Multi-Objective Optimization of Retail Promotional Ordering with Probabilistic Forecast

Decisions regarding optimal order quantity for retail promotion can be formulated as a single period newsvendor problem. Inaccurate forecast, limited access to financial information and packaging constraints are some of the major practical limitations of the single period newsvendor problem. We address all of the above 3 limitations and develop a multi-objective optimization problem with lost sales and leftover inventory as two major conflicting objectives. We propose a 2-stage method to solve the problem. The first stage defines various service level thresholds through inventory classification. The second stage solves the optimization problem with an e-constrained like method. We propose the use of discrete probabilistic forecast, and compare the results with those obtained using point forecast. The results for a real world problem indicate that both solutions outperform the existing ordering policy and the probabilistic approach outperforms the later. Results from the probabilistic approach show 39% reduction in lost sales and 27% reduction in leftover inventory.