Why Employees Hide the AI That Works (and What 95% of Rollouts Get Wrong)
Chris Dyer, named the #1 Leadership Speaker to Follow in 2026 by MSN.com and Inc. Magazine’s #1 Leadership Speaker on Culture, argues that shadow AI is not an employee problem. It’s a rollout problem. Companies spend billions on AI their people won’t touch while those same people hide personal AI they love. The reason has a name, subtraction neglect, and the fix runs in the order most leaders reverse: People first, Process second, Tools third, Technology last. This guide covers why employees hide AI, why most rollouts fail, and two moves any leader can run this week.
Table of Contents
- The thing your team did today
- Two numbers that shouldn’t both be true
- The provisioning gap
- Subtraction neglect: the study that has nothing to do with AI
- The order that actually works: People, Process, Tools, Technology
- What IKEA did instead of layoffs
- Two moves to run this week
- Frequently asked questions
Watch the full breakdown here:
The thing your team did today
Someone on your team used AI today and didn’t tell you. The report you praised on Tuesday was drafted by a chatbot on a personal phone, on the parking lot wifi, and nobody mentioned it.
This is not rare. In more than nine out of ten companies, employees are quietly using AI the company never approved. Over half won’t admit what they use it for. Microsoft went looking at how common this is and found 78 percent of the people using AI at work are bringing their own tools, unsanctioned, off the books.
When I first started hearing this from my clients, I read it the way most leaders do. People fear AI. They don’t understand it. Explain it better and the resistance melts. That reading does not survive the data. Your people are using AI every single day. What they’re avoiding is your version of it.
Sit with the strangeness of that for a second. Companies are spending billions on AI their people won’t touch, while those same people hide AI they can’t stop using. Why would anyone hide the thing that works?
Two numbers that shouldn’t both be true
Start with the official story, the one told on earnings calls. Researchers at MIT went through hundreds of corporate AI deployments and put a number on the gap between the announcement and the reality. After 40 billion dollars of enterprise spending on generative AI, 95 percent of organizations were seeing no measurable return. Ninety-five.
The failures had nothing to do with weak models. The pilots died on brittle workflows and rollouts that ignored how the job actually gets done. Committees. Dashboards. Everything but the work.
Now hold that next to the unofficial version. The same MIT research described a shadow AI economy: employees getting real value from personal AI, off the books, while the official pilot stalled in committee. I watch this happen with clients constantly. The rollout meeting ends, everyone nods, and the real adoption starts in the hallway, on personal phones, unreported. One Fortune 500 insurer reportedly had a sanctioned tool that looked great in the boardroom and collapsed in the field, while employees quietly sped up the same work with chatbots the company had never heard of.
So the technology works. Just not where the company installed it.
The provisioning gap
Before anyone blames the workforce, look at what the workforce was handed. In Microsoft’s data, nearly half of US executives weren’t investing in AI tools for their employees at all, and most people using AI at work had never received a minute of company training.
You know this pattern from the inside. The board sees a demo of the best model on the market. The engineering team gets enterprise seats to build the product. The frontline gets a free tier and a recorded webinar. A license so basic it can summarize an email, as long as the email is short.
That gap has a name now. The provisioning gap. The AI that leadership experiences and the AI employees are given are not the same technology in any sense that matters. But the expectations travel anyway. Be AI-first. Do more with less.
I’ll give you my honest opinion after sitting in these rooms as an advisor. A lot of what gets called an AI strategy is a shiny object with a budget attached. When you ask people to change, and change only ever means do more, they hear you perfectly. They just don’t believe the reason. You can buy the best model on the market, mandate it in every workflow, and still build a rollout where using the official tool feels like punishment and using your own feels like relief.
Your people rejected the deal that came wrapped around the technology. And you wrote the terms.
Subtraction neglect: the study that has nothing to do with AI
In 2021, a team of researchers at the University of Virginia published a series of experiments in the journal Nature that never mentions software. They asked people to improve things. A wobbly Lego structure. An essay. A travel itinerary. Experiment after experiment, people improved by adding. New bricks, new sentences, new stops. Even when removing something was objectively better and cheaper, most people never considered it. The option to subtract barely registered in their minds at all.
