AI can do the heavy lifting on prospect segmentation, but a human needs to own the strategy.

The data processing and tiering steps are structurally repeatable, but the call notes are unstructured and vary widely in quality and content, requiring adaptive interpretation each time. The 1-page strategy deliverable also shifts based on what patterns emerge.

Budget alignment and company size are relatively crisp inputs, but 'buying signals visible in the notes' and 'industry fit' require judgment calls that aren't fully defined. Without explicit scoring rubrics, the agent must make assumptions that a human might dispute.

The Salesforce export is a concrete, uploadable file, which is favorable. However, the agent needs the file actually provided, and it lacks access to live Salesforce, external market data, or the startup's internal ICP definition unless those are supplied explicitly.

Misclassifying a hot prospect as cold could mean a missed enterprise deal — real but recoverable if a human reviews before acting. The output is advisory, not directly executable, which limits blast radius.

Interpreting nuanced call notes, weighing soft buying signals, and crafting a credible prioritization narrative for a founding team all benefit from domain intuition. The agent can produce a solid draft, but a fractional ops director should validate the strategic framing.