The AI Kotter Change Coach skill for Claude

The problem the skill tackles is familiar to anyone who has ever rolled something out. The new way of working is announced, everyone nods in the meeting, and three weeks later half the team is back to doing it the old way. According to Kotter, the cause is rarely the content of the change. The SKILL.md says it plainly: the cause never lies in the content but in the process around it. People are not brought along, urgency is missing, short-term wins fail to appear, or the old behaviour is given room to come back too soon.

This file is written for anyone who needs to land a change without hiring a consultancy to do it: team leads, project leads, business owners, operations managers and internal change agents. The frontmatter names the cases itself: rolling out a new tool, adjusting a process, changing culture, a reorganisation, a merger, or an earlier change that stalled.

The frustrated remarks are in there too as triggers, and those are more recognisable than the jargon: people just won't pick it up, everyone slips back into old behaviour, I can't get them on board. If you don't yet know exactly where the change is getting stuck, start one step earlier with the McKinsey 7S Analyst: it lays structure, systems, style and shared values side by side and points out where they work against each other.

One thing is important to know upfront, and the skill says so itself in the file's first paragraph: the goal is not to invent the change but to make it land and stick. What needs to change stays your decision. What you have here is the coach that guards the steps, not the strategist that picks the direction. It belongs in the skill library NL that we make freely available through our AI and automation service, with no account and no follow-up sales email.

New to this kind of file? It helps to first read what Claude skills are NL and how you install NL them. Want to write one yourself for your own way of working? The structure is explained in writing a SKILL.md NL.

The figure of 70 per cent traces back to Hammer and Champy from 1993, on reengineering. On this page, and in the skill, we use Kotter's own wording. Source: Kotter in Harvard Business Review, 1995.

The urgency is never actually felt. Almost every failed change starts with a presentation explaining why the new system is better. What is rarely in there is what happens if things stay as they are. That is exactly what it forces: step one asks for the scenario over twelve and twenty-four months without intervention, for three hard facts from figures, customer reactions, market shifts or competitors, and for a single line that keeps coming back in every meeting.

There is a project lead but no coalition. A change that hangs on one person collapses the moment that person goes on holiday. This coach asks for four to eight people with a mix of formal power, subject expertise, credibility on the shop floor and leadership, and for a reason per person why they are there and what could hold them back. The first session goes into the diary within fourteen days, not at some point.

There is too little communication. According to Kotter this is the most common mistake, and the SKILL.md takes the rule of thumb literally: plan for a factor of 10 repetition. Hence the ninety-day calendar with at least ten moments in different formats, from an all-hands to a one-to-one conversation, each with a sender, a receiver, a core message and a call to action.

People are willing but unable. Step five is about obstacles: old processes, systems, rules, managers and rewards that punish the new behaviour. It then asks for the three biggest ones, with an owner, a date and a measurement per obstacle, and sets that as the coalition's top agenda item for the first sixty days.

The first win fails to appear. Without a visible result within sixty to ninety days, the energy evaporates. So it asks for two or three wins that can be unambiguously linked to the change and are relevant to the wider group, with the explicit rule attached: no inventing artificial wins, credibility is everything.

In diff form, with the logic of step one: how an announcement turns into an urgency narrative.

Next comes a short chapter on when the skill should switch itself on. Not only for an announced change, but also for signals of resistance that the user often does not recognise as a change question: they don't pick it up, people go back to the old way of working, everyone nods yes but does no, it stalls, I can't get them on board. That is exactly the moment when most people are not looking for a model but a listening ear, and when a model actually delivers the most.

The theory chapter summarises Kotter in half a page and lists the eight steps in their original order. The rule that governs the rest of the file follows right after: skipping steps feels faster but almost always leads to relapse, and the skill guards the order and forces the user to make every step concrete.

