A business demand generation system may generate a demand score based on typical demand and predicted future demand from a business segment mix for a future time period. The demand score is processed through a action recommendation model to generate an actionable recommendation that is tied to an attribution model that evaluates the impact of the recommended action on actual segmental demand of a business. The evaluated impact is fed back through the recommendation model to create a continuous learning loop that refines the model to improve future recommendations to drive better and more accurate results.