Many organizations respond to AI with more documentation. Those that succeed respond with clearer direction and greater room to act.
AI governance is often misunderstood as control. In reality, it is about enablement.
Consider time. If no time is allocated for AI learning, one of two things happens: either nothing happens or it happens anyway, without structure, sharing, or organizational learning. Slack is not a luxury; it is a prerequisite, says Carl Heath.
The same applies to data. The most common misprioritization is not lack of data, but treating data as a technical asset rather than a business concern.
– When organizations measure what is easy to measure instead of what is relevant, decisions drift. This is Goodhart’s law* in practice.
Work practices present the opposite trap: the belief that everything must change. AI fundamentally alters information gathering, analysis, and draft production. It does not replace the value of human relationships, judgment, and trust. Sound governance begins where impact is high and risk is manageable.
The fine combination of direction and space
A functioning governance model must provide both direction and space. Carl Heath suggests three dimensions. First, compass: strategic intent, testable hypotheses, and clear mandate. Second, enablement: competence, interplay between operations and development, and psychological safety. Third, momentum: scaling what works and turning pilots into everyday practice.
– Governance that exists only on paper creates theater. Governance that exists in daily work builds capability.
Three questions for the executive team:
- How can our AI governance create clarity and momentum rather than bureaucracy?
- Where do we need clearer direction and where should we deliberately create more room to act?
- How do we ensure governance lives in daily work, not just in policies and documents?
*Goodhart’s Law states that “when a measure becomes a target, it ceases to be a good measure”. This concept highlights the potential for unintended consequences when metrics are used as targets, leading to distorted behavior and a narrow focus on meeting these targets rather than addressing the underlying reality. The law is attributed to Charles Goodhart, a British economist, who articulated the idea in 1975.
Carl Heath is a senior researcher at RISE and a doctoral researcher at the University of Gothenburg, focusing on leadership, AI and digital resilience. His work explores how organisations can navigate technological change while strengthening human judgement, trust and agency. He approaches AI not as a technical issue, but as a leadership and organisational challenge rooted in learning and everyday practice.