HR leaders are being handed two different pictures of the same rollout. Vendors promise hours returned and complexity erased. The people doing the work, surveyed independently, say AI has added a second job on top of the first. I do not think those pictures can both be right, and I think HR has been too willing to buy the first one on the vendor’s word alone.
The pitch versus the experience
Take DXC Technology’s recent pitch for its workplace services line: a 40% reduction in operational complexity, 60% fewer service-desk calls, half of device issues resolved before an employee even notices, and more than 15 hours of productivity “returned” to every employee every month. “Enterprises should not have to choose between advancing business priorities, improving IT efficiency, and delivering a stronger employee experience,” said Kelly Candler, DXC’s global offering lead for workplace and business process services, arguing the platform delivers all three at once.
Now put that next to what workers themselves report. HERE’s newly released survey of 1,000 finance, healthcare and public-sector professionals found 48% say AI has created roughly as much extra work as it has saved, 30% spend at least half their workday manually re-entering information between AI and legacy systems, and 72% admit to routing around their company’s AI restrictions entirely. These are not the same claim measured two different ways. One says the friction is disappearing. The other says employees are quietly building workarounds because it is not.
The strongest case against my argument
The honest counter is that these are not directly comparable studies. DXC is describing IT-managed workplace services, ticket volume and device provisioning, not general-purpose generative AI tools. HERE surveyed a different population using different products for different tasks. It is possible both numbers are true in their own narrow lane: fewer service-desk calls because painful IT chores are automated away, and more re-entry work because employees are separately layering consumer AI tools onto systems nobody redesigned around them. Vendors would also point out that efficiency gains compound, and that six months of adoption data looks different from six weeks.
I take that seriously, but it does not hold up under the scale of the gap. If AI were genuinely returning 15 hours a month per employee, that relief should be visible somewhere in how workers describe their week, even accounting for different tools and different populations. Instead, independent research keeps landing in the same place: Federal Reserve researchers found the productivity payoff companies tout on earnings calls is still mostly a promise, not something showing up cleanly in output. When the vendor-reported number and the worker-reported experience diverge this consistently, across this many independent surveys, the burden of proof sits with the vendor, not the workforce.
What HR should actually do with vendor numbers
None of this means AI tools are worthless or that DXC’s platform does nothing. It means HR should stop treating a vendor’s productivity claim as a finding and start treating it as a hypothesis the organization has to test on its own workforce. That is a solvable problem: track toggle time and re-entry volume before and after rollout, the same way you would track time-to-fill or attrition, instead of accepting hours-saved figures generated by the company selling the tool.
Concretely, that means writing the verification step into the contract before signing it, not after the rollout stalls. Ask the vendor for the methodology behind any hours-saved or complexity-reduction figure, and ask whether it was measured on a comparable workforce or modeled from an idealized deployment. Then run a 60-day pilot with a control group that keeps the old process, and measure toggle time and re-entry volume directly rather than trusting a satisfaction survey the vendor helped design. If the number holds up under that scrutiny, the tool earns the wider rollout. If it does not, HR has caught the gap before it shows up as unplanned overtime and burnout six months in, which is exactly where too many AI rollouts are quietly landing right now.
It also means being honest about incentives. A vendor’s productivity number is marketing before it is measurement, built to close a deal, not to survive an audit of how a specific workforce actually uses the tool day to day. Independent worker surveys carry their own limits too, self-reported perception is not the same as an output metric, but at least the incentive runs the other way. Until vendors publish rollout data that a third party can verify against actual usage logs, HR teams that budget AI initiatives on the vendor’s hours-saved number are, more often than not, budgeting against a claim nobody outside the vendor has ever checked.
Sources: DXC Technology and HERE via GlobeNewswire
