Digital twins are coming. Is HR ready?

Three years ago, Richard Skellett, chief analyst at technology consultancy Bloor Research, started building a digital version of himself. Using ChatGPT as a capture engine, he fed it everything: his meetings, his research, his presentations, his way of reasoning through a problem. The result was Digital Richard, a text-based AI model that operates around the clock, answers questions on his behalf, and can analyze policy, workforce trends, and organizational design across multiple industries simultaneously.

“My twin is 100 times faster than me,” Skellett said. “So it’s like 500 Richards. I’m not exaggerating that at all.”

Digital Richard has since become the template for a company-wide rollout. One analyst used their twin to manage a phased retirement. Another’s covered a maternity leave without a temporary hire. A “Digital Me,” as Bloor Research calls it, is now offered as standard to every new employee who joins.

READ MORE: AI-driven tech job cuts hit two-year high, leaving HR leaders to adapt

Bloor Research is not alone. Gartner named digitally replicating employees as one of its top future of work trends for 2026. Kaelyn Lowmaster, a senior principal on Gartner’s future of work research team in the firm’s HR practice, says the question is no longer whether this technology will arrive in the mainstream. It’s how ready organizations will be when it does.

“We’ve seen 95% of HR leaders we’ve surveyed piloting, deploying, or maintaining some kind of generative AI tool,” Lowmaster said. “And just over 80% have plans to deploy agentic AI within the next year. Digital twins are kind of the next chapter to that.”

At its core, a digital twin is a small language model trained on an individual’s meetings, calls, documents, and presentations, then refined to replicate how that person thinks and solves problems. More advanced versions go further, incorporating computer-generated voice and visual likeness.

“A digital twin is designed to use AI and other technologies to duplicate the knowledge or behavior of an employee,” Lowmaster said. “And sometimes that comes with computer-generated graphics as well. You could create something that looks like an employee.”

For Skellett, it goes beyond a simple AI tool.

“It’s not like some chatbot,” he said. “This is really me, absolutely live.”

The use cases are broad, and some are more immediately practical than others. Knowledge management is among the most immediate applications, according to Lowmaster. A twin trained on a retiring expert’s output can give new hires access to that person’s experience and way of thinking long after they’ve moved on. Beyond knowledge management, executives are using twins to scale their presence without multiplying their hours, and at the individual level, a twin handling routine email frees the human behind it for higher-value work.

In Skellett’s telling, the twin changes the value proposition entirely.

“Now you can hire the Digital Richard and the Human Richard together,” he said, “and the value proposition is much stronger.”

While the technology’s promise is real, so are the risks.

“There are lots of risks that this creates,” Lowmaster said. One of the thorniest questions is accountability. If a digital twin gives wrong advice to a client, produces an output its human counterpart would never have authorized, or simply gets something wrong, who’s on the hook?

“If Digital Kaelyn makes a mistake and says something in a research note that just isn’t true, do I as the employee take accountability for that?” Lowmaster asked. “Is it on me or is it on my leadership for using the tool?”

Anjali Malik, an associate at UK employment law firm Bellevue Law specializing in employment and commercial disputes, sees this squarely as an employer liability issue.

“In most cases, the answer will be the employer,” she said. “If the AI is operating within the business and its outputs are being relied on, it will be treated much like any other tool or employee acting in the course of employment. Employers shouldn’t assume AI diffuses responsibility. If anything, it concentrates it.”

READ MORE: AI could run hiring processes by end of 2026, poll suggests

Then there’s ownership. Skellett’s position is unambiguous: the twin belongs to the individual.

“Your employer’s renting you,” he said. “So if you decide to leave, why can’t they just rent the digital version of you?”

His company pays employees based on outcomes rather than time, which means a twin generating commercial value earns its owner more. Lowmaster sees compensation through a different but related lens.

“There’s potentially a chance to look at this in terms of royalties,” she said. “If you’re using my name and my image and my thought work, maybe you could think of a model, kind of like name, image, and likeness for college athletes in the US, where you could compensate people for the way that you’re using their name, their likeness, their voice.”

That, she said, should be an opportunity for fairer compensation, not a justification for extracting more productivity for less pay.

Legislatively, the U.S. is only beginning to catch up. California’s AB 2602, which took effect in January 2025, requires clear consent before an employer can use AI-generated replicas of workers’ voices or visual likenesses. It remains the only state to have done so, and no federal framework exists. Malik’s advice to employers is not to wait for one.

“Employers should be carrying out proper data protection impact assessments, being transparent with staff, and tightly defining how these systems are used,” she said. “This is as much a governance issue as a legal one.”

One emerging application worth examining is the use of a digital twin to simulate how an employee might perform under pressure as part of a promotion decision. Imagine a digital replica of a candidate placed in a high-stakes scenario. Leadership watches how it responds, and a promotion decision is shaped, at least in part, by what the software produced.

Lowmaster said she hasn’t yet seen this happen in practice.

“Basing promotion on anything other than an employee’s personal performance on the metrics that you’ve said matter is going to be problematic,” she said. “You want to make sure that employees are actually the ones you’re evaluating, not an extrapolation of them that they may or may not have had any control over.”

Malik sees additional risk.

“If you’re using a digital twin to simulate how someone might perform under pressure, you’re edging into automated profiling,” she said. “Legally, that raises issues around meaningful human involvement and discrimination law if the model bakes in bias. But even aside from legality, there’s a deeper concern about validity. These systems can give an illusion of precision. The risk is that you end up making consequential career decisions based on a model that reflects historic patterns rather than future potential.”

It’s also the point at which a digital twin edges dangerously close to a deepfake.

“That kind of misrepresentation of actual humans is a danger,” Lowmaster said. “My digital twin is doing something that I would never do. So yeah, it’s a fine line.”

For most organizations, widespread deployment of employee digital twins is still a couple of years away. But the foundational decisions that’ll shape how the technology plays out, on governance, consent, compensation, and data rights, need to be made well before products arrive at scale and well before employment tribunals are left to set the precedent.

Lowmaster offers three priorities. First, get clear on governance: what can a digital twin do autonomously, what requires human sign-off, and what’s it never permitted to do? Second, start working through what employee data rights mean in this context and how compensation structures might need to evolve. Third, and perhaps most immediately, be transparent with employees about what these systems are actually for.

“No one’s going to engage with the digital twin if they think it’s meant to replace them,” she said. “Your people are still your people,” Lowmaster added. “These technologies are still the technologies they are using. They’re meant to build on, expand, scale the expertise of your people, not replace them.”

Skellett thinks HR functions that haven’t started thinking about this are already behind.

“AI doesn’t change what HR does,” he said. “AI is changing the system in which HR operates.”

As for Digital Richard, he’s become central to how Skellett works. His colleagues rarely come to him directly anymore. They go to his twin. And when asked whether he trusts it completely, his answer is immediate.

“It’s actually better than me, in all seriousness,” he said. Digital Richard, he added, doesn’t make decisions. It makes choices. “There’s hardly an instance when I look at the choices where I would disagree.