“I’ve reduced it from 9,000 heads to about 5,000, because I need less heads.”
Marc Benioff, Salesforce’s chief executive, said it on a podcast at the end of August 2025. He was talking about customer support. A few days later the company said that “we no longer need to actively backfill support engineer roles.” CNBC’s headline called it layoffs.
In April 2026, a company post told a different story. Salesforce, it said, had redeployed “hundreds” of support engineers into fast-growing parts of the business. By Salesforce’s own account, it had handled the training side well: people trained during work hours, and were already certified when the new jobs opened. None of the accounts says how many of the roughly 4,000 moved.
Then the post zooms in on a single move. An engineer who went from support into contract renewals was, by the company’s account, uncomfortable at first.
Training on company time and certificates in hand before the jobs opened are things a plan can schedule. Whether a person wants to go is a different hurdle.
Five hurdles, in order
A change management firm called Prosci has a model built for exactly that gap. According to Prosci, Jeff Hiatt developed it in 1996, first published it in a 1999 white paper and laid it out in a 2006 book, ADKAR. The name lists five things Prosci says a person needs, in this order: Awareness of the need for change, Desire to take part in it, Knowledge of how to change, Ability to actually do the new work, and Reinforcement so it sticks.
The model’s central claim is about sequence. ADKAR splits change into five hurdles and says to start with the first one a person missed. Prosci’s own wording is to “focus support on the earliest unmet element.” A person who has cleared Awareness but not Desire is stuck at Desire, however many courses they’ve finished.
Why the classroom can’t fix a “no”
Training is the tempting place to start, since it is easy to schedule and easy to count. Prosci calls starting there “a common mistake.” Its chief innovation officer, Tim Creasey, puts it more sharply: “training without the preceding awareness and desire is ineffective and can actually be more frustrating than beneficial for employees.”
Piano lessons work the same way. Book them for a child who doesn’t want to play and you buy an hour of scales a week, not a pianist. The lessons aren’t bad. They are aimed at the wrong hurdle.
Desire is often the hardest, in Prosci’s telling, because it “requires a personal decision.” A manager can explain what the change means for the person and answer “What’s in it for me?” A manager can’t make the decision for them.
Salesforce’s post doesn’t use ADKAR’s terms, so reading its story through the model is an interpretation, not the company’s own account. But the details line up. The post says the renewals job meant a new vocabulary, from APIs and SQL queries to uplift pricing and revenue targets. The new team’s director built extended training with practice before the full workload, and a manager kept up “ongoing reassurance.” By the company’s account, the engineer soon embraced the move. It is a company success story, though, and it can’t show what would have happened without the reassurance.
A consultancy’s checklist, not a law
ADKAR comes from a firm that sells training in it, including a certification program that teaches it, and much of the case for the model comes from that firm too. Prosci says the 2006 book followed “nearly a decade of research by Prosci,” and it cites its own benchmarking surveys. Its website calls ADKAR “a proven approach.”
Independent tests are scarce. A 2021 systematic review led by Reema Harrison looked at 38 studies of change-management models in healthcare. Kotter’s model appeared in 19 and Lewin’s in 11; Prosci’s ADKAR appeared in one, a case study. The reviewers added that the studies were often case examples with little detail on methods. A 2026 review of 17 studies on virtual care found ADKAR in one, and “limited evidence about how the use of such models influenced the success of change management beyond providing a roadmap.” It also listed “linear application of models” among the common barriers.
Neither review reports a controlled test of ADKAR’s core claim, that a person stalls at the first missing element. Prosci’s own FAQ says the elements are “sequential but not linear.” So the model works best as a set of questions to ask about one person, not as proof of what will happen. Other pages here take different angles: Kotter’s eight steps work at the level of the whole organization, and learning anxiety looks at the fear of not being able to learn.
The whiteboard by the loading dock
Imagine a warehouse supervisor whose company is replacing the paper rota with a scheduling app. She has been to two training sessions and handles the app fine in the demo. Back on the floor, she still draws the week on the whiteboard by the loading dock, then types it into the app on Friday.
A third training session is the obvious fix, and it’s aimed at Knowledge, a hurdle she has already cleared. Ask her instead what the board does that the app doesn’t. If the answer is that her team checks it on the way in, the stall is at Desire, maybe even at Awareness of why the app matters to anyone beyond head office. That calls for a different conversation, perhaps a screen by the dock.
Book the third session without asking, and the whiteboard will probably stay up.