My mentor at HP Inc., Cara Curtland, my PhD advisor Dr. Chuanren (Charles) Liu, and I represented Team HP at the Foresight Practitioner Conference 2025, held at UNC Charlotte. Our work was selected as one of the top five global finalists for the International Institute of Forecasters’ Forecasting Practice Competition, and we presented to the judging panel and the conference for the 2025 Impact of Forecasting in Practice Award.

The talk describes HP’s worldwide print demand forecasting system, over 18,000 products across more than 170 countries, and how we moved from simplistic statistical methods to an integrated, enterprise-grade pipeline built on LightGBM, robust machine learning operations (MLOps), and a human-in-the-loop design. The recurring theme is that the system delivers tangible business results, including lower inventory, precisely because machine learning is paired with human insight rather than pitted against it.

As far as I could tell, I was the only student presenting; everyone else came from a forecasting team inside a company, which made it a wonderful place to learn. The other finalists’ problems were fascinating in their own right: a German team forecasting electricity grid load, Maersk forecasting the repositioning of empty shipping containers, and several others.

This is joint work with Chuanren Liu (University of Tennessee, Knoxville), Cara Curtland, and Adam Ghozeil (HP Inc.).