What an AI Change Management Communications Strategy Actually Looks Like - Not a Campaign. A Multi-Year Operating Model.

Here's what usually happens when a company decides it needs an "AI communications strategy."

golden gate bridge

Someone in leadership — usually in HR or corporate communications — gets tasked with building it. They look at what's been done before: the last ERP rollout, the last restructuring, maybe the digital transformation program from three years back. Maybe, if the company is large enough they bring someone in from the Change Management team. They pull out the old playbook. They build a launch campaign. They schedule the town hall. They draft the manager talking points. They send the email from the CEO.

And then, about six months later, the energy disappears. The town halls time slots are swapped out for something more urgent. The email updates get less frequent. The change management budget gets raided. And the program, still officially "in progress", quietly loses altitude while leadership moves on to the next initiative.

I have watched this happen in multiple companies across different industries and different scales of AI integration. The failure mode is almost always the same, and it almost never gets named accurately in the post-mortem.

Side note ->Why does no one ever want to do a post-mortem with me?

It is the answer to all of our problems!

Did no one else watch Greys Anatomy?

The problem isn't the quality of the communications. The problem is that a campaign was built where an operating model was needed.

A campaign has a launch, a peak, and a conclusion. An AI integration - a real one, the kind that actually changes how work gets done across a large enterprise - doesn't have those things. It is culture, organizational change. It has phases, plateaus, setbacks, and re-accelerations that play out over years. Especially if you are building in house - new tools have unexpected delays and engineering has to solve new problems they weren't anticipating. Things happen. The communication architecture has to be built for that reality, not for a product launch.

The 18-Month Wall Is Real, and Almost Nobody Prepares for It

Let me give you a specific pattern I've seen repeat across organizations, because naming it is the first step to surviving it.

In the first six months of a large AI integration, energy is generally high. There's novelty. There's leadership visibility. There's usually real budget. The town halls are well-attended. People are curious, even if anxious. The communication cadence is strong because it has to be the launch is happening and everyone is watching.

Around month seven or eight, something shifts. The novelty fades. The early adopters are using the tools; the majority is still watching. The organizational attention starts moving toward other priorities, a new product, a quarter-end push, a restructuring. Engineering needs more time to solve problems they didn't anticipate and we won't hit that original target date for the soft launch. The AI program is no longer new. It's just there.

By month twelve to eighteen, if the communication architecture hasn't been rebuilt as a standing operating discipline rather than a launch campaign, you hit what I call the wall. Leadership is tired of talking about something they haven't seen fully executed. The program is still running. The tools are still deployed. But nobody is talking about it anymore with any coherence or consistency, and the adoption that was building has plateaued or started sliding back.

The Institute of Internal Communication's State of the Sector survey found that 44% of internal communicators cited change fatigue as one of the biggest barriers to success — the second most significant challenge after limited team capacity.

The organizations that hit the wall and never recover are almost always the ones that treated the initial campaign as the strategy. The ones that survive it are the ones that built something designed to run continuously, adapt based on feedback, and stay alive without requiring a launch event to sustain momentum. (But you should definitely still have a launch event because they are fun and a great way to get people excited and who doesn't love a party?)

AI Is Not a Technology Change. It's an Organizational Behavior Change. That's the Problem.

Before I explain where AI breaks the old model, I want to walk through the change management framework I've used as the primary lens in my integrations, because it's the one that exposes the problem most clearly. Because most internal communications and HR communications people I've met have backgrounds in journalism not organizational psychology or change management.

ADKAR - developed by Prosci and built from research across more than 900 organizations in 59 countries — is an acronym for the five sequential building blocks every individual needs to successfully adopt a change: Awareness, Desire, Knowledge, Ability, and Reinforcement.

The logic of ADKAR is sequential and non-negotiable: you cannot skip stages. Without Awareness, people won't develop Desire. Without Desire, Knowledge doesn't matter because people won't apply it. Without Ability, even motivated and knowledgeable people can't execute. Without Reinforcement, people revert to old habits.

What makes this framework so useful - and so diagnostic - is that it tells you exactly where people are stuck. It's not a soft framework for feelings. It's a precision tool for locating the specific barrier that is blocking adoption for a specific group of people and then deploying the right intervention for that specific barrier.

• If a group of employees has Awareness but no Desire - they understand the change is happening but don't want it - more training won't help. Training addresses Knowledge.

