Amazon, Coca-Cola Europacific Partners and Employment Hero leaders say measuring AI productivity means rethinking outputs, roles and reinvested time
Time saved by agentic artificial intelligence (AI) is being quietly absorbed back into everyday busywork, and HR leaders will only capture the productivity gains if they deliberately redesign roles and workflows, a panel of senior people leaders told delegates at HRFest 2026.
The panel, The Agentic Shift – From AI Tools to AI That Acts, brought together Dean Jackson, director, global digital employee experience at Coca-Cola Europacific Partners (CCEP); Kate Jolly, global head of culture and engagement at Employment Hero; and Pritho Saxena, director HR – China, Australia and South East Asia at Amazon. It was moderated by Jennifer Dobell, chief operating officer at the Australian Financial Security Authority (AFSA).
Dobell framed the discussion around the productivity problem. "The big word with AI is productivity," she said. "It's a really topical area that the Treasurer is even talking about, that Australia is not productive. And everyone knows that they're pinning their hopes on AI."
Where AI productivity gains are showing up
Jackson said CCEP has brought document translation in-house using a partner tool grounded in the company's own language and context. "We were spending a lot of money externally as well, so to actually get things professionally translated," he said. "But now we're saving about 40% of cost doing this all internally. We've got about 50% time reduction doing that as well."
A trial of automated ticket triage in ServiceNow across France and Belgium has delivered "about 15% time reduction there for the people services teams on more generic requests," he added.
Dobell said even government, which she described as a laggard, is changing. "We used to have a lot of teams doing speechwriting because you have to do a lot of presentations. We don't do speechwriting anymore," she said.
How should HR measure AI productivity?
Jolly pointed to Employment Hero's 2026 AI Paradox research, in which three in four employees who use AI said it had boosted their productivity. The open question, she said, is whether organisations see the same result – or whether staff are "just generating absolute slop".
Employment Hero tracks AI adoption first, then measures the real output of each team rather than input metrics. For software engineers, that means features shipped, not lines of code.
"Our target by September next year is to 32x the number of features that we're able to provide for customers," Jolly said. "Think about what the key output is that you want from that team. That should be a productivity measure, not a KPI or an input metric, in my opinion."
Saxena said Amazon applies different measures depending on whether a unit of work is "AI enhanced", "AI native" or "AI frontier" – the last describing an agentic workforce working alongside people. "At Amazon, we don't think that one measure is gonna work for everybody, and we are focusing on the work unit, not on the organisation," he said.
Saxena also cautioned against cutting headcount ahead of the evidence, saying experiments had shown that "the causality between reduction in FTE headcount and business outcomes is not usually as direct as we thought it might be", in reference to full-time equivalent (FTE) roles. "I think the next 6 months is about observation and being comfortable with messy innovation."
Why freed-up time disappears without work redesign
The sharpest warning came from Jackson, whose team runs 150 robotic process automations across CCEP's people and culture processes. "When you look at it, we've saved millions of hours and I still speak to everyone and they're as busy as they've ever been," he said. "So people find work and reasons to do work. So it's actually then going, is this the right work?"
Dobell agreed. "I think if we're truthful, it's getting soaked up in the same work. I don't think we've been as deliberate," she said. "I don't think it's going to happen by osmosis."
Jolly said Employment Hero has asked every leader to redesign their function from first principles. "If you give people the tools and just tell them to use it to make themselves faster, all they will do is their existing job and their existing set of tasks, maybe 10% faster," she said.
The company also now includes an AI proficiency test in every hiring process. "Some people don't pass, candidly," she said.
Saxena added that much of the apparent spare time is not spare at all. "This is not time coming out of anywhere. It's being reinvested into learning, and that's extremely valuable for the next phase of transformation," he said.
Job architecture and data foundations under pressure
Saxena said traditional job families and job codes are "collapsing", and Amazon is now reimagining job archetypes defined by business outcomes – such as stewards, makers, decision makers, operators and scientists. "Job architecture as it existed in the past is not fit for the future," he said.
Jackson said scaling agents across workflows depends on unglamorous groundwork. He said CCEP runs about 117 people and culture applications globally, adding: "The statistics are that only 29% of those are integrated."
"Data, knowledge, integrations, all of those things that no one wants to look at," he said, are what determine whether AI moves beyond individual productivity.
For HR leaders heading into 2027, the panel's message was consistent: pair AI training with deliberate decisions about where freed capacity goes – whether that is customers, candidate care or the HR leadership standards set by APAC's top women in HR.
As Jackson put it: "If you don't do it, people will find reasons to be busy."