Retailers are increasingly vulnerable to overstocking when their demand forecasts are inaccurate, especially for recently launched products with insufficient sales data. This issue highlights a critical need for more sophisticated predictive models in the retail sector.

Ian Yong, a senior consultant at Wavestone Asia, shared insights at the Retail Asia Summit 2026 in Hong Kong, revealing that one beauty retailer's new product forecasts were nearly double the actual consumer demand. This discrepancy resulted in a surplus of slow-moving and ultimately obsolete merchandise. Yong emphasized the difficulty beauty retailers face in forecasting demand for new items, given their frequent product introductions and the lack of historical sales figures for merchandisers to rely upon.

To tackle this challenge, Yong's team implemented artificial intelligence to analyze similar products and their characteristics, including type, market relevance, pricing tiers, and seasonal demand. This AI-driven approach generated a demand curve for new products, guiding the retailer on procurement volumes, timing, placement strategies, and order quantities.

Furthermore, Yong explained that AI can also detect trends in product demand, signaling when seasonal interest or initial excitement for a new item begins to wane. Jeanne Pigassou, Wavestone's senior manager and AI lead, added that traditional methods for allocating products to stores often rely on merchandisers' experience and intuition, which can lead to biases and knowledge gaps when experienced staff retire or move on.

Yong cited a successful collaboration with fashion retailer Zadig & Voltaire, where a K-means algorithm was used for store clustering. This strategy, incorporating 28 numerical features such as customer profiles, store environments, and sizes, boosted sell-through performance by ten percentage points. He noted that additional factors like average air-conditioning temperatures and proximity to public transport could also be integrated into these models, while stressing that human oversight remains crucial for interpreting AI outputs and making informed decisions.

Retailers face distinct forecasting hurdles depending on their supply chain models. Push-based systems involve deciding initial stock allocation without historical demand, common for new products. In contrast, pull-based models use customer purchases to generate inventory signals, guiding replenishment. Despite these differences, Pigassou underscored the universal need for retailers to strike a balance between product availability and inventory levels. Yong also observed that many retailers continue to manually calculate safety stock and inventory signals on a monthly or quarterly basis, despite having access to vast amounts of data.

Looking ahead, AI is expected to automate more routine merchandising tasks, such as processing customer orders and generating purchase orders. This automation would free up merchandisers to focus on more strategic initiatives, including creating demand, developing business rules, and managing exceptions, thereby enhancing overall operational efficiency.