  
  

• The intervention for Desire is different. It involves addressing personal concerns, demonstrating specific benefit to that role, and securing the buy-in that makes people willing participants rather than reluctant compliance cases.

  
  

• If a group has Desire and Knowledge but lacks Ability - they want to use the tool and know the theory but can't execute in their real workflow - more communication won't help. They need hands-on practice embedded in actual work, not classroom instruction.

Prosci's research across 2,600 change practitioners is definitive: projects with excellent change management are seven times more likely to meet their objectives than those with poor change management. Even moving from poor to fair change management triples the likelihood of success.

Seven times. That number alone should settle the debate about whether change management is a support function or a core program driver. It doesn't settle it in most organizations, but it should.

Now here's where AI breaks this model. Not because ADKAR is wrong. But because AI introduces conditions at each stage that no previous technology rollout has ever produced at the same scale and intensity. The framework still applies. But the interventions required at each stage are categorically different and most organizations are applying the old interventions to the new conditions and wondering why nothing works.

Stage One: Awareness - When the Message Lands But Creates the Wrong Understanding

In a standard technology rollout, building Awareness is largely a communications problem. You explain why the change is happening, what it means for the organization, and what employees need to know. Do it clearly, from the right voices, through the right channels, with enough repetition - and Awareness builds.

With AI, the Awareness problem is inverted. Employees already have enormous amounts of awareness. They are aware and they aren't neutral, in fact they are incredibly biased - negatively. They've read the headlines. They've seen the press releases from companies citing AI as the reason for layoffs. They've watched colleagues lose jobs in announcements that framed AI as the cause.

The issue isn't that they don't know AI is coming. The issue is what they know - and what they believe about what it means for them specifically.

Harvard Business Review's research published in early 2026 found that AI initiatives stall because employees' industry-shaped anxiety about relevance, identity, and job security drives surface-level use without real commitment. Leaders who treat AI adoption as a psychological and contextual challenge, not just a technical rollout, are far more likely to convert experimentation into sustained impact. Harvard Business Review

This is the Awareness problem that most AI programs are not solving. They're building awareness of the tool - what it does, how to access it, what the business rationale is. They're not building accurate awareness of what the change means for individual roles, individual careers, and individual futures inside the organization.

And because they're not addressing that question directly and honestly, employees fill in the gap themselves. With fear. With skepticism. With the stories they've read and the colleagues they've watched leave.

Research consistently finds that employees say a lack of clarity around the reason for change makes them most resistant to it. Without that clarity, individuals fill the gaps with their own assumptions. (almost always negative)

With AI, those assumptions are running in a context saturated with layoff news and job displacement anxiety. The assumptions are not irrational. They are informed. And the standard Awareness campaign - a CEO email about transformation and innovation - does almost nothing to correct them.

What changes: Awareness for AI requires explicitly addressing the job security question, by role, from senior leaders, before any other message about the tool lands. Not once. Repeatedly. With specificity about what AI will and won't do in that particular job, in that particular team. Generic reassurance that "AI will augment, not replace" has been heard so many times it no longer registers. The specificity is what builds actual Awareness.

Stage Two: Desire - The Hardest Stage, Now Harder Than It's Ever Been

ADKAR practitioners consistently identify Desire as the most difficult stage to manage. It depends on personal factors - values, fears, professional interests - that vary from person to person and cannot be produced by organizational communication alone.

Research indicates that 68% of change initiatives die at the Desire stage precisely because organizations focus on logical benefits while ignoring the emotional resistance underneath.

There is a scene in the Wolf of Wall Street.

Jordan Belfort's first day and he is having lunch with his boss Mark Hanna (Matthew McConaughey). Jordan suggests that making money for clients is advantageous for everyone and Hanna stops him.

No one knows that the stock market is going to do. The stocks are fiction. The market is unknowable. So what are you actually selling? You're selling the feeling. The dream. The electricity of possibility. You keep the client pumped up, emotionally invested, convinced the next move is the big one - because the moment they stop feeling that, they cash out, it becomes real, and the game is over.

The logic never enters the room. The logic is the thing you use to justify the decision after the emotion has already made it. Hanna's whole lesson - delivered over martinis, and chest thumps (which you will have stuck in your head the rest of the day) - is that the sale lives entirely in the emotional state of the client. Keep the dream alive, keep the feeling running, and the money follows. The moment you try to lead with facts and figures, you've already lost them.

