Prosci published a post on 22 August with a title that gives away the argument: “Change Fatigue in the Age of AI: The Half of Saturation Nobody Measures” (Prosci, 2026). It contains no statistics of any kind, which for a change management post is unusual enough to be worth noticing. What it has instead is an inequality.

Change disruption > change capacity = change saturation.

Everything the industry sells about change fatigue works on the left-hand term, which is disruption. Slowing the roadmap reduces disruption, and so does hunting for the collisions where three programs hit one team in a single quarter. Neither of those disruption levers touches capacity, which Prosci defines as “how well the organization can absorb those changes”.

The strongest claim in the post is Prosci’s own and deserves the credit: “the issue is not that sequencing stopped working; it’s that AI makes fewer changes sequenceable.” An organisation does not choose when a model provider ships a major new capability, and cannot defer it to the third quarter. “If you don’t control when all the disruptions arrive, you can’t fully control the sequence.” Which makes “do less” an incomplete strategy.

Sit with that, because it invalidates a product line. Heat maps and sequencing workshops are disruption instruments, and if the disruption lever is running out of travel, the only thing left to work on is capacity, which nobody measures.

Four figures appear together in every change fatigue deck: 71 per cent of employees overwhelmed by the amount of change during the pandemic, 54 per cent of fatigued employees considering a new job, 48 per cent more tired or stressed, and 37 per cent trusting their employer less. All four get attributed to Capterra’s 2022 Change Fatigue Survey, 970 US employees fielded in March 2022 (Westfall, 2022).

Three of them are in it. The 37 per cent is not.

I know that because it was in my own notes for this piece, sourced the way everybody sources it. Checking meant reading every percentage on the Capterra page, and there are twenty-one. None of them is 37. None is 38 either, which matters because a companion figure of 38 per cent enjoying their jobs less usually travels beside it. The one sentence about trust in the whole survey has no number attached to it: “Too much backtracking can decrease employee trust that company leaders know what they’re doing.”

That 37 per cent has an ecosystem of its own now. The trust figure turns up across statistics roundups and consultancy blogs as some version of “37% feel less trust toward their employers after workplace changes”, credited every time to a survey that quantifies trust nowhere. Nobody invented the trust figure maliciously. Somebody paraphrased an unquantified sentence about trust into a number once, and the citation chain did the rest.

The 73 per cent makes a better example, because there are three of them. Prosci reports that “more than 73%” of respondents in its Best Practices in Change Management research described their organisations as approaching or past the saturation point (Prosci, n.d.). Gartner has one about employees affected by change facing moderate to high stress. Gartner has a second, from 473 HR leaders surveyed in July 2024, about employees being fatigued by change. Three numbers measuring three different things, stacked on each other in slide decks constantly.

Even the Prosci figure will not hold still. The benchmarking is genuinely theirs and has tracked saturation since 2007, which is the impressive part. But no page carrying the number names its edition or year, and Prosci has published 78 per cent for the same measure under the same “most recent study” wording. The honest version says roughly three-quarters of respondents have sat at or past saturation across editions spanning two decades, which beats the round figure, because the durability is the point.

The 53 per cent who say too much change is happening at once belongs to PwC’s 2024 Global Workforce Hopes and Fears Survey of more than 56,000 workers, whose own page never prints that figure, saying only “more than half” (PwC, 2024). The decimal precision arrived in trade coverage.

A field that cannot keep its four best-known figures attached to the right surveys does not have a measurement problem confined to capacity. It has one everywhere. Capacity is simply where the absence shows, because there was never a number to launder in the first place.

What I did not expect to find, while checking all of that, is that Australia already measures capacity. Every year, of about 150,000 people, at no cost to anyone who wants to use it.

