft, Google and Amazon, among others, to build AI-powered services for brands. Leveling up in AI can help brands strategically and creatively.

For instance, in September, Coca-Cola, which is widely viewed as being on the leading edge of AI, launched a limited-edition soft drink it called Y3000. The new drink and flavor were concocted based on AI research, which Coca-Cola performed through its partnership with OpenAI and Bain & Company.

Read more about the Coke-Bain partnership here

“It has huge potential, and we are already using it,” said Pratik Thakar, Coca-Cola’s head of global creative strategy and content, referring to the idea of AI-generated consumer consulting panels. “We used some of those [AI] tools to create the taste palette, but also to test whether it works, what people think and whether it’s going to be palatable for our young audience.”

“So, research and development, using AI, is part and parcel of how we use AI,” Thakar said.

Coca-Cola also is using consumer-facing generative AI in its marketing, with a site it launched this year called Create Real Magic. That’s where consumers go to make AI-generated images infused with Coca-Cola’s brand iconography. The technology is built on DALL-E 3, OpenAI’s image creation platform.

Coca-Cola had to work with OpenAI to create the custom site so that the AI could understand the brand. Coca-Cola helped train the model to replicate the famous Coca-Cola script, the exact shade of red the brand uses and other nuances, Thakar said. Consumers type in prompts that return personalized images. While in this case the AI is predominantly being used for marketing, this type of interactivity keeps educating the model, too, Thakar said.

“There’s millions or billions of people actually training the model,” Thakar said, “and I see that as a giant focus group, happening every second of the day and night throughout the world, across the world. So, the machine is learning on every interaction and that’s why this kind of research is really fascinating.”

Always-on personas

Last month, Pereira O’Dell, the independent ad agency, launched an insights tool with “AI personas,” built on top of Chat-GPT 4 and trained with extra consumer data that is readily available to marketers. The personas are based on broad categories of consumers, so they are not digital twins of real people, but the chatbots can be enhanced by brands that have their own data, according to PJ Pereira, co-founder and creative chairman at Pereira O’Dell.

In the travel category, a brand could customize the chatbots with, say, frequent flyer data to better represent the customer base. Pereira O’Dell started with travel personas for the first AI focus group and is expanding to auto, retail and finance.

“You can click on one [of the chatbots] and ask questions about specific things, about travel or anything else you want, to get a better sense of who that person is,” Pereira said, “but when it gets really interesting is when you click on more than one persona, because then you can ask the same question for multiple people, and they answer at the same time, and you can see the contrast between the different personas.”

The evolution of AI into true digital twins will be difficult because of the sheer amount of computing power it requires. “That’s time-consuming and a little bit expensive right now,” said Todd Thiessen, managing partner at Plan.Net Americas, a digital services firm that partnered with Pereira O’Dell on the AI personas tool.

Still, the industry is interested in testing the limits of digital twins and using AI as super-consultants, Thiessen said. “That’s where this is going and it’s probably going to happen before we know it,” Thiessen said.

There are privacy issues to consider with the advent of consumer avatars, and it is part of a broader conversation in advertising about how brands collect and use consumer data.

“It kind of scares me when I hear digital twin,” Thiessen said.

With enough data fidelity, AI will be able to approximate a consumer with close to 95% accuracy, Thiessen said. For its AI personas, Pereira O’Dell used data sources that are commonly distributed in other parts of the advertising ecosystem—including anonymous site surfing data, Adobe Analytics, and anonymized Salesforce data. If more types of information are ever used, such as a brand’s proprietary customer data, it would require more guardrails and consent from consumers, Thiessen said.

There has been a movement in ad tech, phasing out third-party internet trackers such as cookies, to deploy new forms of consent and transparency. When a consumer visits Coca-Cola’s Create Real Magic site, for instance, they are immediately met with a link to the “Privacy Center Preference” page, where they can read about how cookies and personal information is used.

Brands are trying to build direct relationships with their customers, through internet services and other forms of marketing, including through these new AI-based promotions. The industry also is developing data ecosystems, through cloud computing and clean rooms, designed to prevent leaking data or using it in ways that violate consumers’ data choices.

Twin speaks

Nigel Vaz, CEO of Publicis Sapient, already created a digital twin of himself that has his voice and can manifest as a 3D virtual model. Vaz’s AI doppelganger was trained by ingesting the text of his book “Digital Business Transformation,” and using video of his public appearances. When the AI speaks it’s almost indistinguishable from himself, Vaz said in a recent phone interview, while assuring that the voice on the other end was really him.

It takes about 15 hours of computing to train the chatbot to speak convincingly as Vaz for just five minutes. The avatar gets updated frequently based on material that Vaz generates, such as his latest media appearances.

The chatbot also is available to communicate with employees in the agency, which is all part of the much deeper work Publicis Sapient is doing with digital customer twins. Publicis Sapient is part of Publicis Groupe, which also owns Epsilon, a massive consumer data firm.

“Using that ‘digital twin’ metaphor,” Vaz said, “you can essentially embody all the characteristics of specific consumer groups and start interrogating them.”

What’s powerful about the digital consulting is that a brand can experiment with thousands of different types of people and they rapidly return insights. Also, with a human focus group, the responses depend on a person’s mood at any given time, and responses change based on everything from the weather to the type of day a person is having. “Filter in all the context of my day then ask me to make a [consumer] choice,” Vaz said, “and it will be a very different answer.”

As for AI-generated customers, they can give feedback for any scenario and any conditions, and all variables can be tested, Vaz said.

Digital twins also help test products. When a brand is about to launch a new product, it often asks a human panel to describe all the ways they might use the product.

“We apply digital twins in a consumer scenario and start to see how a consumer might use a product,” Vaz said. “How many uses for a particular thing can you come up with in a minute? Ask AI the same thing, and it’s significantly more.”

Eventually, Vaz predicts that consumers will manage their own digital twins—virtual representations that successfully capture a person’s preferences. That could fundamentally reshape advertising. “I might allow my digital twin to go grocery shopping without me actually having to pick what brand of chicken I want,” Vaz said.