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

2026-04-20 · Source: DataFramed · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Advanced, extended

Summary

Rami Krispin, Senior Director of Data Science and Engineering at Apple Finance, discusses the evolution and scaling of time series forecasting, particularly with the emergence of time series foundation models. Traditional methods, like ARIMA and Holt-Winters, struggle with the exponential growth of data from sources like electronic devices and retail SKUs, necessitating a shift towards machine learning and foundation models. These new models, exemplified by Amazon's Chronos and Salesforce's Moirai, are trained on vast datasets to handle diverse use cases at scale, enabling businesses like Walmart to optimize inventory and capacity planning for hundreds of thousands of products. Krispin emphasizes the critical role of feature engineering in business forecasting, the importance of backtesting for model selection and stability, and the need for robust risk management and transparent communication of forecast uncertainty to stakeholders, especially in critical applications like energy capacity planning.

Key takeaway

For Data Scientists and MLOps Engineers building forecasting systems, you should embrace time series foundation models to handle large-scale data, but critically combine them with rigorous backtesting and feature engineering. Prioritize transparent communication of forecast uncertainty and model limitations to business stakeholders to build trust and ensure informed decision-making, especially for high-stakes applications like energy capacity planning.

Key insights

Time series foundation models enable scalable forecasting for massive datasets, moving beyond traditional methods.

Principles

• Scale is critical for modern time series analysis.
• Feature engineering drives business forecasting success.
• Backtesting ensures model stability and consistency.

Method

The workflow for production forecasting involves data pipeline building, model experimentation with backtesting, model deployment, and continuous monitoring for data drift and accuracy degradation.

In practice

• Benchmark foundation models against simpler alternatives.
• Use prediction intervals to communicate forecast uncertainty.
• Decompose time series (e.g., STL) to identify trends and anomalies.

Topics

• Time Series Foundation Models
• Production Forecasting
• Backtesting
• Feature Engineering
• Forecast Uncertainty

Best for: Data Scientist, MLOps Engineer, Director of AI/ML

Related on AIssential

• Interview with AAAI Fellow Yan Liu: machine learning for time series — Physics-informed foundation models enable accurate, adaptable time series analysis, especially in data-scarce scientific and real-world applications.
• Extra #6 - Taking Your RNN From Beginner to Pro — Robust model validation requires more than visual fit, focusing on reproducibility, input visualization, error analysis, and baseline comparison.
• How to Use a Time Series Foundation Model for Anomaly Detection — A single, general time series foundation model can replace many narrow, problem-specific anomaly detectors.
• 7 Steps to Mastering Time Series Analysis with Python — Mastering time series analysis requires a sequential, structured approach, from understanding unique data properties to robust deployment.
• FAME: Forecastability-Aware Mixture of Experts for Heterogeneous Time Series Forecasting — FAME uses forecastability fingerprints and a sparse router to dynamically select optimal experts for heterogeneous time series.
• google-research / timesfm — TimesFM 2.5 is a more efficient, capable time-series foundation model with reduced parameters and extended context.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by DataFramed.