The premise behind most workplace AI adoption is that employees end up with less to do. Marsh's global digital leader for employee benefits, Kate Brown, says that is not what her team is observing. Employees using AI are reporting that they are busier than before, not less busy, because the technology allows them to handle more work in parallel rather than genuinely reducing their overall load. The tasks do not disappear. The throughput increases. And the cognitive strain moves with it.
Brown, speaking on Marsh's Transforming Benefits with Technology panel series alongside CTO Andrew Owens and head of product development Simon Jarvis, put the implication directly: most employers responding to AI adoption by assuming it reduces the case for updating employee support programmes are operating on a flawed assumption. The mental health and wellbeing benefits sitting underneath an AI-enabled workforce were, in many cases, designed for a pre-digital working environment. Brown's characterisation - that some employers are still relying on mental health benefits created in 1984 - is pointed precisely because the gap between what AI is doing to cognitive load and what benefits programmes are designed to address is widening rather than closing.
For benefits brokers, that gap is the immediate practical question. A client rolling out AI tools across their workforce is not automatically reducing the stress, burnout or cognitive overload risk their benefits programme was built to manage. In some cases they are increasing it, and the benefits package has not been revisited to reflect that.
Jarvis framed the broader arc of AI adoption in benefits administration as moving through three stages: automation of discrete tasks, redesign of whole processes, and eventually the elimination of certain tasks altogether. At the far end of that arc, benefits decisions are increasingly made or suggested for employees based on data the organisation already holds, with active employee choice reduced rather than expanded. HR's function shifts from running processes to governing whether AI is being deployed appropriately - a materially different role that requires different skills and a different relationship with benefits data.
Owens, whose remit covers the technology architecture underpinning Marsh's benefits platform, pointed to the connectivity question as the near-term challenge: most employers are running benefits across systems that do not share data effectively, which limits both the quality of AI-assisted decision-making and the ability to spot wellbeing trends before they become claims. More connected infrastructure is a prerequisite for the more sophisticated AI applications the panel described, and most employers are not there yet.
The panel identified hyper-personalisation and digital twin technology as the developments most likely to reshape employee benefits over the next several years. A digital twin in this context means a model of an individual employee's likely future health trajectory, built from wearable data, biometric information and lifestyle inputs, used to guide preventive benefit design rather than responding to conditions after they develop.
The opportunity is genuine. Preventive intervention delivered at the right moment, before a condition becomes acute, reduces both the human and financial cost of employee health risk. For self-funded employers in particular, shifting spend from reactive claims to proactive prevention is one of the most commercially compelling arguments in benefits strategy.
But the panel was direct about the barrier. Employees will not share the data that makes personalisation meaningful unless they trust that the organisation will protect it and can demonstrate clearly how sharing it improves their outcomes. That is not a technology problem - the infrastructure for collecting and processing biometric data exists and is improving. It is a communication and governance problem, and most employers have not resolved it. Asking employees to share sensitive health data without a clear, credible explanation of how it is used and who can access it will produce low participation rates that render the personalisation model ineffective regardless of the underlying technology.
For brokers advising on benefits design, the trust question is worth surfacing before a client invests in personalisation infrastructure. The technology is only as useful as the employee participation rate it generates, and that rate is a function of trust that has to be built deliberately rather than assumed.
Three things follow from what Marsh's panel described, each of which is a legitimate reason to bring a client's benefits programme back to the table outside the standard renewal cycle.
First, if a client has deployed AI tools across their workforce in the past 12 to 18 months, the mental health and wellbeing support sitting beneath that deployment is worth reviewing. The cognitive load profile of their workforce has changed. The benefits have not.
Second, the data connectivity question has direct implications for how a benefits programme is structured going forward. Employers whose benefits systems do not share data are not positioned to use AI-assisted decision support effectively, and that gap will widen as competitors invest in more connected infrastructure.
Third, for self-funded clients in particular, the shift from reactive claims management to preventive intervention is a strategic argument worth making now, while the employer is still forming its view of what AI means for its workforce. That conversation is easier to have before a client has locked in another year of the same programme than after.
The Marsh panel's framing - that AI is moving beyond isolated automation and into core benefits strategy - is correct. The practical corollary for brokers is that a client's AI adoption decisions and their benefits programme are no longer separate conversations.
