What to Do in the Next 90 Days: A Practical Reset for AI Programs That Are Measuring the Wrong Things and Managing the Wrong Way

rmclements10

Mar 30

12 min read

Yay! You made it through 5 very tough and humbling challenges.

If you're 24 months into an AI integration and everything is going sideways. That's okay. If it's month 3 and you've already done everything wrong, that's okay too. Because we are all building the plane as we learn to fly it (as they say).

If you've read the previous five posts in this series, you have a reasonably complete picture of what's actually wrong with most enterprise AI programs.

The technology isn't the problem. The measurement is measuring the wrong things. The change management is built for a different kind of change. The communications strategy is producing compliance theater and calling it adoption. The internal communications leader was brought in too late and handed the wrong job.

And underneath all of it, the employees who are supposed to be doing something different with AI are mostly doing what people always do when they're anxious and uncertain about something that feels like a threat: the minimum required to avoid scrutiny, and no more.

You know all of this. Many of you knew some version of it before you read a word I wrote.

The harder question is what to do about it when you're already eighteen months in, when you have a board asking for ROI evidence, when your dashboard is green and your behavior change is flat, and when the people who made the original decisions are not especially interested in being told those decisions were built on the wrong foundation.

That's the conversation this post is actually about.

First: The Honest Assessment Nobody Wants to Have

Before any of the tactical steps matter, something harder has to happen. Leadership has to be willing to look at the program honestly.

Not the dashboard version of honest - the version where you audit the metrics, find a few places where the numbers could be higher, and call that a program review.

The actual version, where you ask the questions that the dashboard was never designed to answer.

New research shows that AI initiatives often 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

The questions that force that honest conversation are these:

Do you know what your employees actually believe about AI?
Not what they said in the survey that their manager saw, but what they believe privately about whether AI threatens their role, whether leadership is being straight with them about the workforce implications, and whether using AI will help or hurt their career at this company? Not as a general read, by role, by function, by tenure, by geography?
Do you know what behavior has actually changed?
Not who logged in and how often, but whether the way people approach specific tasks has measurably shifted from the baseline you established before the program launched? (And: did you establish a baseline before the program launched?)
Do you know why it isn't working in the places where it isn't working?
Not the assumption, what the data says. Is it an Awareness problem, where people genuinely don't understand what they're supposed to do differently? Is it a Desire problem, where they understand but don't want to? Is it a Knowledge problem, where they want to but don't feel equipped? Is it an Ability problem, where the training hasn't translated into actual proficiency? Or is it a Reinforcement problem, where the behavior starts and then reverts because the system doesn't reward it?

Each of those is a different problem. Each requires a different intervention. And you cannot know which one you have without a diagnostic that most programs have never run.

Without strong change management, organizations settle for cosmetic adoption: licenses distributed, tools in use, yet no actual change in how work gets done. That's when "regret spend" shows up - money poured into pilots that never scale. Leaders who measure success by license counts will miss the real picture. Fast Company

The honest assessment is the prerequisite. Everything else in this post assumes you've done it.

I know it's hard to have these conversations, Its hard to admit you might have made a mistake. But just because you're a few stops in the wrong direction doesn't mean you need to take the train to the end of the line. You can get off now. Better a few wasted months than a few wasted years.

The 90-Day Reset - Not a Relaunch

I wholeheartedly believe in re-sets. Starting over.

I love re-launches. I love January 1st and Day 1's of a new habit challanege. But within hours or maybe a few days when it's raining and I don't want to go for a morning run or I have an unexpected event and don't have time to meditate or I am traveling and that diet goes out the window and the entire plan blows up.

Never to be mentioned or touched again.

So instead, I try to stick to re-starts.

Re-starts don't need a Monday. They don't need a new month.

Re-starts can start any time any day.

They don't need fanfare or a celebration or a social media post.

Just the small, quiet decision to remember why I wanted this in the first place and the willinness to try again.

The wrong approach to this is a relaunch. New campaign. Fresh energy. Recommitment messaging from leadership. Town hall with renewed momentum. And then - three months later - the same flat behavior change, now compounded by employee cynicism about the fact that leadership launched this thing twice and it still isn't working.

The right version is quieter and harder. It's a diagnostic, a redesign of the interventions that are failing, and a shift in what gets measured and who is accountable for it. It doesn't require a big communications moment. It requires genuine organizational honesty about what the program was missing and a structured plan to address the actual barrier — not the assumed one.

Here's the framework, organized by the most critical decisions and actions in each phase.

Days 1 through 30: Stop Adding and Start Listening

The instinct when an AI program is underperforming is to add. Add more communications. Add another training module. Add a refreshed campaign. Add a town hall.