This is why Coca cola sells nostalgia not a beverage. Lexus sells luxury not a vehicle. And why Lululemon sells prestige wellness not yoga pants.

Desire is always personal. It cannot be mandated. It cannot be trained into people. It can only be influenced - by addressing the concerns that stand between a person and their willingness to engage.

In a CRM rollout, the Desire barriers are manageable: the tool might feel like extra work, people might miss the old system, there might be worry about performance visibility. Real concerns, but bounded ones. You can address them with honesty, with demonstrations of benefit, with early wins that make the value tangible.

In an AI integration, the Desire barriers are existential. Employees are being asked to embrace a technology that has been publicly associated with job elimination, that makes their expertise feel less certain, that changes the nature of judgment in their work in ways nobody has fully mapped yet. That is not a normal change management problem. It is an identity-level challenge that the standard Desire playbook was never designed for.

Harvard Business Review's research on organizational barriers to AI adoption found that most firms struggle to capture real value not because the technology fails — but because their people, processes, and politics do. Fear of replacement, rigid workflows, and entrenched power structures quietly derail AI initiatives, even in companies with advanced tools. Harvard Business Review

The uncomfortable truth I've landed on after watching this play out multiple times is this: you cannot build Desire in an environment where the organization has simultaneously been using AI as the stated reason for workforce reductions. Those two messages - "AI is replacing people, so we're cutting teams" and "AI is a tool that will empower you" - cancel each other out. Employees are not confused about which message to believe. They believe the one backed by evidence. The layoffs are evidence. The communications are words.

What changes: Building Desire for AI requires decisions before it requires communications. Decisions about what the organization is actually committing to in terms of workforce protection, skill investment, and role evolution. Those decisions have to come first, and they have to be visible and credible, before any amount of communication about AI's benefits will build genuine Desire in the people you're asking to change. Please go back to the previous blog post to get more into this - because you CANNOT move forward effectively if you don't nail this down first I promise.

Stage Three: Knowledge - Why Training Completion Is Not Knowledge

This is the stage where most organizations invest the most - and measure the least accurately.

I get it. I love data and I love a good dashboard.

But we don't get to green based on how many employees took a training.

Knowledge, in ADKAR terms, means understanding how to change: the skills, the behaviors, the processes required to actually operate differently. Training is the primary intervention here. And most organizations build training programs, deploy them, measure completion rates, and report those numbers as evidence that the Knowledge stage has been addressed.

It hasn't. Completion is not comprehension. Comprehension is not capability.

McKinsey's research on AI upskilling found that training alone rarely drives sustained behavior change. In a study of Microsoft 365 Copilot adoption, nine in ten participants said formal training would be useful — yet seven in ten ignored the onboarding videos entirely, relying instead on trial and error and peer discussion. McKinsey & Company

The format problem is real - but there's a deeper issue. Most AI training programs teach the tool, not the judgment. They show employees how to access the interface, how to structure a prompt, how to interpret an output. What they don't address is the harder, slower, more human question: how does my judgment change when AI is part of my workflow? What do I trust, what do I verify, what does quality look like in an AI-assisted output, and how do I know when the tool is wrong?

Those are not software questions. They are professional judgment questions. And they cannot be answered in a ninety-minute onboarding module.

Prosci's research with 1,107 professionals found that user proficiency challenges account for 38% of all AI implementation difficulties. Technical implementation issues account for only 16%. The Knowledge gap is more than twice the size of the technology gap. Most organizations are investing in the smaller problem. Prosci

What changes: Knowledge for AI requires contextual, role-specific, ongoing learning embedded in real workflows - not a training event with a completion certificate. It also requires building judgment alongside skill: explicit guidance on when to trust AI outputs, how to audit them, and what human oversight looks like in each specific role. The majority of the work is completed AFTER the tool is deployed and the launch party has died down. The organizations succeeding here are building what some practitioners call AI academies - continuous learning communities with peer-to-peer networks, shared experience documentation, and coaching structures that keep learning alive after the launch event ends.

Stage Four: Ability - The Last Mile That Everyone Misses

There is a gap between knowing how to do something and being able to do it under real conditions, with real deadlines, with real scrutiny, while also managing everything else that was already part of the job.

