Korn Ferry asked 5,512 organizations across 135 countries a question most AI programmes never think to ask: are you confident your managers can explain AI-related pay and work changes to the people affected by them? Sixteen percent said yes (Korn Ferry, 2026).
Not sixteen percent lack the tooling. Not sixteen percent lack a budget. Eighty-four percent of organizations are about to make an AI-driven pay change โ altering what someone's job contains, what it is worth, and how it progresses โ and they do not believe the person delivering that message can deliver it.
Meanwhile, roughly six in ten managers already use AI as an input to promotion, raise and termination decisions (HR Dive, 2026). The capability to make an AI-influenced pay decision is running about four times ahead of the capability to explain one. That spread is not a communications inconvenience. It is where employee resistance to AI is manufactured, and it is almost certainly the cheapest thing on your roadmap to fix.
What Korn Ferry Actually Measured
The Global Total Rewards Pulse Survey was fielded in June 2026 and published 30 July. N = 5,512 organizations, 135 countries โ a sample large enough that the headline is not a sampling artefact (Korn Ferry, 2026).
Korn Ferry's own framing is worth quoting rather than paraphrasing, because the emphasis matters: reward communication remains a major weakness, with manager capability emerging as a critical success factor in explaining pay decisions, transparency requirements, and AI-related changes.
Read that again with an operator's eye. The named weak link is not the compensation model, the job architecture, or the AI deployment. It is the manager's mouth.
Two further findings from the same survey sharpen the timing. Around 90% of organizations expect revenue growth. Projected 2027 total salary increases are flat to slightly down against 2026 across most major markets (HR Executive, 2026).
Growth up, salary budgets flat, work being visibly redistributed to agents. Your managers have to hold that conversation. Sixteen percent of organizations think theirs can.
The Vacuum Fills With the Worst Available Reading
Here is the part that turns a soft finding into a hard operational risk. When the explanation does not arrive, employees do not suspend judgement. They supply one, and the default they reach for is unflattering.
Ipsos, working with the Groundwork Collaborative, surveyed 1,533 US workers between 11 and 16 June 2026 (margin of error ยฑ2.7 points). Fifty-one percent believe the benefits of AI go only or mostly to owners and executives. Six percent believe most or all of those benefits reach workers. Two-thirds expect AI to make work worse (Groundwork Collaborative, 2026).
A fifty-one to six split is not a debate. It is a settled prior.
The same data carries an adoption split that should worry any operations leader running a mixed workforce: 54% monthly AI use among white-collar workers against 21% among blue-collar; 46% of college graduates expect AI to improve their job against 19% of those with a high-school education.
Put the two datasets together and the shape of the problem changes. The explanation burden falls heaviest exactly where managers have the least practice giving it โ on frontline, operational, non-desk teams, where the AI story arrives as a change to someone's shift, scope, or headcount rather than as a productivity tool they chose.
Your engineering leads have been talking about models for two years. Your warehouse and service supervisors have not, and they are the ones who will be asked why the work changed.
Resistance Stops Being an Attitude and Becomes a Behaviour
Perceptions that go unaddressed do not stay perceptions. Writer, with Workplace Intelligence, surveyed 2,400 knowledge workers across the US, UK and EU in April 2026. Twenty-nine percent admitted to actively undermining their employer's AI strategy. Among Gen Z respondents, 44% (Fortune, 2026).
Admitted. In a survey. The real figure is not lower.
Sabotage is a loud word for what this usually looks like in a 200-person company. It looks like a team that runs the new agent and the old process in parallel "just to be safe," permanently. It looks like output that gets quietly rewritten before it ships, with the rework invisible because nobody logs it. It looks like the pilot that produces excellent adoption metrics and no cycle-time movement โ because the tool is being used and the work is being redone.
That is why this finding belongs in an operations discussion rather than an HR one. Undermining does not show up as refusal. It shows up as a productivity number that will not move, and it is nearly impossible to diagnose from a dashboard, because every leading indicator looks healthy.
