Why It Matters
Accurate, scalable forecasting is critical for businesses ranging from retail inventory to energy grid management, where mis‑predictions can cost millions or disrupt essential services. Understanding the capabilities and limits of new foundation models helps data teams adopt more efficient tools while maintaining safeguards against costly forecasting errors.
Key Takeaways
- •Foundation models enable scalable time‑series forecasting across millions of series.
- •Retail inventory planning gains cost savings via accurate demand forecasts.
- •Outlier events remain hard for models to predict reliably.
- •Productionizing forecasts requires rigorous backtesting and risk management.
- •Hybrid models outperform single approaches for diverse business use cases.
Pulse Analysis
The episode highlights a paradigm shift from decades‑old statistical time‑series methods to large‑scale foundation models. Rami Krispin explains that traditional TS objects and ARIMA‑type algorithms struggle with the exponential growth of high‑frequency data generated by devices and logs. By training on massive, heterogeneous datasets, foundation models such as Chronos, Moirai, and Google’s offerings can ingest thousands of series simultaneously, delivering forecasts with far less manual wrangling. This scalability is especially relevant for enterprises that need to predict demand across hundreds of thousands of SKUs or monitor energy consumption patterns in real time.
Business leaders hear concrete value: more precise inventory allocation, reduced warehousing and spoilage costs, and smarter capacity planning for utilities. Retail giants like Walmart and Amazon can fine‑tune stock levels for bread, ketchup, or seasonal items, while energy firms avoid costly under‑ or over‑production. The conversation underscores that even modest accuracy improvements at scale translate into millions of dollars saved, making AI‑driven forecasting a competitive differentiator across supply‑chain, retail, and infrastructure sectors.
Despite the promise, Krispin warns that foundation models still stumble on rare, outlier events—concert tours, World Cups, or sudden policy shifts—that can dramatically distort demand. Deploying forecasts in production therefore demands rigorous backtesting, cross‑validation, and a risk‑management framework that prioritizes high‑impact series. A hybrid strategy—mixing traditional statistical models, machine learning, and foundation models—often yields the most reliable results, allowing teams to benchmark, iterate, and select the optimal approach for each unique time series.
Episode Description
Time series data is everywhere — from inventory systems and energy grids to financial planning and product demand. As data volumes grow, the old ways of building individual forecasting models simply don't scale. How do you forecast hundreds of thousands of products without spending months on manual modeling? How do you know when to trust automation and when to step in? And what does it actually take to produce forecasts that business stakeholders will act on?
Rami Krispin is Senior Director of Data Science and Engineering at Apple Finance, where he leads teams working at the intersection of statistical modeling, machine learning, and production forecasting. He is the author of Hands-On Time Series Analysis with R, an open-source contributor, Docker Captain, and instructor. He holds an MA in Applied Economics and an MS in Actuarial Mathematics from the University of Michigan, where he began his journey learning time series on DataCamp — before going on to build his own course there.
In the episode, Richie and Rami explore time series foundation models and the case for scaling, traditional versus modern forecasting approaches, feature engineering in the business world, backtesting and model selection, risk management in automated forecasting, communicating forecast uncertainty to stakeholders, the evolving role of data scientists as architects, and much more.
Links Mentioned in the Show:
Forecasting: Principles and Practice (Rob Hyndman)
Nixtla
skforecast
Prophet
Connect with Rami
AI-Native Course: Intro to AI for Work
Related Episode: Developing Better Predictive Models with Graph Transformers
New to DataCamp?
Learn on the go using the DataCamp mobile app
Empower your business with world-class data and AI skills with DataCamp for business
Comments
Want to join the conversation?
Loading comments...