AI Strategy

5 Steps for Change Management in the Age of Gen AI

By Alyssa Caridi, VP, Strategy & Innovation · Last updated July 4, 2026

• Change Management
• Gen AI
• AI Adoption
• AI Training

Change management for gen AI is the work of moving a team from occasional experimenters to daily users who trust the tools enough to rebuild their work around them. In the rollouts I have led, the deciding factor is almost never the strategy deck. It is the quality of the very first thing people receive from the system. Adoption sticks when the first delivery is accurate, when automation removes guesswork instead of handing it to the user, and when it is obvious what someone is supposed to do with the output. The five steps below cover the goal, the trust, the workflow, the expectations, and the people. A 2025 Udacity survey found that three in four workers abandon AI tools mid-task, which is what happens when those five things are missing.

What is change management for gen AI?

Change management for gen AI is the work of getting a team to adopt, trust, and rebuild how they work around AI so the investment produces results. It is a different job from a traditional software rollout. Gen AI is closer to a capability than a feature, so getting value from it means changing how work happens. Teaching people a new interface does not get you there. The part most plans underinvest in is the quality of the output itself. You can run flawless training and still lose the team if the first report the system produces is wrong, or if it lands on someone's desk and they cannot tell what they are meant to do with it. Trust in the tool is built or broken at the point of delivery.

Why do most AI rollouts lose people?

Most rollouts lose people because the first thing the system delivers is almost right instead of right, and almost right is expensive. Stack Overflow's 2025 developer survey found that the number one frustration, for 45 percent of respondents, is AI solutions that are almost right but not quite, because fixing the near-miss costs more time than doing the work by hand. Udacity found the same pattern across roles: nine in ten workers use AI on the job, yet three in four abandon a tool mid-task over accuracy, refinement time, and poor workflow fit. Confidence is falling even as usage rises. ManpowerGroup's 2026 barometer reported that regular AI use grew while worker confidence in AI dropped 18 percent over the year. Once a team learns to distrust the output, no amount of change messaging wins them back.

Step 1: How do you set a North Star tied to outcomes, not tools?

Start by naming the outcome you want, then find the capability that moves it. A North Star should be simple enough for everyone to understand and specific enough to measure, such as cutting the time to produce a monthly report or removing a manual data-entry step from a daily process. Define how the work will change and what "done" looks like before you evaluate a single tool. McKinsey makes the same point in its analysis of change management in the gen AI age: treat gen AI as a capability and lead with the outcome, because that is what lets you tell whether the tool actually worked. Tools bought without a target become cost. An outcome you can measure keeps the whole effort honest. If you are starting the plan itself, the 90-day guide to building an AI strategy from zero lays out the sequence.

Step 2: How do you build trust by getting the first delivery right?

Trust is set at the first delivery, so hold a 100 percent accuracy bar before anything reaches the team. In my experience, the fastest way to kill a rollout is to ship a result that is close and tell people this is the way we work now. They try it, it is wrong, and they quietly go back to the old process while nodding along in the meeting. The fix is to validate outputs against a known correct answer before release, keep a human reviewing the edge cases, and refuse to lower the bar to hit a launch date. A tool that is right the first time earns repeat use. One that is almost right teaches people to re-check everything it produces, which erases the time the tool was supposed to give back. Accuracy is not a polish step at the end. It is the thing that makes adoption possible.

Step 3: How do you automate the guesswork out, not into, the workday?

Automation should take guesswork away from the person, not move it onto them. A tool that hands back something a person has to interpret, second-guess, or reverse-engineer has relocated the work rather than reduced it, and people feel that immediately. The rollouts that hold are the ones where validation is built into the workflow so the output arrives already checked, formatted, and ready to use, with no manual cleanup between the system and the next step. This is also where usage is won. McKinsey found that 45 percent of US employees would use gen AI more if it were built into their daily workflows, and 48 percent would use it more with proper training. Both numbers describe the same friction: the guessing and the cleanup that still sit between people and the tools. Remove that, and use climbs on its own.

Step 4: How do you set explicit expectations for every handoff?

Make every handoff clear about three things: what the output is, how far it has been checked, and what the person is supposed to do with it. Ambiguity is where adoption stalls. Someone who cannot tell whether an output is final, a draft, or a suggestion will treat it as none of those and set it aside. A good handoff looks concrete. The report arrives labeled "validated, ready to send," or "draft, needs your review of the highlighted figures," or "flagged three exceptions, please confirm before it goes out." That one line of context is the difference between a person acting on the output in seconds and a person opening it, feeling unsure, and closing it again. Set the expectation at the point of delivery, every time, so no one has to guess what a given output means or what to do next.

Step 5: How do you turn your team into change agents?

A workflow redesign does not hold unless the people doing the work help build it. McKinsey's transformation research found that only about 2 percent of employees are directly involved in a typical transformation, while organizations that involve at least 7 percent roughly double their odds of a strong financial outcome, and the best performers involve 21 to 30 percent. The practical version is to invite the people closest to a process to shape how AI fits it, since they know where the real friction is and where an output would actually save time. Find the early superusers and let them show peers what good looks like. Have leaders use the tools in plain sight rather than delegating them. Change that comes from the middle of the team outlasts change handed down as a mandate. Deciding who holds the mandate in the first place is its own question, covered in who should lead AI transformation at your agency.

How do you measure whether change management is working?

Measure change management by regular use and abandonment, not by licenses purchased. The signals that matter are how many people use the tool on their own after the launch push ends, how often they abandon it mid-task, and whether the outcome you named in step one is moving. Abandonment is the early warning. When people quietly stop using a tool, it is almost always because the output failed them once and they decided not to risk it again. Track that, ask why, and fix the accuracy or the clarity problem behind it before you add more tools. Set your target metric before you begin, review adoption and business impact together, and treat sustained voluntary use as the clearest sign the change has taken hold.

Change management in the gen AI age comes down to what you put in front of people and how you treat their time. Set an outcome, hold the accuracy bar, automate the guesswork away, make every handoff clear, and let the team help build it. Do that and the tools stop being something people are told to use and become something they reach for.