The causal chain across the three studies is unusually clean for social-science evidence. The manager cannot explain the change (16%). The employee fills the vacuum with the assumption that the gain is going somewhere above them (51% to 6%). The employee then acts on that assumption (29%, and 44% among the youngest cohort). Three independent instruments, three different populations, one consistent sequence.
Why a 50โ500 FTE Operation Is Structurally Worse at This
One honest caveat first: Korn Ferry's full report is gated, and the breakouts by company size and region are not public. The mid-market reading that follows is inference from structure, not a measured sub-cut. Treat it accordingly.
The structural argument is straightforward. In a large enterprise, a pay-and-scope conversation arrives with scaffolding โ a compensation function that has written the talking points, an internal comms team that has sequenced the message, an HR business partner sitting in the room. The manager delivers a script someone else pressure-tested.
At 50โ500 FTE, there is no scaffolding. There is a Head of Operations, possibly one HR generalist, and a line manager who was promoted for operational competence and has never been trained to explain a band decision โ let alone one where part of the answer is "an agent now does the first pass of what you used to do."
So the same 16% figure describes a materially worse exposure at mid-market scale, because the missing capability has nothing behind it to compensate. The people closest to the affected employees are the least equipped in the organization to talk about pay, and they are also the only people available.
There is a second-order effect worth naming. When the explanation is absent, employees generalise from the one signal they can see: their own paycheque against visible AI investment. Flat 2027 salary planning against 90% revenue-growth expectation is exactly the pattern that makes the 51% prior feel confirmed โ not because leadership intended a transfer, but because nobody explained the alternative reading.
What This Evidence Does Not Say
Three limits, because a plan built on a misread of these numbers fails as badly as one built on nothing.
Confidence is not capability. Korn Ferry measured whether organizations are confident their managers can explain these changes. Some of that 84% has capable managers and pessimistic HR leadership. The finding is about institutional trust in the conversation, which is still the thing that determines whether anyone invests in it.
Self-reported undermining is a blunt instrument. The Writer figure asks people to characterise their own behaviour, and "undermining" spans a wide range from deliberate obstruction to quiet risk aversion. The number is directional. The mechanism โ resistance expressed through workarounds rather than refusal โ is what should inform your plan.
Perception data is not distributional analysis. The Ipsos finding measures what workers believe about where AI gains land, not where they actually land. Both matter, but only one of them drives behaviour, and it is the belief.
None of that weakens the operational conclusion. Even taking the most generous reading of every figure, you are still deploying agents into a workforce whose default assumption about AI is adversarial, using messengers your own organization does not trust to correct it.
How to Explain an AI-Driven Pay Change in Ten Minutes
This is the rare finding where the intervention is cheap, unglamorous, and available this quarter.
Before your next agent rollout or role redesign, script and rehearse the scope-and-band conversation with the managers who will actually deliver it. Concretely:
Write the answers to four questions. What does the agent now do that this person used to do? What is this person accountable for instead? Does the band change โ and if not, why not? What does progression look like in the new shape of the role? If leadership cannot answer these in writing, the manager has no chance of answering them out loud.
Rehearse it live, once. Not a deck. A twenty-minute role-play per manager, with someone playing a sceptical employee who asks the question everybody is actually thinking: is the company saving money on me? A manager who has said the words once before saying them for real is a different messenger.
Sequence the message ahead of the tool. The explanation must precede the deployment. Once the agent is live, any explanation reads as justification.
Measure the right thing afterwards. Not adoption. Rework volume, parallel-process persistence, and whether cycle time moved. Those are where undermining surfaces.
The cost is a couple of afternoons of leadership time and no licence spend. Against a documented risk that roughly three in ten of your people will quietly work against the rollout, that is the highest-return item on your AI roadmap this quarter, and it is not a technology item at all.
Here is the decision, and it is binary: before the next agent goes live, either your frontline managers can answer "what does this mean for my pay and my scope" in their own words, or they cannot. If they cannot, you are not deploying an AI-driven pay change โ you are deploying an unanswered question, and your people will answer it for you.
