Case Study: Flieber achieves 34% forecast error reduction with Nixtla's TimeGPT

Flieber cuts forecasting error by 34% with Nixtla

Flieber, a multichannel inventory planning platform, faced significant complexity in its demand forecasting. Its previous system required a large ensemble of statistical models that were resource-intensive to maintain and took hours to run. The challenge was to accurately forecast daily demand up to a year ahead across over 20,000 product-level time series, accounting for holidays, events, and sparse data. To address this, Flieber partnered with vendor Nixtla to implement a custom TimeGPT model.

Nixtla's solution was a unified TimeGPT forecasting engine tailored to incorporate holidays, domain knowledge, and temporal reconciliation. This custom model replaced the previous complex ensemble. The implementation by Nixtla resulted in a 34% reduction in forecast error and slashed the forecasting runtime from five hours to under five minutes. This greatly improved operational scalability and provided Flieber's customers with more efficient inventory planning.

Flieber