The APS Employee Census contains a communication and change block, and one of its items is “Change is managed well in my agency”. In 2025 it went to 186,849 public servants and 151,771 answered, an 81 per cent response rate, and 48 per cent agreed (APSC, 2025). Two questions sit beside it in the same block. Fifty-two per cent said staff are consulted about change at work, and 67 per cent reported that the impacts of change are communicated well within their own workgroup. Consultation scores lowest of the three, which is its own quiet finding.

That question does not ask how many changes are running, or whether the portfolio is sequenced. It asks people to assess the quality of what they have already been through. That is what Prosci names as the input to capacity: an organisation’s history with change, and what previous experiences have taught people to expect from the next one. Prosci Principal Change Advisor Evelyn Williams defines change fatigue as being “at least one change project past their capacity”. The census asks about the projects that came before that one.

The spread underneath that national figure is wide. In 2025 the Commission on Safety and Quality in Health Care recorded 29 per cent on the same question, and the ACCC 59 per cent: a thirty-point gap between two Commonwealth bodies operating under identical policy settings.

So the measure the sector says does not exist has existed here for years, with a sample no vendor survey approaches. It gets used as a report card, never as a sequencing input, which is what a capacity measure is for.

Some of that is the cycle the census runs on. Fielded in May and June and published in September, the census reports its result well after the decisions it should have informed. Some of it is granularity, because an agency-level census figure tells a deputy secretary something and tells a program director almost nothing. And some of it is that the census number gets filed under engagement instead of delivery, so the people building the roadmap never look at it.

The timing is not kind either. Under the AI Plan for the APS 2025, non-corporate Commonwealth entities were required to appoint a Chief AI Officer by July 2026, and Finance’s implementation guidance says the appointee comes from the existing senior leadership team (Department of Finance, 2026). Of 106 entities required to appoint one, all but two met that deadline. GovAI Chat entered beta on 27 July. Agencies must implement AI impact assessments for in-scope use cases by 15 December 2026 (DTA, 2026).

Three changes to how public servants work, inside eight months, arriving on the same senior people, one of whom now holds a new title that reportedly carries no additional pay. That last detail is a capacity decision made explicitly, and nobody appears to have measured what it spent.

Meanwhile people are already ahead of the program. SAP’s Value of AI 2026 research with Oxford Economics, covering 2,600 leaders across 13 countries, found that 77 per cent of Australian organisations report shadow AI occurring at least occasionally, and that 42 per cent say they are deploying agents faster than they can standardise and govern them, rising to 68 per cent once the unsure are included. Only 22 per cent rate their skills and expertise as mostly or fully ready for AI governance, against 33 per cent globally (SAP, 2026). It is vendor research and should be read as such.

Seventy-seven per cent measures the distance between the pace of the sanctioned program and the pace people already work at, which makes it a capacity reading taken by accident, and not a staff discipline problem.

The institutional reflex to a number like that is monitoring, and monitoring is not free. Chen (2026) argues that AI-enabled workplace surveillance alters control qualitatively, moving it from observable conduct to inferred states, against legal controls that arrive mostly after the fact. Responding to a capacity signal by inferring more about people spends the very thing you were protecting.

Consultants are paid to deliver initiatives. We are almost never paid to reduce their number, and never to repair the credibility damage from the last three. The incentive runs against the diagnosis, and Prosci sits inside that same incentive, since its post closes by promoting the company’s own enterprise training program. The argument is still right, and I write it as somebody who sells change work.

The measure I would test costs nothing and nobody can sell it to you. Ask a team what proportion of their last five change initiatives delivered what was promised, and have the team answer rather than the PMO. Five is enough to catch a pattern and short enough that people remember. It sits on no dashboard, it takes about ten minutes, and it would alter a sequencing decision immediately, because a team at two out of five is not a team you schedule another launch onto.

The census question is that same measure at national scale, asked annually and sitting in a published PDF. Forty-eight per cent of the public servants counted in 2025 said change is managed well where they work. We have known that number for years and kept sequencing off the volume number instead, because volume is the easier thing to count. That is how the easy count became the only one.

References