Resist it. The first thirty days of a reset are about stopping the additions and doing the diagnostic work that should have happened at the start.

Run a barrier audit by role.
This is the most important thing you can do and almost nobody does it. Go talk to people — not in a focus group format that signals "leadership wants to know what you think" and produces polished responses, but in real conversations with real employees doing real jobs. What is actually making it hard for them to use AI in the way the program envisions? Is it a training problem? A workflow problem? A trust problem? A job security problem? A manager problem, where their direct leader isn't modeling AI use and is actually, quietly, creating a signal that it's optional? You cannot fix the right problem until you know which problem you have, and you cannot know which problem you have without asking people who are not performing for an audience.
Segment your employee population by stage of change, not by job title.
Some portion of your workforce is at Awareness — they still don't clearly understand what AI means for their specific role. Some is at Desire — they understand but aren't convinced it's worth the investment of their effort. Some is at Knowledge — they want to but feel undertrained. Some is at Ability — they've been through training but can't yet perform the skills confidently in real work. And some is already at Reinforcement — they've adopted and are at risk of reverting without ongoing support. These are four completely different intervention requirements, and treating them with the same communications campaign will fail every one of them.
Audit your baseline.
This is painful if you didn't establish one before launch, but it's not too late. Identify the specific behaviors the program was designed to change. Find whatever proxy measures for the pre-AI state you can — historical data on time-on-task, output volume, quality scores, whatever is available by role. It won't be perfect. It will be significantly better than nothing, and it will give you the before-state you need to measure from.
Stop the metrics that are misleading leadership.
If your weekly update to the CHRO or COO leads with login rates and training completion percentages, have the conversation about what those numbers actually mean — and don't mean. This is uncomfortable. It is significantly less uncomfortable than having the ROI conversation with the board in six months with no better data.

Days 31 through 60: Redesign the Interventions That Are Actually Failing

Once you know which stage of change is the actual barrier - by role, by function, by segment of the workforce - you can stop running generic interventions and start designing targeted ones.

For Awareness failures:
The problem is almost never that people haven't heard about the program. It's that they don't understand what it means for them specifically. Generic AI messaging doesn't fix a role-specific understanding problem. The intervention is role-specific, concrete, and honest. Here is what changes about how a person in your exact job will work, here is what AI will take off your plate, here is what you'll do differently, and here is what we are not asking you to do. Specificity. Vague is the enemy.
For Desire failures:
This is the hardest stage and the one most programs underestimate. Creating foundational trust in AI use throughout the organization is essential. If employees don't trust AI output, they won't trust the decisions it makes — and the technology will have little chance of attaining scale. When companies invest in building trust in AI and digital technologies, they are nearly two times more likely to see revenue growth rates of 10% or higher. McKinsey & Company
The Desire barrier is almost always a trust problem underneath. Trust in whether AI actually works for their specific job. Trust in whether leadership is being honest about the workforce implications. Trust in whether investing their effort in this program will benefit them personally or just make their role easier to automate. Each of those trust barriers requires a different response, and none of them are resolved by a positive-framing communications campaign. They require direct, honest, specific conversation and visible leadership behavior that makes the stated intentions credible.
For Knowledge failures:
The most common form is training that covers the tool but not the workflow. Employees can complete the module and still have no idea how to use the capability in the actual context of their job, on the actual tasks they do daily. The fix is role-specific, workflow-embedded practice, not more modules. Find the superusers in each function who have figured out how AI integrates with their specific work and build peer learning structures around them. McKinsey's research shows the most enthusiastic AI adopters are millennial managers 62% of employees aged 35 to 44 report high levels of AI expertise. These employees, identified and supported, can become powerful change agents who mentor their peers and lead practice groups.
For Ability failures:
People know what to do but can't yet do it reliably. This requires practice and feedback, not more information. The intervention is structured opportunities to use AI skills on real work with someone available to help when it goes wrong - sandbox environments, paired practice, micro-coaching from peers. Time-to-proficiency is the metric, and it needs to be tracked by role.
For Reinforcement failures:
This is where the system fights the change. People adopted the behavior and then stopped. Because their manager doesn't use AI themselves so there's no modeling, because the performance management system doesn't reward it, because the workflow wasn't actually redesigned so using AI feels like additional work rather than better work. Reinforcement failures require structural changes, not communications. If the system doesn't reward the behavior, the behavior will not persist. That is not a people problem. It's an organizational design problem.

Days 61 through 90: Rebuild What Gets Measured and Who Gets Held Accountable

By day sixty you have a diagnostic, a segmented understanding of where the actual barriers are, and redesigned interventions targeting those specific barriers. The third phase is about changing the accountability architecture so the progress you're building doesn't quietly erode when attention moves on.

