AI-generated avatars are becoming easier to create and more realistic—and are beginning to afford marketers new ways to connect with consumers.

D-ID, a software development company, is rolling out a marketing suite for brands to develop avatars to serve as digital spokespeople. These personas, which can be modeled after real or fictional people, are meant to advance communication with consumers by offering personalized interactions, such as on a brand’s website. 

“For every visitor, you can have a new avatar,” said Gil Perry, co-founder and CEO of D-ID.

Pepsi and Warner Bros. have previously worked with D-ID, and plan to do so with these new tools, Perry said. The suite is also meant to help brands deploy significantly more personalized video content, such as on social media, than would be possible with human talent.

D-ID’s new launch illustrates the growing usefulness that AI-generated personas are offering marketers. The technology is allowing organizations such as Le Pain Quotidien and Pereira O’Dell to digitally clone and manufacture artificial customers, to develop consumer insights. Meanwhile, brands such as Qatar Airways are testing digital versions of support specialists. And social platforms including TikTok are leveraging avatars to help streamline content creation. 

As with all AI use cases, however, deploying avatars is not without its imperfections. Media agency Noble People has experimented with HeyGen, an avatar platform similar to D-ID’s, and while the firm was able to successfully model a persona off of one of its employees, doing so required significant work just to get the avatar to sound somewhat natural.

Time efficiency is often promoted as one of generative AI’s key strengths, but using avatars “certainly wasn’t a time-saver,” said Todd Alchin, partner and CMO of Noble People. 

More generally, consumers are skeptical of brands using generative AI in their marketing. And 64% of consumers prefer that companies not use AI in their customer service, according to a Gartner survey published this summer.

Also read: AI in ads divides consumers

Despite these reservations, D-ID sees consumer interaction as a primary use case for AI avatars. Perry stressed the realism of D-ID’s personas, both as stock models and as representations of real people. The technology leverages advancements in synching mouth movements with speech, as well as processing new speech with natural sounding inflection, pauses and other details of communication.

In order to create an avatar that can represent an individual brand, the bot is trained on brand data. It can use retrieval-augmented generation (RAG) to access a knowledge base in real time as a way to surface more information, such as from a website or social media platform. For example, an anonymous D-ID client created a bot intended to be an expert on environmental, social and governance (ESG) issues. The client published this bot on its website so that consumers could interact with it to learn more about the company’s ESG efforts.

These personas can also be personalized to specific consumers. This happens through a process of data collection performed by the bot in real time. A brand can assess this data and then tweak avatars to fit cohorts of customers. Still, it remains to be seen just how bespoke these avatars can be. Location data can easily be determined so that the avatar is more likely to speak in a language that the consumer speaks, but D-ID did not provide evidence that more fine-grained information, such as interests, can be understood and leveraged. 

Beyond consumer interaction, media optimization is another use case that D-ID is aiming to advance through the use of its avatars. In a hypothetical example, Perry said that Pepsi could A/B test ads that feature avatars expressing different messages, allowing for real-time adjustments based on how consumers are responding. The avatar itself could just as easily be swapped out with a different one, Perry said, which is not as feasible when using real people. 

Similar to TikTok’s avatar rollout, D-ID is hoping that brands and creators will develop AI-generated iterations of real people so that they can also reach more consumers with personalized media. Video, in particular, is an expensive format to develop and customize, Perry said, making it a prime target for replicating advantages offered by using avatars. Indeed, video has been one of the fastest growing areas for AI marketing use cases, thanks to its resolving bottlenecks typically faced in the production process.

As D-ID’s avatars get more face time with brands and customers, their emotional aspects could improve, which is key to increasing the believability of digital personas, said Noble People’s Alchin. But already, Alchin can detect significant improvements in the technology, which are evident by the fact that avatars are no longer being billed as mere support specialists, but as dynamic spokespeople for brands themselves.

“It’s like they’ve gone from being in the kitchen to front-of-house,” Alchin said.