Every AI program your organisation has started AI governance from scratch. The organisation stood up a steering committee, drafted principles, set risk tolerances, and defined approval thresholds. Then the program ended. The committee disbanded, the sponsor moved to the next priority, and everything the organisation learned about governing AI left with the people who held it.
In Brief
- Organisations that retain AI governance learning across programs start each new cycle further ahead than they started the last one.
- Governance that resets with each program forces the organisation to re-pay the governance start-up cost every cycle.
- Documentation captures what was decided, not the judgement gained from applying it under operational pressure.
- After three or four cycles, the organisation that resets cannot close the gap in a single cycle.
- Attaching governance to the operating function rather than the program is a single structural decision available now.
Six months later, the next AI program began. A new committee convened, drafted fresh principles, and debated risk tolerances as though none of this had happened before. The organisation paid the full start-up cost of governance again because nothing from the first cycle carried forward. Senior leaders who have governed three or four AI programs recognise this. The third steering committee rehearses the same early conversations as the first. The same risk questions come up and the same delegation debates run again. The governance quality stays flat across cycles. What shifts is the executive’s growing awareness that the organisation keeps starting from the same place.
Governance resets with every program
The pattern looks like responsible governance because every program has a committee, a framework, and a set of delegations. The structural problem is that none of it persists. Governance is scoped to the program, not to the function. When the program closes, its governance closes with it, and no mechanism exists to carry what was learned into the next cycle’s governance design. The gap is invisible from the inside because each individual program appears well-governed. It only becomes visible when the executive looks across programs rather than within them and notices that the third program’s governance committee is re-answering questions the first one already settled.
Boston Consulting Group’s (BCG) 2024 research into what separates AI leaders from the rest found the differentiator was organisational: how companies structured their AI operating model determined whether early gains extended into sustained performance (BCG, 2024). The structural factor that receives the least executive attention is whether the AI governance function retains what it learned across operating cycles.
Zen Ex Machina’s governance maturity tiers describe this structural gap in concrete terms.
A governance function operating at Tier 3 maturity carries specific assets forward. It retains a settled view on which use cases need executive approval and which do not, along with the thresholds for autonomous AI decision-making that the organisation learned to trust. It holds the approval pathways that were redesigned mid-cycle because the originals bottlenecked. Each of these positions was expensive to reach. A governance function that retains them starts the next cycle from a materially different position because it does not need to re-establish them.
A Tier 1 governance function, one that resets with each program, carries none of these forward. The new committee does not know which delegations were tested or which risk thresholds proved too conservative. The new principles document does not reflect what the previous one learned under operational conditions. Over three or four planning cycles, the distance between these two organisations widens until the one that resets cannot close it inside a single cycle.
Organisations that recognise this gap often try to solve it with documentation: governance wikis, lessons-learned documents, archived risk registers. Documentation is not retained learning. A wiki page captures what was known at a point in time. It does not carry the judgement that came from applying that knowledge under pressure, or the institutional memory of why a specific approval pathway was abandoned mid-cycle. The governance function that retained its people and its sponsor knows why decisions were made and what happened when they were applied. Those are the assets that shift the starting position of the next cycle.
Competitors pull further ahead each cycle
The immediate consequence is visible inside the organisation: it re-pays the governance start-up cost with every AI program. Decisions that were settled in the previous cycle get re-opened. Risk tolerances that were calibrated are debated from scratch. Executive time and committee bandwidth that could have been directed at genuinely new problems go instead to re-establishing what was already known.
With each planning cycle that passes without retained governance learning, the number of cycles the organisation needs to catch up grows.
By the third or fourth cycle, the organisation that resets has fallen several cycles behind. It spent each of those cycles re-establishing its governance baseline rather than advancing from it. The organisation that retained its governance learning started each cycle with a higher baseline and moved further in the same period. It is now making the harder decisions available only to a governance function with that history: which AI capabilities to embed permanently, which approval pathways to remove because the delegations proved reliable, and whether to lower risk thresholds that two cycles of operation showed were set too conservatively.
These decisions produce durable advantage, and they are accessible only to governance functions that carried their learning forward. An organisation that resets cannot reach them in one cycle because the prerequisite decisions, the ones it keeps re-opening, stand between it and the harder choices.
Governance lives in the function
The structural decision that separates these two outcomes is whether governance is attached to the program or to the operating function. Attach it to the program and it is created and dissolved with the program lifecycle. Attach it to the function and it persists: the governance team and sponsor stay in place, principles evolve rather than restart, and the decisions settled in the last cycle become the starting position for the next one.
This is a single operating decision about where the governance function sits, not a governance maturity exercise or a documentation initiative. It has to be made before the next AI program starts. Making it afterward means the reset has already taken hold.
What this means for senior leaders: Before the next AI program is approved, put one directive to the transformation lead: this program’s AI governance will carry forward the team, the sponsor, and the learning from the last one. That decision determines whether the organisation enters its next cycle further ahead or from the same starting position.
The executive who makes it ensures their organisation carries settled governance positions into the next operating cycle and starts from a place that competitors who reset cannot match.
References
BCG. (2024). Where’s the value in AI? Boston Consulting Group. https://www.bcg.com/publications/2024/wheres-value-in-ai
