Free AI ROI Calculator for PE Firms
Back to Blog
Industry InsightsNovember 16, 2026

We've Run AI Change Management at 40+ Companies. Here Are 5 Mistakes to Avoid.

Send to
Send to

Key Takeaways

  • The people side is the hard part. BCG's research puts about 70% of AI implementation challenges in people and process, and only 10% in the algorithms.
  • Every role needs its own reason. A company-wide announcement tells people a tool exists. It doesn't tell anyone what changes in their week.
  • Champions spread adoption farther than memos. Peers closest to the work carry the message better than leadership can.
  • Train on real work, and give it real time. Regular use is sharply higher for employees who get at least five hours of training, and only a third say they've been properly trained.
  • Measure adoption from week one. Usage data and feedback show where AI is sticking and where people need help, before momentum fades.

Here's what doesn't add up. The models keep getting better every quarter, and most companies still can't point to what AI has done for them. In BCG's 2024 survey of 1,000 CxOs and senior executives, 74% of companies had yet to show tangible value from AI. The same research split the blame. Around 70% of the challenges came from people and process. About 20% came from technology. Only 10% came from the AI algorithms themselves.

So the part everyone spends the most time picking (the model, the vendor, the architecture) is the part that causes the fewest problems.

That matches what we see. Getting the system working is rarely what holds a rollout back. People are. A rollout can have a great model, clean data and a working integration, and six weeks later half the team is back to the old spreadsheet. Nothing broke. People just never changed how they work.

I think of it this way: every AI rollout ships two things. One is the system. The other is a new habit for the people who are supposed to use it. Most project plans have a line item for the first one and nothing for the second.

The mistakes below are what happens when the habit has no owner. None of them are exotic. That's the point. They're easy to make because they feel like reasonable shortcuts at the time.

Mistake 1: Shipping the Tool Before the Reason

The most common launch email says some version of "We're excited to announce that everyone now has access to [tool]." Then a link. Then silence.

That email answers a question nobody asked. People don't wonder whether a tool exists. They wonder what changes for them. Does this replace part of my job? Am I expected to use it? What happens if I don't? Who do I ask when it gives me a wrong answer?

When those questions go unanswered, people fill in the blanks themselves, and the answers they invent are rarely generous. Some ignore the tool. Some quietly assume it's a headcount play. A few try it once on a hard problem, get a mediocre result, and decide it doesn't work.

Write the "what changes for me" line for every role

Before anything rolls out, write one or two sentences per role. What this team will do differently, what stays the same, and how success will be measured. Finance gets a different line than field operations. Managers get a different line than the people they manage. If you can't write that sentence for a role, that role isn't ready to launch. Better to find that out before launch day.

Mistake 2: Rolling Out From the Top Down Instead of Through Champions

Leadership sponsorship matters. It matters a lot, actually. In BCG's AI at Work 2025 survey of more than 10,600 leaders, managers and frontline employees, the share of employees who feel positive about GenAI rises from 15% to 55% with strong leadership support. Only about a quarter of frontline employees say they get that support.

But sponsorship and rollout are different jobs. A sponsor makes it clear the change matters. The rollout itself spreads person to person, and it spreads fastest through people who already do the work.

A memo from the COO saying "use AI more" changes very little. A colleague on your team showing you how she uses it on the weekly report you both dread changes a lot. Same message, very different messenger.

Pick champions from the people closest to the work

Find one or two people on each team who are curious, respected by their peers and close to the daily workflow. Give them early access, a short playbook and a direct line to whoever is building the system. Let them shape the workflows and test early versions. They'll find the rough edges faster than any QA plan, and when the tool launches, they're the ones their teammates ask.

Mistake 3: Training People on Features Instead of Their Own Work

Most AI training I've seen is a tour of buttons. Here's the chat box. Here's how to upload a file. Here's a prompt that writes a haiku. People nod, go back to their desks and don't use it, because nothing in the session looked like their actual job.

The BCG data is blunt on this. Regular usage is sharply higher for employees who get at least five hours of training, and only one-third of employees say they've been properly trained. Five hours is not a lot. Most rollouts don't come close.

(I'd add a caveat here. Hours alone won't fix it. Five hours of generic demos is still five hours of generic demos.)

Build sessions around one real workflow per team

Pick the workflow each team does every week and train on that, with their real documents and data. The goal is for people to leave knowing exactly where AI fits into their week, with something they can use the next morning. Then keep going. Short refreshers, quick-reference guides and office hours beat one big kickoff session. We've written up how we think about AI proficiency levels if you want a way to benchmark where each team starts.

What BCG found about AI adoption at work
What BCG found about AI adoption at work. Source: BCG, AI at Work 2025: Momentum Builds, but Gaps Remain

Mistake 4: Leaving the Job Question Unanswered

Everyone in the room is thinking it. Is this coming for my job?

