Becoming AI Ready through Governance

Boards mistake milestone reporting for governance. A programme can be green every month, hit every deadline, and still fail, because the board never asked the one question that mattered: what problem will still exist the day after go-live?

I have seen this play out where an eighteen-month transformation programme was eventually cancelled after significant investment. The issue was not poor reporting; the issue was that the board was measuring delivery progress rather than business impact.  

• The same mistake shows up in a sharper form with AI specifically: a project is approved because the demonstration was compelling, not because anyone defined the value it was expected to create or how that value would be measured once it was live. I have watched a utilities organisation deploy AI to help call centre agents document conversations. It shipped. The board could point to a live AI use case. Underneath it, the business was losing recoverable revenue through a property ownership dataset nobody had ever been made accountable for. The AI that shipped made no difference to that number, because nobody had asked what it was actually meant to move before it was approved.

This persists because programme teams under delivery pressure report what is on track, not what is at risk. Boards accept RAG status, and they accept a good demonstration, because both feel like oversight without anyone challenging the number underneath. Until a board asks what business outcome it expects, how that outcome will be measured, and who is accountable for delivering it, it is not governing. It is simply approving activity.

Who owns this, by name? Not a function. Not a committee. A named individual who owns the decisions, the data, and the consequences. If the answer is a function rather than a person, accountability does not yet exist.

What measurable outcome will tell us it is working, and is this the highest-value problem we could point AI at, or just the easiest one to ship? Most AI business cases describe capability, not value, and approval usually follows the demonstration rather than the return. A use case can work exactly as designed and still fail the organisation, if it was never pointed at the problem that actually moves a number the board cares about.

What controls and monitoring are in place for when this gets it wrong, and who is named to press the button if it needs to be stopped? Not if. Every AI system fails somewhere. The question is whether testing, thresholds, and one accountable person with the authority to halt it exist before deployment, not after.

• A board's job is not to predict whether AI will succeed. It is to ensure the organisation knows who owns the outcome, how success will be measured and who has the authority to intervene when it fails. If those questions cannot be answered, the investment case is incomplete, however impressive the demonstration.

The fastest way to recognise an organisational problem disguised as a technology problem is to ask what would change if the technology arrived tomorrow. If a new platform is proposed and nobody today can say who will be answerable for the outcome it produces, the organisation is buying technology to avoid an organisational decision it has not yet made.

I look for three signals. First, when ownership is described only as 'the data team' or 'IT', rather than a business leader accountable for the outcome, the organisation does not have robust ownership, and no platform fixes that. Second, when two senior leaders quote different figures for the same metric in the same meeting, the gap is not a tooling problem; it is an unresolved question of whose number is authoritative. Third, when governance or data is treated as a project with a start and end date, rather than part of the organisation's operating model, the organisation is treating an operating model problem as a delivery problem.

The practical test is simple: imagine deploying the technology tomorrow without changing a thing. If no accountable person or role already exists for its decisions and outputs, the barrier was never the technology. It was governance.

Governance and innovation are often presented as opposing forces, but I think that's the wrong view. Good governance is what allows organisations to innovate repeatedly without recreating the same problems at greater speed. AI does not create organisational capability; it amplifies the capability already in place. Where accountability, data quality or decision-making are weak, AI will accelerate those weaknesses rather than resolve them. Organisations with strong governance become faster. Organisations with weak governance become faster at making inconsistent decisions.  

• The organisations moving quickest are not the ones with the least governance. They are the ones with the greatest clarity. Everyone knows who owns the data, who owns the business outcome, what success looks like and who has the authority to intervene when something goes wrong. That clarity removes friction rather than creating it.  
• I describe the hidden cost of poor governance as the Friction Tax. Teams spend time reconciling reports, debating whose numbers are correct, repeating work and rebuilding trust after avoidable failures. None of that is innovation, it’s the Friction Tax. Governance should eliminate that tax by establishing clear accountability and consistent decision-making before technology is introduced.  

The role of executives is not to choose between innovation and governance. It is to ensure governance becomes part of how innovation happens. When governance is treated as infrastructure rather than a gate, transformation accelerates because teams spend their time solving business problems instead of resolving uncertainty. Governance is not the cost of innovation; it's what allows innovation to scale.

The defining capability will be organisational readiness: the ability to combine trustworthy data, accountable decision-making and adaptive ways of working.  

The capability is created through honest self-assessment against foundations, not ambition against use cases. I built a 2030 Convergence Readiness Scorecard for exactly this reason, and the detail that surprises people every time is where it starts. The first dimension is called AI Amplification Readiness, and it does not ask what AI can do. It asks about error rates in the underlying data, whether a single source of truth exists, whether human-in-loop checkpoints are defined, and whether a bad AI decision can be rolled back. That ordering is deliberate. AI does not fix a weak foundation, it amplifies whatever foundation is already there, so the organisations that score well on capability but poorly on foundation are the ones scaling their existing errors faster, not solving them.

Boards often ask, "Are we ready for AI?" because they are focused on adoption. The more important question is, "Is AI ready for the organisation?" because success depends less on what the technology can do and more on whether the organisation has the data, accountability and operating model needed to use it effectively. AI does not fix organisational weaknesses; it exposes and scales them.

-----------------------------------------------------------------------------------------------------------------------------------------------------

* These responses draw on the Data Leadership Series, the Friction Tax and Rehearsal Gap frameworks, the whitepaper What Will Be Different This Time, and board advisory work in Digital, Data and AI.