Step 5: Empower Broad-Based Action

Even when people are convinced that ethical AI matters and inspired by the vision for it, they will struggle to act if the organisation’s systems, structures and cultures are working against them.

Kotter’s fifth step is about removing the obstacles that prevent people from acting on the vision. For ethical AI, these obstacles are real and structural:

Structural obstacles: no budget allocated for ethics work, no time protected in project timelines for impact assessments, no clear process for raising ethical concerns about an AI system.

Cultural obstacles: a culture that rewards speed above all else, where raising ethical concerns is seen as obstructing progress, where “we’ll fix it later” is an acceptable answer to a serious risk.

Skills obstacles: people who want to behave ethically but lack the knowledge to do so. A product manager who genuinely cares about fairness but has never heard of algorithmic bias. A data scientist who wants to build responsibly but has never been trained in privacy-preserving techniques.

Empowering action means addressing all three. It means resourcing ethics work properly. It means protecting time and creating processes. It means building capability across the organisation, not just in a specialist ethics team. And it means changing the culture so that raising ethical concerns is celebrated, not penalised.

Step 6: Generate Short-Term Wins

Long transformation journeys need early victories. Kotter’s sixth step recognises that sustained change requires evidence that the effort is working: visible, credible proof that the new direction is producing results.

For ethical AI, short-term wins might include:
– Successfully completing the first AI impact assessment and using it to improve a deployment decision
– Identifying and correcting a bias in a live AI system
– Launching a cross-functional AI ethics training programme with strong participation
– Receiving positive external recognition for ethical AI governance
– A specific case where the ethics process prevented a harmful deployment

Short-term wins serve multiple purposes. They maintain momentum. They demonstrate to sceptics that ethical AI is practical, not just aspirational. They give the guiding coalition evidence to share with senior leaders. And they give everyone involved in the change effort a moment to recognise their own progress.

Critically, wins must be genuine. Manufactured or superficial victories: announcing a set of principles without any accompanying change in practice, for example: are quickly seen through and undermine rather than build credibility.

Reflection question:  What would a meaningful short-term win for ethical AI look like in your organisation?