The way of working itself consists of three phases. Phase one is the intake with five questions, and the answers become the context for all eight steps. Phase two works out the eight steps, each with its own instruction list and its own output, as shown above in FIG.02. Phase three is the risk section: always close with three risks, each in the same format of what could go wrong, how you spot it going wrong through an early signal, and what you do then.

The output format is described separately and consists of four parts: a summary in five lines, the eight steps each worked out separately, the risk section and an action list for the first thirty days with an owner and a deadline per action. Then follows a chapter on tone that is surprisingly strict for an instruction file: write businesslike and concrete, no consultancy language, do not open with it is important to, and no English except terms that appear in the book itself. The user is the change leader, the skill is the coach that guards the steps.

The file closes with a source list of three titles and with the worked-out example of the CRM migration you see in FIG.01 above. That example is not there for decoration: it shows the level of detail the skill expects of itself, right down to the remark that CRM use gets built into the bonus structure.

Leading Change, John P. Kotter, Harvard Business School Press, 1996. The book in which the eight steps appear in their definitive form. The order in the skill is the order from this book, including the emphasis on anchoring as a final step rather than as an afterthought.

Leading Change Why Transformation Efforts Fail, Harvard Business Review, March and April 1995. The article that preceded the book. This is where the analysis the skill summarises in its theory chapter comes from: most changes fail not on content but on process, and the mistakes are consistently the same eight.

The Heart of Change, John P. Kotter and Dan S. Cohen, Harvard Business School Press, 2002. The follow-up, which turns on the question of how people actually get moving. You can see that reflected in how the skill handles communication: no memo but ten moments in different formats, each with a question to the receiver.

What the skill does not do is pad out those sources with research that is not in them. The figure of seventy per cent that appears in the theory chapter comes from this work. So you will not find success rates, benchmarks per sector or timelines on this page. Change processes vary too much to make any sensible promise about that, and the file does not do so either.

Want to look at the behavioural side separately from the process? There are two skills that go deeper into that. The Fogg Behaviour Model Coach looks at what needs to be right per action in terms of motivation, ability and prompt, and the EAST Framework Behaviour Coach makes the desired behaviour easy, attractive, social and timely. Where Kotter organises the process, those two are about the individual action within that process.

After that, using it is simple: describe in plain language what you want to change and for whom. It first asks the five intake questions, then starts at step one. You don't need to know anything about Kotter to work with it, and that is the point of a skill: the model sits in the file, not in your head. A fuller explanation of installing in different environments is at Installing Claude skills NL, and the broader background on working with AI is in the knowledge base.

There are also situations where you are better off leaving it aside. The first is the phase before that: if it is not yet settled what needs to change, the skill has no subject and you get a plan for something that may never happen. Choose first, and use an analysis or decision skill for that instead. The second is the individual conversation. An employee who is stuck is not an organisational change, and the eight steps turn a personal conversation into a project it is not. The Radical Candor Coach NL or the Nonviolent Communication Coach fits better there.

The third limit is the intake itself. Skip the five questions and ask straight for a plan, and you get eight headings of generic text. That is not a flaw in the skill but a result of the missing context, and the file is clear that the answers to the five questions form the basis for all eight steps.

And then the most honest limit: the skill makes your plan better, not your organisation. If there is no mandate, no time and not a single manager on board, a well worked-out eight-step plan will not make it succeed anyway. What it does do is make visible where it goes wrong, and that is useful information: a step two you cannot get filled in tells you something about your organisation before you are three months further down the line.

If the bottleneck sits in the team itself rather than in the process, look at the Tuckman Team Stages Coach. A team that a reorganisation knocks back to the starting phase behaves differently from a team that does not want the change, and those two need a different approach.

Download the zip, drop it into your AI environment and answer the five questions. Within half an hour you have an eight-step plan with a risk section and an action list for the first thirty days, and that is more than most changes ever get down on paper. If you would rather not explain it all over again every week afterwards, or if you want this kind of work ready and prepared with a person approving it, we simply talk about that. If your team already works with AI but without agreements, the AI training NL is a more logical first step than a new subscription.