3.1.4. When AI Is Not the Right Solution

💡 First Principle: AI is a tool for handling uncertainty and variety at scale. When a problem has no meaningful uncertainty, too little data, too little volume, no tolerance for error, or an underlying process that is simply broken, AI adds cost and risk without adding value. Saying "not AI" is a core strategic skill, and the exam tests it directly.

Signals that AI is not the appropriate solution:

• The logic is deterministic and stable. Tax calculations, statutory thresholds, eligibility rules: use rules (2.2.1).
• Data is insufficient, poor, or unavailable. If the patterns are not captured in data, or data cannot legally be used, the model has nothing reliable to learn from. Fix data first or choose a different approach.
• Zero error tolerance without the ability to add oversight. If any error is unacceptable and no human or system check is possible, probabilistic AI is a poor fit.
• Low volume or low value. Automating a task performed twenty times a month rarely justifies AI build and operating costs.
• The process itself is the problem. Automating a broken approval chain makes broken approvals faster. Redesign the process first.
• Explainability requirements exceed what the approach can provide. Some regulated decisions require reasons that certain models cannot give reliably.
• A simpler tool already solves it. Search, a report, a form, a lookup table, or standard workflow software may meet the need.
• Cost exceeds value once full ownership cost is counted (3.2.3).

Consider a city permit office with a twelve-week backlog. A proposal suggests an AI assistant to answer applicant questions. Analysis shows that 70% of the delay comes from three sequential sign-offs that could run in parallel and a form that requires the same information four times. Redesigning the process and form cuts the backlog by half at almost no cost. AI might help later, after the process is sound, but leading with AI would have automated the waiting.

⚠️ Exam Trap: When a scenario reveals that a problem is caused by process design, policy, missing data, or simple deterministic logic, the best answer is often a non-AI solution (process redesign, rules, data remediation, standard software). Answers that add AI to a broken or deterministic process are distractors, even when the AI option sounds innovative.

Reflection Question: An executive asks for an AI model to predict which purchase orders will be approved late. Investigation shows that approvals are late because one director must sign every order above $500. What would you recommend, and why is that a stronger strategic answer than the model?