Systems and methods are disclosed for automated generation and adaptive refinement of data-driven presentation content using a multi-agent workflow architecture. A campaign processing system retrieves advertiser-level and campaign-level performance data, applies eligibility criteria to identify actionable opportunities, and generates structured presentation content including visualizations, narrative recommendations, and data tables organized into discrete content slots. Generated content is presented to a human reviewer for approval or rejection. Upon rejection, an adaptive feedback engine employing a large language model extracts structured feedback elements from unstructured user input and may iteratively request clarification. Structured feedback is aggregated at slot, advertiser, and program levels to selectively regenerate content and improve subsequent generation workflows