AI change management: your team is rarely the problem
AI transformation fails more often for lack of learning time than because of the team. Why pushback is a review brief, and five questions before any new tool.
An AI transformation fails because of leadership more often than because of the team. The usual response to skepticism is access for everyone, a demo, and the expectation that the rest will sort itself out on the side. I think that's the surest way to end up with the verdict that AI doesn't help. Hand out tools without planning time to learn, and you create the very pushback you'll blame on the team later. At SUMAX in Dortmund, Germany, I've been moving the whole company onto AI since 2023 as COO and Chief AI Officer: how we work, roles, services and rules. We're coming from an old CRM and Excel lists, and the new tools get built with the people who use them.
Access for everyone isn't a transformation.
I see two mistakes with AI, and they sit at opposite ends. One is trusting it too much: output gets passed on unchecked because it sounds right. The other is using it too little: everyone gets a $20-a-month subscription and that's it. That doesn't even scratch the surface.
Classic change models like Kotter's or ADKAR describe change in phases that end at some point. With AI the tools keep changing, and every result a model produces has to be checked. A phased plan with an end date doesn't fit that well. Treat the transformation as lasting rules for how work gets done, not as a project with a finish line.
Skepticism is usually a review brief.
Bringing in AI means stepping into other people's work. Behind the skepticism there are almost always two worries: about one's own role and about the quality of the work. Both deserve a straight answer.
On roles: if you promise that nothing will change for anyone, you lose credibility the moment something does. Tasks shift, some shrink, new ones come in. My answer to the fear of losing one's job is a conviction I act on: AI takes the busywork, not the people. The time that frees up goes into review and expert decisions.
On quality, the skeptics are often just right. A model produces mistakes that sound convincing, and whoever points that out is helping the rollout. When someone tells me a result isn't usable, I want to know how they can tell. Those criteria become review rules that apply to everyone. Ask every skeptic for their criteria and turn them into rules for the whole team.
Whoever passes it on checks it. Leadership included.
A model can prepare a first analysis. Putting it in context, spotting the gaps and making the right call takes experience in the field. That role has to be visible in the workflow, with its own time in the plan. Without it, technical progress looks like a put-down of the people who've carried the business for years, and the experienced ones are the first to pull back.
So expert judgment stays with people. Whoever passes on an AI result has checked it and owns it. Skip that and you're just a pass-through between the machine and the client. Leadership plays by the same rule: an announcement, a training handout or a conversation guide that ChatGPT drafted goes out only after you've checked it yourself. Model the review rule yourself, or nobody on the team will take it seriously.
Learning time goes in the plan, or nobody learns.
If leadership expects people to learn on top of a full workload, it gets half-hearted attempts and, in the end, the verdict that AI doesn't help. I'm a certified vocational trainer in Germany, and I know from training apprentices that learning needs structure and time. AI is no different.
This is what it looks like in a lot of companies: a team gets access to a new tool but no time to try it on its own tasks. A few weeks later two people use it heavily and the rest not at all. Leadership reads that as unwillingness. What was missing was a clear use case, set time to experiment and someone to ask.
We train on real tasks from each person's own area instead of generic demos. On a task someone has done for years, they see a model's limits faster than with any slide-deck example. Lawmakers are moving the same way: in March the European Parliament adopted the EU AI Act, which will require companies to make sure their staff have sufficient AI literacy. Training on real tasks serves that purpose better than a mandatory presentation. Schedule learning time like any other work, with hours in the plan instead of on the side.
Expect learning on top of a full workload and you get half-hearted attempts.
The job description has to change too.
Access to a language model doesn't change a job description. When AI prepares part of the work, roles shift, and that has to be written down somewhere: who prepares results, who checks them and who signs them off. Judging the output belongs in the business unit, because errors show up there sooner than in any outside check. Technical support is set up separately.
Before the tools comes the foundation for everyone: users, permissions, access and clear rules on which data may go into which tool. Security belongs in from day one, because it's very hard to bolt on later. If you have a works council, bring it in early. Rewrite the role profiles before the second tool arrives.
Feedback that gets no answer is the last you'll get.
Feedback from daily work decides what we build next. That only works if feedback visibly leads somewhere. Someone who reports a bug three times and sees nothing happen won't report it a fourth time. So I treat it as a leadership duty to answer every report: done, planned, or declined with a reason. When leadership hands a transformation off to a project team, the team treats it as a side issue. Leadership sets the frame itself: priorities, rules, time to learn.
Five questions before the next tool.
Before another AI tool goes in, you want a yes to each of these:
- Does the team have set hours in the plan to try the tool on its own tasks?
- Do the role profiles say who prepares results, who checks them and who signs them off?
- Are the skeptics' objections written down as review rules?
- Is it settled which data may go into the tool and who has which permissions?
- Does every piece of feedback get an answer, including a no with a reason?
A no isn't a reason to drop the tool. It shows you where the rollout will snag first.
Bringing people along is slower, and it holds.
Not every doubt goes away with training. Some people prefer to keep working the way they always have, and some tasks just aren't a good fit for AI. A leader has to be able to accept both without giving up on the direction. A transformation that brings the team along moves more slowly than one that hands out logins and waits for results. I can't give you a fixed timeline. Ours has been running since 2023.
I think the slower path is the only one that holds up. Take the tool that gets used least in your company and ask the people who were supposed to use it whether they ever had time for it in their plan.