Ignoring that question doesn't make it go away. It makes it louder. And the anxiety isn't limited to the frontline. In the same BCG survey, 43% of leaders and managers worry about losing their job in the next decade, compared with 36% of frontline employees. Employees at companies that are redesigning their operations around AI worry more, too: 46% versus 34% at less advanced companies.

That second number is the uncomfortable one. The deeper you go with AI, the more people worry. Which means the companies doing this well have to talk about it more, not less.

Say what the plan is for people's time

You don't need to promise that no role will ever change. People will see through that. What helps is being specific about the plan for the time AI frees up. Which work moves off people's plates, and what they're expected to spend that time on instead. "We want the team spending less time on data entry and more time with customers" is a real answer. "AI will empower our people" is not.

This is a very common conversation in the PE-backed and industrial companies we work with. It goes better when leadership has the answer ready before anyone asks.

Mistake 5: Assuming Adoption Instead of Measuring It

Launch day feels like the finish line. Licenses are provisioned, the training happened, the announcement went out. So the project gets marked done.

Then nobody looks again for three months.

The BCG numbers show what happens in that gap. Regular AI use among frontline employees has stalled at 51%. And when employees don't have the AI tools they need, more than half say they'll find alternatives and use them anyway. So the usage you can't see is often the usage you should worry about most.

Trust works the same way. It's built or lost in the first few weeks. When we built an AI search assistant for Education Delta, one of the main design constraints was keeping school staff's trust. The system has guardrails so that when it can't find a relevant document, it points people to another source instead of guessing. It launched to 700 pilot users who asked more than 1,600 questions in the first two weeks. You only know numbers like that if you're counting from day one.

Pick three numbers and review them every week

Keep it simple. Weekly active users by team. The workflows people actually run. And one feedback signal, whether that's a pulse survey, champion requests or office-hours questions. Review them weekly for the first couple of months. When a team lags, add support there. When a champion proposes a new workflow, that's your next build.

Run the change plan alongside the build: where each of the five mistakes gets fixed
Run the change plan alongside the build: where each of the five mistakes gets fixed. PressW

A Change Plan Fits on One Page

None of this needs a 60-page change management deck. Here's the one-page version we come back to. Each mistake has an early warning sign you can spot in the first month, and a fix that's cheaper to do up front than to retrofit.

MistakeEarly warning signThe fix
Shipping the tool before the reasonPeople ask "are we supposed to use this?" after launchWrite a "what changes for me" line for every role before rollout
Top-down rolloutUsage spikes after the announcement, then dropsName champions on each team and give them early access
Feature trainingPeople attend training but don't change how they workTrain each team on one real weekly workflow, then keep reinforcing
Leaving the job question openQuiet resistance, rumors, managers hedgingSay plainly what freed-up time is for
Assuming adoptionNobody can say who is using it or for whatTrack active users, workflows and feedback weekly

If you look closely, all five fixes are cheap. A few sentences per role. A handful of champions. Training built on real work. An honest conversation. A weekly report. What makes them hard is that nobody owns them, because the project plan only has a line for the system.

The People Side Is the Project

I'm bullish on where AI adoption goes from here. The tools are good enough that most teams will get real value once they actually use them every day.

But I don't have a clean answer for every part of this. How fast should you push a team that's skeptical? When does a champion network need to become a formal role? How do you talk about job changes when you don't yet know which jobs will change? I think every company, including the ones we work with, is still figuring that out as the tools move.

What I'm sure of is that the companies treating adoption as part of the build get there first. That's how we run AI Change Management at PressW: stakeholder alignment, champion networks, role-based training and adoption tracking that run alongside every rollout.

If you have a rollout coming up, start with one question: can you write the "what changes for me" sentence for every team it touches? If not, that's the place to start. And if you want to compare notes on what you're seeing, Book a Discovery Call.

FAQ

AI change management is the work of getting people to actually adopt a new AI tool or workflow. It covers stakeholder alignment, communication, training, usage policies and adoption tracking. It runs alongside the technical build so teams are ready the day the system goes live.

Most AI adoption problems are people and process problems. BCG's research puts about 70% of AI implementation challenges there, versus 10% in the algorithms. Common causes are unclear expectations, generic training, unanswered job concerns and no one tracking usage after launch.

The five we see most are launching the tool without explaining what changes for each role, rolling out top-down instead of through champions, training on features instead of real work, ignoring job-security concerns, and assuming adoption instead of measuring it. Each is cheaper to fix before launch than after.

BCG's AI at Work 2025 survey found regular usage is sharply higher for employees who get at least five hours of training. The content matters as much as the hours, though. Training built around each team's real workflows sticks better than generic tool demos.

An AI champion network is a group of employees, usually one or two per team, who get early access to new AI tools, help shape workflows and support their peers. Champions spread adoption person to person, which tends to work better than top-down announcements.

Start with weekly active users by team, the workflows people actually run and one feedback signal such as pulse surveys or office-hours questions. Review them weekly in the first months of a rollout. The data shows where AI is sticking and where a team needs more support.

AI Change ManagementAI AdoptionChange Leadership

MORE FROM THE BLOG

Related Posts