2.2.1. Rule-Based Automation vs. AI
💡 First Principle: If you can write the complete logic down and it rarely changes, write it down: rules execute perfectly, cost little, and are easy to audit. Use AI only where the logic cannot be written down because inputs are too varied or patterns too subtle, and where the business can absorb some error.
| Choose rule-based automation when… | Choose AI when… |
|---|---|
| Logic is fully known and can be enumerated | Patterns are too complex or numerous to write as rules |
| Inputs are structured and predictable (forms, codes, thresholds) | Inputs are varied or unstructured (free text, images, speech) |
| Every decision must be identical and fully explainable (regulatory calculations, eligibility thresholds) | Some error is acceptable or can be caught by review |
| Requirements change rarely and changes are deliberate | Patterns shift and the system benefits from learning from examples |
| Volume or complexity is too low to justify AI cost | Volume and variety make manual handling or rule maintenance impractical |
Real systems are often hybrid. An expense-approval workflow can apply firm rules ("receipts required over $75," "alcohol not reimbursable") while an ML model flags unusual patterns for audit and a generative AI step reads receipt images. The rules provide guarantees; the AI provides perception and pattern detection. A hybrid design also gives a natural safety net: rules can act as hard limits on what AI-driven steps may do.
Watch for rule explosion as a signal to consider AI. If a rules-based email router has grown to 2,000 keyword rules that still misroute a quarter of messages, the problem has outgrown rules. Conversely, watch for AI where a lookup would do. If a team proposes a generative AI assistant to tell employees how many vacation days they have left, a database query and a simple interface will be faster, cheaper, and always correct.
⚠️ Exam Trap: When a scenario stresses that decisions must be consistent, fully auditable, based on fixed criteria, and identical for identical inputs (tax calculations, statutory eligibility, contractual thresholds), rule-based automation is usually the best answer, even if an AI option sounds more advanced.
Reflection Question: A utility company calculates late fees using a regulated formula and also wants to identify customers likely to fall behind on payments. Which of these is a rules problem and which is an AI problem, and how could the two work together?
