ML Optimization System for Paid Marketing

Atlassian's ML Optimization System for Paid Marketing has been enhanced with a two-stage ML paradigm for optimal budget allocation. The system comprises a payout model that predicts business outcomes as a function of marketing spend, incorporating global model transfer learning and capturing spend diminishing returns. A budget optimizer then uses these predictions to find the best allocation across all segments simultaneously, maximizing overall business value under real-world constraints. The system operates with offline and online layers and a continuous learning loop, enabling dynamic budget refreshes, cross-channel global optimization, constraint-aware optimization, and multi-objective optimization across funnel metrics. Future work includes better extrapolation handling, incorporating prediction uncertainty into optimization, and feature enrichment.