High-Frequency Pricing at Scale for E-Commerce

A new algorithmic pricing tool deployed at online fashion retailer Zalando delivered approximately 6% higher profit during sales campaigns while keeping sales and revenue steady, according to a paper posted to arXiv. The system replaces a hybrid manual-algorithmic approach with a forecast-then-optimize architecture that cuts pricing decision time from hours to minutes. The tool combines daily-resolution demand forecasting using gradient-boosted trees with a multi-objective optimization framework that maximizes long-term profit and net merchandise value across more than 5 million articles [1]. It was validated through 23 A/B tests conducted across 12 markets during the 2023-2024 sales campaign cycle [1]. The paper states the algorithm now handles the majority of algorithmic pricing decisions for sales campaigns at the company [1]. Zalando, founded in 2008 and headquartered in Berlin, operates in 25 European markets and reported revenue of 10.3 billion Euro in 2022 [7]. The company employs roughly 17,000 people as of early 2023 [8]. In July 2025, Zalando acquired a majority stake in Hamburg-based fashion online retailer About You, which is active in 26 European countries [9]. The shift from a weekly-granularity pricing system to daily-resolution decisions addresses what economists call menu costs — the expenses firms incur when changing prices, including administrative planning, system updates, and consumer communication [3]. By reducing the time required to set prices from hours to minutes, the tool lowers these operational frictions [1]. Fashion e-commerce amplifies such challenges: earlier research on Zalando's platform noted small stocks, high return rates, seasonality, and continuous article turnover as persistent difficulties for automated systems [6]. The paper's authors frame the work as a response to the unique demands of sales events, where volatile demand patterns require rapid pricing decisions that balance short-term revenue against long-term profitability [1]. The experimental results showed the new system achieved the profit gain while maintaining equivalent performance on sales and revenue compared to the prior approach [1].