AI & decision-making
When is a rules-based workflow better than an AI model?
If the decision can be written as a small set of stable conditions, a model may add uncertainty rather than capability.
Published 25 September 2026
A request should go to a manager when its value exceeds an agreed threshold and two documents are present.
A rule can check those facts and show why it routed the case. A model is harder to justify unless the documents themselves require interpretation.
Prefer rules when the criteria are known
Thresholds, required fields, permissions and fixed routing conditions are easier to test and audit as rules. They also make it obvious when policy changes. Do not pay for a probabilistic answer to a deterministic question.
Rules still need ownership. Someone must decide what happens when a field is missing or two conditions conflict.
Use a model for the part that needs language or judgement support
A model may help summarise an unstructured request or suggest a category, leaving a rule or person to decide the action. Keep its input bounded and test it against difficult examples.
The right design is often a combination: model assistance for interpretation, explicit rules for authority and an accountable person for exceptions.