Using custom digital twins to better target messaging

The emergence of generative AI has brought significant changes to many areas of the public relations sector. One gaining traction right now is “digital twins” – virtual replicas of target personas.

In essence, digital twins are similar to the character profiles used in communications and marketing plans for decades, according to Ephraim Cohen, global managing director of media, platforms and storytelling at FleishmanHillard. The key difference is that digital twins are more dynamic in that they can use a broader range of data, which leads to better insights based on real-world results.

Cohen’s FleishmanHillard team first delved into generative AI more than five years ago as part of its ongoing effort to better understand audiences. When ChatGPT and other next gen products burst onto the scene around 2022, the agency “almost immediately started looking at digital twins” to make harvesting those insights “much faster, much easier and much more cost-effective to develop” than the traditional personas created by hand.

Jon Lombardo, co-founder of synthetic research platform Evidenza, noted that digital twins  also offer enhanced flexibility, allowing teams to test multiple messages simultaneously and compare the responses to each version.

“Things that used to take months now take literally minutes,” he said.

Developing a program

FleishmanHillard assembled a team of about 50 people to begin developing and testing its digital twin frameworks. The initial approach involved training bots on datasets related to audience behaviors, preferences and online conversations.

Cohen couldn’t go into detail about specific types, but he said FleishmanHillard’s main focus has centered on B2B and B2C audiences.

“We didn’t want to do it with real people, because there are a lot of legal and ethical ramifications there,” Cohen explained. “So we started by taking data sets on how people behaved, their favorite brands, purchase habits and online conversations, and then training bots on those data sets.”

The agency also used qualitative data sources such as academic papers, books and news clippings to build a more comprehensive understanding of their target audiences. That gave insight into things like behaviors, word choices and even their general thought processes.

The goal was not only to gain deeper audience insights, but to also create interactive tools that could assist with media relations and content strategy, Cohen said. He shared that the agency has even experimented with creating profiles of journalists and influencers to better understand how to position stories and content in a “way that resonates.”

Digital twins have made the once static persona “come to life,” Lombardo said. They can have actual names, roles and financial information, allowing for in-depth questioning. By asking the virtual person questions, they can gain the immediate feedback needed to model customer preferences, motivations and pain points.

“(Digital twins can) model the entire sample and give you a more robust view of what the market thinks,” Lombardo said. He added that Evidenza’s clients have had the most success using the platform to reach hard-to-access communities.

“Most of the people that PR people want to impress are not taking surveys or picking up the phone,” Lombardo said. “And in some ways, the only way to talk to them or model them is to use AI.”

Another area digital twins can help with, Lombardo said, is narrative and message testing – understanding how different stakeholders will respond to new campaigns or messaging. Beyond just helping to generate ideas, Lombardo advised PR pros to start asking their AI personas if they like the story angle or the messaging and why they feel that way.

Not as effective with real-time analysis

While the technology is promising, Lombardo highlighted that digital twins have limitations, particularly in real-time situations, such as crises or campaign results as they’re coming in.

“It’s very good at things that have a broader view, like research or segmentation or narrative testing,” he said. “It’s not as good at things that depend on real-time, immediate assessment.”

Cohen largely agreed with those sentiments, especially for digital twin programs just getting started.

Digital twins won’t be perfect from the start, especially when it comes to real-time processes. This is largely because most standard platforms aren’t designed for real-time use, but rather to learn from past data.

Cohen said that many of the initial results from FleishmanHillard’s trials weren’t relevant to their work because they were based on outdated information. It’s possible to train tools and keep them up-to-date, Cohen said, but a team needs to constantly feed and update them with new information. While it’s possible in theory, Cohen said that in practice, most digital twins would require a custom application to draw on real-time data.

While the technology isn’t there yet, Cohen said he feels the technology is moving in that direction. He said many augmentation technologies, like APIs, that can connect to bring in real-time data to a gen AI application.

“As it stands, we’re not there yet,” Cohen said. “But we’ll get there.”

Casey Weldon is a reporter for PR Daily. Follow him on LinkedIn.