When the problem is still unclear.
If the team cannot agree what needs improving, selecting a tool is premature. Start with the work: what happens, where it gets stuck and what a better outcome would look like. That conversation may reveal that the real issue is ownership, information or an unnecessary step.
When you cannot check the result.
An answer that sounds confident is not necessarily correct. Before relying on an AI output, decide how it will be checked and whether the person checking it has the knowledge and time to do so. If meaningful verification is not possible, reconsider the task or the level of reliance.
When the information is not appropriate for the proposed tool.
Do not treat convenience as a reason to skip an information-handling review. The suitability of a tool depends on its setup, the information involved and the organisation’s requirements. Ask the responsible person or specialist when those boundaries are unclear.
When the relationship matters more than the speed.
A sensitive client conversation, a difficult team discussion or a decision with significant consequences may need direct human attention. AI might assist with preparation, but that does not mean it should own the interaction or the decision.
When a simpler solution works.
A checklist, a clearer template or a rule-based workflow may solve the problem with less complexity. Consider the effort of introducing, supervising and maintaining the AI step as well as its potential benefit.
A useful boundary is a positive decision.
Choosing not to automate something is not falling behind. It is a decision about what the business values and where its attention belongs. Review the boundary as the context changes, but start with an approach you can explain.