Now watch that finding run inside an AI rollout. Leadership wants transformation, so leadership adds. A new tool. New logins. Training modules stacked onto a job that did not shrink to make room. Nothing gets retired. Not a report, not a meeting. And every person on the receiving end does quiet math: same job, plus a learning curve, plus higher output expectations. That equals more.
So they respond rationally. In the official rollout, they keep their heads down, because participation there means addition. Privately, they take relief from a chatbot in their pocket that subtracts an hour from a task with no meetings attached. Then they hide it, because revealing it invites the two outcomes they fear most. More work, or replacement.
This is not a corporate disease. Consumer researchers studying feature fatigue found that shoppers pick the gadget loaded with the most capabilities in the store, then get worn down by those same features at home, where satisfaction collapses. Capability wins the purchase. Usability decides the experience.
I need to own something here. I ran PeopleG2 for two decades. I took it fully remote back when that was considered reckless, and I still made this exact mistake more than once. New software, big launch meeting, me at the front of the room genuinely excited. And I retired nothing. Not one report, not one meeting. I believed I was handing my team a gift. What they received was a heavier job, plus a boss wondering why nobody seemed grateful. Change that only adds is not change. It’s accumulation.
The order that actually works: People, Process, Tools, Technology
Most rollouts run the order backwards, so get it straight.
Technology is the capability itself. AI. Tools are the form it takes in someone’s hands: Claude, ChatGPT, Copilot. Process is where a tool lives in the work, which tasks it touches and what counts as good output now. And people are where all of it survives or dies on contact with an actual Monday.
Most companies buy the technology, pick a tool, skip the process, and inform the people. The order that works is the reverse. People first. Process second. Tools third. Technology last. The technology is the least important decision on that list, which is exactly why it’s the one leaders enjoy making most.
The people part means participation, and the evidence for that is older than most of the companies currently failing at this. In 1948, researchers studying a Virginia pajama factory ran one of the most famous experiments in workplace history. Production changes imposed on workers produced grievances and quitting. The identical changes designed with the workers recovered productivity faster and held. Kotter’s steps, ADKAR, the frameworks you’ve heard of are fine maps. But under every map that works is the same road. The people affected helped design the change.
Now the part leaders don’t enjoy hearing. More explanation is not the fix either. A study in the Journal of Marketing found that people with lower AI literacy are actually more receptive to AI. The less someone understands it, the more it feels like magic. Which means some of your most resistant people are your most AI-literate. They can count precisely what this rollout adds to their week, and what it might subtract from their future. Sometimes the resistance is the most accurate feedback in the building. Not all of it. Some resistance is habit, some is fear, and some ideas deserve to be pushed through anyway. But treat resistance as data before you treat it as defiance.
What IKEA did instead of layoffs
Does the people-first order actually pay, or is it just kinder? Watch this comparison.
Same technology as every failed pilot in this story, a customer service chatbot. Same event, the automation got good fast. In 2021, IKEA’s parent company Ingka launched a bot called Billie, and Billie grew into handling roughly half of all customer inquiries. At that point, every company in the 95 percent knows the next move. Half the workload is automated, so cut headcount to match, book the savings, announce the transformation.
Ingka ran the other play. They studied the half Billie could not resolve and found a pattern inside it. Customers kept asking for something bigger than order status. They wanted help designing their homes. Consultative demand, sitting unanswered in the queue. So instead of severance, Ingka built a retraining program and turned 8,500 call center workers into remote interior design advisers. The transactional work went to the bot. The human work moved up a level. Their CEO called most of the old tasks soul crushing.
