AI Fluency Is Not AI Capability: Can Leaders Apply the Strategy Consistently?
An AI strategy becomes credible when leaders can apply it to competing investments, operating constraints and changed assumptions without contradicting its stated priorities.
Strategic language is easy to repeat
An AI strategy may promise responsible innovation, customer value, efficiency and human oversight. Each phrase can be reasonable. However, the combination remains incomplete until leaders use it to choose between incompatible alternatives.
The test is consequential. Which proposal receives funding? Which use case remains outside scope? What evidence is sufficient for deployment? Who may stop the system? If the strategy cannot guide these choices, it has not yet become an operating capability.
A research result suggests a useful test
Mancoridis, Weeks, Vafa and Mullainathan study whether large language models apply concepts consistently across linked tasks. A model may explain a concept correctly but fail to generate or revise an example that satisfies it. The authors call this pattern “Potemkin understanding” and interpret it as evidence of incoherent concept representations.
The paper studies language models, not management teams or corporate strategy. The organisational application that follows is a Marketways synthesis: an organisation may use the language of AI strategy without applying its decision logic consistently.
Classification tests the boundary
Give leaders a mixed set of proposed investments and ask which proposals fit the AI strategy. Require a reason for every inclusion and exclusion. A strategy that prioritises customer trust, for example, should affect the treatment of a high-return use case that relies on weak consent or opaque decisions.
The classification test reveals whether the AI strategy defines a real investment boundary. If every attractive proposal qualifies, the strategy has not established a priority.
Construction tests whether the choices can coexist
Ask the leadership team to allocate a fixed budget across data preparation, workflow redesign, technology, assurance, adoption and capability. The resulting plan must still meet the strategy's stated objectives and risk limits.
A team may discover that its aspirations require more operating change than the budget permits. Leaders must then narrow the scope, change the timing or revise the ambition instead of preserving incompatible promises.
Revision tests whether the logic survives change
Change one material assumption. A supplier may withdraw a capability, the evidence may show weaker performance or a regulator may impose a new control. Ask which decisions must change and which should remain.
Consistent revision demonstrates that leaders understand the relationships inside the strategy. Random reversal suggests that the original choices depended on unstated preferences rather than an agreed decision rule.
Coherence does not prove that the strategy is right
A leadership team can apply a strategy consistently even when its assumptions about customers, cost, capability or risk are wrong. Linked decision tests establish coherence. They do not establish commercial validity.
Management must therefore test the assumptions separately through market evidence, operational data, pilots and evaluation. AI Opportunity Assessment tests the opportunity, while Decision Assurance examines whether the final commitment follows from adequate evidence.
Turn the strategy into decision rules
A usable AI strategy states where the organisation seeks advantage, which operating capabilities it will build, what it will not pursue and what evidence each investment stage requires. It also assigns authority for deployment, restriction and withdrawal.
Explicit scope, capability, exclusion and evidence rules allow leaders to classify, construct and revise actual plans. Strategic fluency may improve the discussion. Consistent decisions demonstrate the capability.