Artificial intelligence initiatives are increasingly positioned as strategic imperatives. Yet many fail, not because the technology is immature, but because the decision to adopt it is insufficiently interrogated. Executives are often presented with compelling narratives, polished demonstrations, and competitive pressure, all of which can obscure foundational weaknesses.
Before approving any AI initiative, leaders should impose disciplined scrutiny. The following five questions are not technical, they are strategic. They are designed to separate signal from noise and ensure that investment decisions are grounded in operational reality rather than technological optimism.
1. What specific decision or process are we trying to improve?
AI does not create value in abstraction: It only delivers value when tightly coupled to a defined decision point or operational process.
Executives should demand precise articulation:
- What decision will be augmented or automated?
- What is the current baseline performance (cost, speed, accuracy)?
- Where exactly does AI intervene in the workflow?
If the answer remains vague, e.g., “improve insights” or “enhance customer experience”, the initiative is not ready. Ambiguity at this stage almost always translates into diffuse implementation and weak outcomes.
Test: If you cannot map the AI system to a specific operational lever, you are funding exploration, not execution.
2. What is the mechanism of value creation?
AI initiatives often rely on implied value rather than explicit economic logic. Executives must force clarity on how the system produces measurable benefit.
Key considerations:
- Does the system reduce cost, increase revenue, mitigate risk, or improve decision quality?
- What are the causal pathways between model output and business outcome?
- Are these pathways direct or dependent on human interpretation?
Many AI systems generate probabilistic outputs that require human judgment. If the human layer is not accounted for, projected value will be overstated.
Test: If the value cannot be translated into a defensible financial model, the initiative is speculative.
3. What are the system’s actual capabilities and limitations?
There is a persistent gap between what AI systems appear to do and what they can reliably do under operational conditions.
Executives should insist on:
- Performance metrics under realistic conditions (not curated demos)
- Failure modes and error distributions
- Sensitivity to data quality, context, and edge cases
Particularly with language-based systems, outputs can be fluent but incorrect. This creates a dangerous illusion of competence.
Test: If the system’s failure modes are not clearly understood, you are assuming risk without quantification.
4. What organizational changes are required for this to work?
AI is not a plug-in capability. It reshapes workflows, roles, accountability structures, and sometimes decision authority itself.
Critical questions include:
- Who is responsible for acting on AI outputs?
- How will workflows change?
- What new skills or training are required?
- Where does accountability reside when the system is wrong?
Most AI initiatives fail not at the model level, but at the integration layer, where human and machine processes intersect.
Test: If implementation assumes minimal disruption, it is almost certainly underestimated.
5. What is the downside risk, and is it acceptable?
AI introduces new categories of risk:
- Operational (system errors, brittleness)
- Reputational (incorrect or harmful outputs)
- Legal and regulatory exposure
- Strategic dependency on opaque systems
Executives must evaluate not only expected value, but risk-adjusted value.
This requires:
- Scenario analysis of failure cases
- Understanding of reversibility (Can we roll this back?)
- Clear governance mechanisms
In some cases, the downside risk outweighs the incremental benefit, particularly when systems operate in high-stakes or externally visible contexts.
Test: If the risk discussion is superficial, the approval process is incomplete.
Conclusion: Discipline Over Momentum
AI initiatives often gain momentum through external pressure: competitors adopting; vendors promoting, and boards inquiring. But momentum is not a substitute for judgment.
These five questions impose a necessary discipline:
- Define the decision
- Clarify the value mechanism
- Understand real capabilities
- Anticipate organizational impact
- Evaluate risk rigorously
Executives who apply this framework do not reject AI: they allocate capital more intelligently. In a landscape saturated with promise, the differentiator is not adoption speed, but decision quality.
J. Michael Dennis ll.l., ll.m.
AI Foresight Strategic Advisor
Based in Kingston, Ontario, Canada, J. Michael Dennis is a former barrister and solicitor, a Crisis & Reputation Management Expert, a Public Affairs & Corporate Communications Specialist, a Warrior for Common Sense and Free Speech. Today, J. Michael Dennis help executives and professionals understand, evaluate, and responsibly deploy AI without hype, technical overload, or strategic blindness.
You must be logged in to post a comment.