When an operator declines an automation effort, the reason given is rarely technical. It is that the people will resist it, that adoption will fail, that the floor will quietly route around whatever gets built. The concern deserves a serious answer rather than reassurance, because part of it is correct and part of it rests on a number that does not survive inspection. Sorting the two is the leadership task: the fear in the workforce is real and measurable, while the statistic most often cited to justify giving up because of it is not.
What the workforce actually feels
The anxiety is well documented and it is high. EY's AI Anxiety in Business survey, which polled a thousand US office workers in October 2023, found that seventy-five percent are concerned AI will make certain jobs obsolete and sixty-five percent are anxious about AI replacing their own job. That is a majority reporting a personal economic threat, not a vague unease about technology in general. From the leadership side, Kyndryl's 2025 survey found that forty-five percent of chief executives say their employees are reluctant or hostile toward AI adoption. Both halves of the picture agree: a large share of workers are afraid, and nearly half of the leaders watching them already see that fear expressed as resistance on the ground.
Those figures describe a state of mind rather than a forecast of outcomes, which is why they are worth taking at face value. They tell you the starting conditions of any automation effort inside a mid-market firm. They do not tell you what happens next, and the gap between the two is where most of the reasoning about this goes wrong.
The number that does not hold up
The forecast usually attached to the fear is the claim that seventy percent of change initiatives fail. It appears in board decks, vendor pitches, and the internal case against trying, repeated often enough that it has acquired the texture of established fact. It is not one. Mark Hughes traced the figure through its citation chain in the Journal of Change Management in 2011 and found that the trail never reaches valid empirical evidence. Each source citing the seventy percent number points to another source, which points to another, and the chain either loops back on itself or ends at an assertion with no study behind it. The figure persists as folklore. It has never functioned as a measurement.
This matters for the leadership decision in a precise way. If seventy percent of change efforts genuinely failed, the rational posture would be caution bordering on refusal. Because the figure has no empirical basis, it cannot carry that weight. You can use it as a narrative about how hard change is, which is fair, but you cannot use it as evidence that yours will fail, because no one has established the base rate it claims to report.
Where the failures actually come from
Stripping out the folklore does not mean AI efforts reliably succeed. BCG's work on AI-project outcomes finds that a large majority fail to deliver the benefits expected of them, and that the dominant barriers are organizational rather than technical. The model is rarely the problem. The problem is that the thing built never gets used, never gets owned, never gets wired into how the work is actually done, which is the same failure mode the change-management research has described for decades, now wearing an AI label.
The more reliable research points at specific causes rather than a single doom statistic. Resistance correlates with two things in particular: fear about job security, which the EY figures already quantify, and weak communication and sponsorship from the top. The corollary is the useful part. Involvement, clear communication, and visible senior ownership measurably improve adoption. Prosci's change-management research attributes a meaningful lift in success rates to active and visible senior sponsorship specifically, meaning not the existence of a change-management plan on paper but leaders who are seen to back the work. The driver of resistance and the lever against it sit on the same axis: the fear is about security and about being kept in the dark, and the remedy is security and being kept informed.
The design choice that answers the fear
This is where the shape of the redesign stops being a technical detail and becomes the most important thing a leader communicates. An automation effort can be built to confirm the workforce's fear or to address it, and the two look different from the first week.
flowscope has written about the operator's role after the work is mechanized, in the operator's role after automation: the agent takes the mechanical majority of a process and routes the small set of cases that genuinely need judgment back to a person, who handles the exceptions and owns the calls the machine should not make. When a redesign is built that way, the message to the floor is that the repetitive part is going and the part that required them to be there in the first place, the judgment, is what remains and gets more of their time. That is a claim the employee can verify within weeks by watching what actually lands on their queue.
The way the work is observed in the first place carries the same signal. flowscope has described the difference in shadowing instead of surveillance: capture is framed as learning how the work is done, with the people doing it as the source of record rather than the target of monitoring. A redesign that begins by asking the experienced clerk how the process really runs, including the workarounds that never made it into the documented procedure, treats that person as the authority on their own work. That is the reverse of the experience that produces the seventy-five-percent fear.
What this asks of leadership
The honest version of the objection is not that people will resist, full stop. It is that people will resist if the effort is done in a way that earns resistance: built in secret, communicated as a cost program, aimed at headcount, with sponsorship that evaporates the moment it gets hard. The research says the fear is high at the start, that the most-quoted reason to quit is unfounded, and that the variable under leadership control is whether the work is visibly owned and honestly explained. None of that is a guarantee. It does mean the outcome turns on choices the leader makes, not on a fixed failure rate the workforce imposes.
A reasonable counter is that this puts too much faith in communication, that some workforces will resist any change regardless of how it is framed, and that involvement and transparency are easy to promise and hard to sustain across a year. There is real truth in that, and a redesign aimed at jobs rather than at the mechanical work will be seen for what it is no matter how it is described. But the failures BCG documents are concentrated in efforts that were never owned and never used, and the choice that decides which side of that line an effort falls on is made by leadership at the start, not by the floor at the end. The place to apply that judgment, including the cases where the right answer is not to automate at all, is covered in when not to automate.