Abstract

Predicting demand for new fashion products has traditionally been more art than science, more guesswork than precision. Yet in an industry where inventory missteps can be catastrophic, “we think it’ll sell well” simply doesn’t cut it anymore.

In this talk, I’ll dive into how we are tackling this “cold start” forecasting problem. I’ll explore our feature engineering approaches for quantifying product similarities, our multi-tiered, “explainable-ish” modelling techniques, and our careful balancing act of blending machine learning methodologies with real-world domain expertise.

Speaker

Judit Kisistok

Data Scientist @ Hakio

Judit Kisistok, PhD, is a data scientist at Hakio, where she develops AI solutions to help fashion brands optimize their demand forecasting. With expertise in data visualization and AI explainability, she is passionate about transforming complex retail analytics into actionable insights. Her unique background, transitioning from academic research to fashion technology, brings a fresh perspective to predictive modeling that enables brands to reduce waste and respond more effectively to market trends.