Learning on the Job: What the Board of Directors Needs to Demand when 89% of the AI Strategy is Improvised

01.06.2026

5 min read

6 Min. read

89 percent of companies say their AI strategy follows a “learning on the go” approach. Meanwhile, investment appetite is hitting record levels. This growing gap is the real boardroom question for 2026: Who is liable when improvised governance meets production systems?

Key Takeaways

Boardroom Briefing

Speed beats structure. 94 percent report a greater appetite for AI investment, yet 89 percent are moving forward without a fixed governance framework. The board sees the spending, not the guardrails.
The competency gap is internal. 62 percent of CIOs admit they make compromises on governance because they lack the know-how. Only 44 percent fully understand the risks.
Liability becomes a boardroom issue. The CIO is no longer just a technology operator but a coordinator of risk and accountability. The board must actively mandate this role, not simply assume it.

The gap between appetite and maturity

The latest global CIO survey delivers an uncomfortably precise finding. The willingness to invest in AI is at a record high, while the ability to operate those investments in a controlled manner lags considerably behind. More than half of respondents already feel their own pace is too high.

For the supervisory board, this is a familiar pattern in new clothing. An organisation spends money on a capability it hasn’t yet mastered and fails to document the gap. As long as nothing goes wrong, the board sees only the investment sum. When something does go wrong, it sees the missing governance – and sees it first.

The question in the boardroom is shifting accordingly. It is no longer about whether the company is investing enough in AI. It is about whether someone is exercising control over what that investment is actually doing in production. That’s a question of structures, not budgets.

The number that should alarm the board

If a single metric can sum up the maturity of AI governance, it is the admission from those in charge themselves.

Boardroom Insight

89 %

of companies describe their AI approach as “we learn on the go”, while 94 percent are expanding their investments.

Source: Logicalis Global CIO Report 2026

Improvisation is legitimate in an early phase. It becomes a risk when improvised rules encounter systems that make decisions in day-to-day operations or have an external impact. That shift is happening right now, and it’s happening faster than the building of the corresponding oversight.

What the supervisory board should specifically demand

A board doesn’t steer through technical details but through questions that require a robust answer. Three questions separate a controlled AI programme from an improvised one.

Who bears accountability? Is there a named person who answers for AI risks across the company – with a mandate and resources, not just a title?
Where does autonomy end? For which decisions may a system act independently, and at what level of impact must a human be mandatorily involved?
What is documented? Can data flows, model decisions, and escalations be evidenced in an emergency, or does the organisation rely on individual memory?

Anyone who cannot answer these three questions within a single meeting doesn’t have a technology problem. They have a governance gap that lands directly on the board’s desk when things go wrong.

The Silent Side Bill: Energy and Sustainability

One aspect is almost entirely missing from most boardroom debates on AI. Only 39 percent of CIOs are very confident that their company actively manages the ecological footprint of AI. Confidence in operational energy efficiency is hardly any higher.

For supervisory boards with reporting obligations, this is no mere footnote. AI operations consume measurable amounts of energy, and these figures are increasingly finding their way into regulatory reporting. An AI program without an energy balance sheet is a program with an open flank in its sustainability report.

From Investment Topic to Control Question

The shift defining 2026 is not technical in nature. AI has arrived in the executive suite; the budget is secured. What is missing is the equivalent establishment of oversight, accountability, and documentation. As long as this gap remains open, the supervisory board carries a risk it cannot quantify.

The productive step is unspectacular. The board no longer demands visions from the CIO, but defensibility: named responsibility, clear boundaries of autonomy, and documentation that holds up in a crisis. This costs less than the next model upgrade and protects against the most expensive scenario: an incident with no one accountable.

Frequently Asked Questions

Why is “we’ll learn as we go” a boardroom risk with AI?

Because improvised governance meets production systems that make decisions or have external impact. Learning is normal during the pilot phase, but in live operation, the lack of structure becomes a liability risk that lands on the executive board in the event of damage.

What questions should a supervisory board ask about AI?

Three are enough to start: Who bears accountability for AI risks, where does system autonomy end, and what is documented in case of emergency? If these cannot be answered in a single session, there is a governance gap.

How is the role of the CIO changing in the executive suite?

The CIO is no longer just a technology operator but coordinates risk, ensures accountability, and drives value creation. The board must actively mandate this expanded role and equip it with resources, rather than silently assuming it exists.

Why does the AI energy balance belong in the boardroom?

Because AI operations consume measurable energy, and these figures are moving into regulatory sustainability reporting. Only 39 percent of CIOs actively manage the ecological footprint, which represents an open flank in the report.

What does defensibility instead of vision mean for AI strategy?

It means the board demands robust evidence instead of future scenarios: named responsibility, clear autonomy limits, and documentation that stands up in a crisis. This is cheaper than any model upgrade and protects against incidents with no one accountable.