AI in change management is no longer an abstract idea for me. It is part of my working day. On a global ERP implementation, I am expected to use Microsoft Copilot when developing and reviewing deliverables and when checking that the work reflects recognised good practice.
That does not mean the transition was effortless. When I first began using AI regularly, I found it overwhelming. I was learning a new tool while still being responsible for the quality, relevance and accuracy of work produced in a complex change environment. Six months later, I am more comfortable with it, but the most useful lesson has not been about a particular feature. It has been about the quality of the instruction.
The tool is as good as the instructions you give it.
Caroline Potter
AI needs a clear change management question
A vague request usually produces a vague result. Before asking Copilot for support, I need to be clear about the outcome I am trying to reach, the audience involved and the context surrounding the change. I also need to decide what part of the task genuinely benefits from AI and what still requires my own judgement.
In practical terms, I find it useful to think through four things before writing an instruction:
- Purpose: What am I trying to understand, test or produce?
- Context: What does the tool need to know about the programme, audience or business environment?
- Boundaries: What should it avoid assuming, changing or presenting as fact?
- Output: What would a useful response look like and how will I assess it?
This pause is important. It prevents AI from becoming another source of noise and makes it more likely that the response can contribute to the work.
The first answer is a starting point, not the deliverable
Using AI well is iterative. I may need to refine the question, add missing context or challenge an answer that feels too generic. I still need to check the result against the realities of the organisation and the people affected by the change.
That review cannot be delegated. A tool does not sit in stakeholder meetings, hear the hesitation behind a question or understand the history that has shaped a team鈥檚 response. It can help organise information and test thinking, but I remain accountable for what is used.
Start with a useful, contained task
For anyone feeling overwhelmed by AI, my advice is to begin with one contained part of the working day. Frame the question carefully, review the response critically and notice what makes the next attempt more useful. Confidence develops through repeated, thoughtful use rather than trying to transform every process at once.
AI can support change management work, but useful outcomes still begin with a person who understands the problem. In a related Insight, I look at why AI should augment experience rather than replace it.