Redefine success metrics in leadership conversations.
This is the moment to replace the activity dashboard with the behavioral change framework from Post 4. Not all at once, not with a dramatic announcement, but in the ongoing rhythm of program updates: this is what we're now tracking, this is what it tells us, and this is why it's a better read on whether the program is working than the numbers we were using before.
Establish role-specific proficiency benchmarks.
For each priority role, define what demonstrated proficiency looks like at 30, 60, and 90 days post-intervention. Not training completion, actual proficiency. The ability to perform a specific AI-assisted task to a defined quality standard. This gives managers something concrete to observe and support, and it gives the program a leading indicator that predicts business outcome impact before it shows up in the lagging metrics.
Build the listening infrastructure that continues beyond the reset.
The barrier audit you ran in days one through thirty shouldn't be a one-time event. It should be the beginning of a continuous read on where the program is actually landing — fear-to-curiosity ratio by role, proficiency confidence, trust in leadership's stated AI intentions, behavioral retention at 30/60/90 days. This infrastructure exists in almost no enterprise AI program right now. Building it is what separates a program that catches adoption failure early from one that discovers it when the board asks for the ROI report.
Make manager enablement a first-class deliverable.
If the manager layer isn't bought in, using AI themselves, and equipped to have honest conversations with their teams about what AI means for their specific roles, nothing else you've built will hold. The 90-day reset isn't complete without a clear picture of manager AI adoption and a plan for the managers who are still the quiet signal that AI is optional.

What This Reset Is Really Asking Leadership to Do

I've described this as a practical framework, and it is. But underneath the tactics is something harder.

The reset is asking leadership to accept that the way they originally approached the program - as a technology deployment with a communications layer - was the wrong model for the kind of change AI actually requires. That's not a comfortable thing to accept, especially when significant investment has already been made and the board is watching.

Most firms struggle to capture real value from AI not because the technology fails — but because their people, processes, and politics do. Organizations that redesign incentives, workflows, and governance to align human behavior with technological capability don't just adopt AI — they transform how value is created across the enterprise. Harvard Business Review

The companies that do this reset well are not the ones with the most sophisticated AI tools or the largest budgets. They're the ones where leadership is willing to ask the questions that the dashboard doesn't answer, to hear what employees are actually experiencing, and to redesign the program around what's actually needed rather than defending what was originally built.

That kind of organizational honesty is rarer than it should be. When it exists, it tends to correlate strongly with the programs that eventually work.

What This Series Has Been About

What I've tried to do across these six posts is build the argument that most enterprise AI programs are failing not because the technology doesn't work but because leadership is managing a behavioral transformation like a software deployment. And that the specific functions responsible for changing that - internal communications, change management, people strategy - are being used in versions of themselves that are about ten years out of date.

The argument isn't that the technology doesn't matter. It matters enormously. It's that the organizations getting real, sustained, measurable value from AI are the ones treating the human side of the transformation with the same rigor, the same investment, and the same accountability structure as the technical side.

The posts have covered the full arc of what that actually requires.

Post 1 named the problem: most companies aren't getting a return on their AI investment, and the part nobody talks about is that the failure is behavioral, not technical.

Post 2 named why employee buy-in is failing and what leadership is doing wrong — the assumptions they're making about employee enthusiasm that don't match what employees are actually experiencing.

Post 3 applied a real change management framework to AI specifically — why ADKAR looks different when the change has no defined end state, when the fear is identity-level rather than task-level, and when reinforcement has to be designed for a program that never truly finishes.

Post 4 showed how to measure whether the program is actually working — what the dashboard is hiding, what a real behavioral change metric looks like, and what the three questions are that every metric has to pass before it goes in front of leadership.

Post 5 made the case for what the internal communications leader's role in an AI transformation actually is — and why the version most companies are running is too narrow, too late, and too focused on output rather than outcome to produce what the program needs.

And this post is the practical close: what to do in the next 90 days if you're reading this and recognizing your program in it.

The One Thing to Do Before You Close This Post

If there's one action I'd ask you to take in the next 48 hours, it's this: have a conversation with three employees who are in the target population for your AI program - not in a structured focus group, not with their manager in the room, not with a survey - and ask them one question. Not "are you using the tools" and not "how do you feel about AI."

>> Ask them: "What would it take for AI to actually make your job better?"<<

Listen to the answers. Don't defend the program. Don't explain the training resources.

Don't redirect to the official messaging. Just listen.

What you hear will tell you more about why the program is or isn't working than any dashboard you have access to. And it will tell you exactly where to start.

This is the sixth and final post in a series on AI integration as organizational change management. I'm sure I'll learn a new lesson this week. If any of this landed - or if you're in the middle of a program that sounds like what I've been describing - I'd be glad to hear about it.