ADKAR calls this the Knowledge-to-Ability gap. It's where most technology change programs quietly fail. Training delivered in isolation, away from real work, builds theoretical Knowledge. Ability requires practice in context — supervised, supported, with feedback loops that help people course-correct when the tool doesn't behave the way the training said it would.

With AI, the ability gap is compounded by a confidence problem I haven't seen at this scale in any previous technology implementation. ManpowerGroup's 2026 Global Talent Barometer found that while AI usage jumped 13% in 2025, confidence in using AI dropped 18% in the same period. "Workers are being handed tools without training, context, or support," said Mara Stefan, ManpowerGroup's VP of Global Insights. Fortune

Adoption is going up while confidence is going down. Those two numbers moving in opposite directions is a signal that what's being measured as adoption is actually compliance - people opening the tool because they're supposed to, not because they feel capable of using it effectively. That is not Ability. That is the performance of Ability in the absence of actual capability development.

What changes: Building Ability for AI requires structured experimentation with psychological safety - environments where trying the tool and failing with it is not a career risk. Prosci's research found that the single strongest predictor of smooth AI implementation is whether the organization strongly encourages employees to try new tools.

Organizations that discourage experimentation are consistently in the struggling group. That encouragement cannot be a poster in the break room or mandated use. It has to be structural: manager behavior, incentive design, and explicit permission from senior leaders to not have all the answers.

Stage Five: Reinforcement — The Stage Nobody Funds

This is where I have seen more AI programs fail than at any other stage. Not because the programs didn't understand Reinforcement's importance — most change practitioners do — but because the resources, attention, and leadership visibility that sustain Reinforcement are almost always the first thing redirected when the next initiative starts competing for attention.

Prosci's research on Reinforcement is direct: reverting to old behaviors is not just a natural tendency — it's a physiological one. The brain is built to favor familiar patterns. Making a change is difficult; sustaining it is harder. The reason Reinforcement fails so often is that once a change goes live, organizations are already moving on to the next one. Prosci

In a CRM or ERP rollout, Reinforcement looks like performance metrics tied to usage, manager coaching on the new workflow, recognition for early adopters, and audits that catch backsliding before it becomes cultural norm. All manageable. All bounded.

For AI, Reinforcement requires something more demanding: it requires that the change never actually ends, because the tool keeps evolving. New capabilities, new use cases, new governance questions, new organizational decisions about where AI applies and where human judgment stays primary. The Reinforcement infrastructure has to be built for a permanently evolving target — not a stable destination that can be locked down with an audit.

This is the place where the difference between a campaign and an operating model becomes most consequential. Without reinforcement systems — regular coaching, recognition, performance integration, and visible leadership support for the new behaviors — regression to previous behaviors is almost inevitable within the first 90 days. isEazy Ninety days. Most AI programs don't even finish their initial rollout in ninety days.

What changes: Reinforcement for AI requires it to be built into standing operating cadences, not bolted on as a post-launch activity. AI usage belongs in performance conversations. AI learning belongs in manager coaching rhythms. Leadership visibility — executives talking publicly and specifically about how they're using AI, what they're learning, what isn't working — is the most powerful reinforcement mechanism available. It costs nothing. Most leaders don't do it.

The One Thing Leadership Needs to Take From This

Every stage of ADKAR looks different with AI than it did with every technology change that came before it. The framework and sequence still work. The interventions are more complex, and the timeline is longer and less predictable.

But the root cause underneath all of it is the same at every stage: AI is being managed as though it's primarily a technology problem, when it's primarily a behavioral one.

The most common barriers to AI adoption are structural and behavioral, not technical. Organizations that embed AI into daily workflows and leadership decision-making dramatically increase long-term adoption and value realization.

The organizations I've watched succeed are not the ones with the best tools or the biggest AI budgets. They're the ones where leadership understood that what they were asking of their workforce was not "please learn new software." They were asking people to change how they think, how they exercise judgment, how they define professional expertise, and what it means to be good at their jobs.

That is not a technology change. That is one of the deepest behavioral changes an organization can ask of its people. And it deserves to be managed with the rigor, the honesty, and the sustained investment that kind of change actually requires.

The Next Post: 4 of 6 → How to Measure Whether AI Adoption Is Actually Working — And Why Almost Every Dashboard You Have Right Now Is Lying to You