AI can crunch the channel data well, but the ranking call still needs a human.

The analytical structure is repeatable — pull data, segment by source, compute metrics — but the interpretation layer shifts with business context each cycle. A quarterly refresh is structurally similar, but the judgment about what 'good' looks like changes as strategy evolves.

Success criteria are underspecified: 'highest quality,' 'best customer fit,' and 'longest retention' each require a definition the task doesn't provide. Without explicit weighting of these dimensions, the agent must make assumptions that could silently misalign with what the business actually values.

This depends entirely on whether the agent has clean, structured access to CRM data, attribution data, and retention metrics — which is often not the case in practice. Data silos, inconsistent UTM tagging, and missing retention fields are common blockers that require human intervention to resolve.

A flawed channel ranking could lead to cutting a high-performing source or scaling a low-quality one, with real budget and pipeline consequences. Errors here are not immediately visible and may take quarters to surface, making them costly and slow to reverse.

Deciding which channels to scale vs. reduce is a strategic call that depends on competitive positioning, team capacity, brand goals, and risk tolerance — none of which live in the data. The agent can surface what happened; it cannot reliably say what should happen next.