#356 The Forecast for Time Series Forecasts with Rami Krispin, Senior Manager of Data Science at Apple

In this episode of DataFramed, host explores the transformative impact of time series foundation models on forecasting with Rami Krispin, Senior Manager of Data Science at Apple. Rami discusses how traditional time series methods like ARIMA and ETS have long struggled with scalability and real-world complexity, especially in large-scale retail and infrastructure applications. He highlights the emergence of foundation models from companies like Amazon (Kronos), Salesforce (Moirai), and Google, which can process vast datasets more efficiently than ever before. However, he cautions that these models still face critical challenges—particularly in handling rare, non-recurring events like Taylor Swift concerts or the World Cup—since they cannot predict future anomalies without external context. Rami emphasizes that while automation is advancing rapidly, human expertise remains essential for feature engineering, model validation, and risk management, especially in high-stakes domains like energy forecasting where errors can have serious societal consequences. The conversation shifts to practical workflows for deploying forecasts in production, stressing the importance of data pipelines, backtesting, and continuous monitoring. Rami advocates for transparency with stakeholders, using visualization tools like STL decomposition and prediction intervals to communicate uncertainty and build trust. He also reflects on the evolving role of data scientists, who are transitioning from coders to architects, leveraging AI tools to accelerate development while maintaining control over model design and interpretability. Ultimately, Rami underscores that foundational statistical knowledge remains crucial—even as AI automates more tasks—because understanding the 'why' behind models enables better decision-making and innovation.