AI can crunch 120 ad campaigns, but the budget call still needs a human in the room.

This is a quarterly recurring task with a consistent structure: pull data, rank campaigns, segment by metric, identify outliers. The schema is stable enough that an agent can run the same playbook each cycle with minimal reconfiguration.

Top and bottom 10% by metric is mathematically crisp, but 'what's working' and 'budget reallocation' are underspecified — the agent must make judgment calls about which metrics to weight, how to handle campaigns with low statistical significance, and what constraints apply to the reallocation.

Google Ads and LinkedIn both have APIs, but the agent needs authenticated access, a unified data schema across both platforms, and likely a data warehouse or export pipeline already in place. Without pre-built connectors, setup friction is high and data quality issues are common.

A flawed budget reallocation recommendation acted on without scrutiny could misallocate tens of thousands of dollars in ad spend for an entire quarter. The output is advisory, not directly executed, which limits damage — but a confident-sounding wrong recommendation is dangerous precisely because it looks authoritative.

Budget decisions depend on context the agent cannot access: upcoming product launches, sales pipeline health, brand positioning trade-offs, and executive risk tolerance. Pattern recognition on historical data is automatable; strategic prioritization is not.