That story reads different to me than it does to the consultants, because I’ve lived a small version of it. In 2009, mortgage lenders were a huge share of my client base, and one of them went from hiring thirty people a day to hiring nobody, overnight. Everyone said the responsible move was layoffs. I cut the office lease, the phone lines, everything but the humans. Expenses down 38 percent. Headcount down zero. And the part that still stings, we hid being remote from clients for years, because back then remote meant you weren’t a real company. I know what it’s like to hide the thing that works.
Notice the mechanism running in reverse. Billie subtracted work first. The new role replaced the old one instead of stacking on top of it, with training to match. Process before tools. People before process. The payoff: that remote design channel generated 1.3 billion euros in a single fiscal year, and the department every consultant calls a cost center became a business line. All of it years before MIT measured the 95 percent failure rate. The proof was on the record before the problem had a number.
IKEA did not use AI to shrink the old jobs. It used AI to find better ones.
Two moves to run this week
Do not try to fix all of this next week. Pick one move, the one that made you slightly uncomfortable.
First, run the subtraction audit. Before anything launches, two questions, asked out loud, answered in writing. What does this replace? What are we retiring to make room for it? If the answers are nothing and nothing, you are stacking weight and calling it change.
Second, if enterprise seats for everyone is real money you don’t have, shrink the pilot, not the tool. Pick twenty people who actually do the work, not twenty managers. Give them the real tool and real training. Then take something off their plates on day one, publicly, so the whole company watches change arrive as relief instead of weight. Thirty days later, ask that group what to roll out next. They’ll know.
And one line for your next one-on-one, exact words. “What are you already using that helps? Full amnesty. I’m asking to learn, not to audit.” Then mean the amnesty.
Getting people to use AI was always the easy half. The real job is making it safe for them to stop hiding it. This is the work I take into rooms in my keynote Beyond the Prompt: Building Your Personal AI System, where teams build the operational system that turns scattered, hidden AI use into something the whole company can actually run on.
Somewhere in your company tomorrow, someone will finish a task in twenty minutes that used to take three hours. And they’ll sit on it until the timing looks normal. Start with one question, and mean the amnesty. Start there.
Frequently asked questions
What is shadow AI?
Shadow AI is employees using AI tools the company never approved, usually on personal devices, and not reporting it. Chris Dyer’s position is that shadow AI is a symptom of a rollout problem, not an employee problem. In more than nine out of ten companies it’s already happening, and Microsoft found 78 percent of people using AI at work bring their own tools.
Why do employees hide their AI use at work?
Two fears drive it: more work, or replacement. When a tool saves three hours, revealing it invites a heavier workload or a question about whether the role is still needed. Hiding it is self-defense inside a system that only ever adds. Chris Dyer calls the root cause subtraction neglect: leaders pile on tools and training without retiring anything, so employees keep the shortcut quiet.
Why do most corporate AI rollouts fail?
MIT found that after 40 billion dollars of enterprise spending, 95 percent of organizations saw no measurable return, and the failures traced to brittle workflows and rollouts that ignored how work actually gets done. Chris Dyer adds a second cause. Nothing gets subtracted, so every new tool lands as extra weight, and people quietly opt out of the official version.
What is the right order for an AI rollout?
People first, process second, tools third, technology last. Chris Dyer argues most companies run it backwards: they buy the technology, pick a tool, skip the process, and inform the people. The technology is the least important decision, which is why leaders enjoy making it most.
What is the subtraction audit?
Two questions asked before any rollout, answered in writing. What does this replace? What are we retiring to make room for it? If both answers are nothing, you’re stacking weight and calling it change. It’s the fastest way to keep an AI rollout from becoming one more thing on an already full plate.
Chris Dyer is a keynote speaker on leadership, company culture, and AI at work. He was named the #1 Leadership Speaker to Follow in 2026 by MSN.com and Inc. Magazine’s #1 Leadership Speaker on Culture, and ranks #15 on the Global Gurus Top 30 for Organizational Culture. He is a five-time Inc. 5000 CEO and the author of four books, including Moments That Matter (March 2026).
To bring this conversation to your team, visit chrisdyer.com. For a free companion workbook, go to chrisdyer.com/moments.
