Skip to main content

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

Richie and Rami explore time series foundation models and the case for scaling, feature engineering in the business world, communicating forecast uncertainty to stakeholders, the evolving role of data scientists as architects, and much more.
· 53 min read
Rami Krispin

Guest

Rami Krispin

Rami is a leading thinker on forecasting at scale. He is the author of "Hands-On Time Series Analysis with R" and the forthcoming "Applied Time Series Analysis and Forecasting" as well as the DataCamp course "Designing Forecasting Pipelines for Production".

Richie Cotton

Host

Richie Cotton

Richie helps individuals and organizations get better at using data and AI. He's been a data scientist since before it was called data science, and has written two books and created many DataCamp courses on the subject. He is a host of the DataFramed podcast, and runs DataCamp's webinar program.

Chat with AI Richie about every episode of DataFramed - all data champs welcome!

Key Quotes

We are becoming more architects or designers rather than writing code and building modules from scratch. Think about someone doing architecture for buildings — they understand the engineering, they understand the materials needed to make it stable. This is the same transition we're going through, where we are vibe coding and becoming more designers rather than hard coders. And in this transition, when you understand what's happening under the hood, you can get better results.

If you remember a few years ago, Zillow were relying on their forecast to estimate the growth of prices — and they got it wrong. That really damaged the stock price. They were relying on an auto forecaster, and when it's doing great, it's great. But you need to understand that when you're using auto algorithms, sometimes you get it wrong, and this is where you need to think about how to mitigate those risks. Blaming the algorithm isn't fair. It's risk management rather than just algorithm.

Key Takeaways

  1. Foundation models for time series are best suited for scale — when you have hundreds or thousands of series to forecast, they reduce manual effort significantly, but accuracy trade-offs are real and need to be actively managed.

  2. Feature engineering remains a critical skill even as AI automates more of the forecasting pipeline. Knowing how to represent events, seasonality, and trend shifts in your data still separates good forecasts from bad ones — foundation models haven't solved this.

  3. Risk management matters as much as model accuracy when forecasting at scale. Not every series carries the same weight — identifying your highest-priority forecasts and applying more scrutiny there is a more effective strategy than treating every series the same way.

Transcript

Richie Cotton: Hi Rami, welcome to the show. 

Rami Krispin: Thank you. I'm happy to be here. Thank you for the invitation. 

Richie Cotton: Yeah, great to have you here. So the hot new thing with time series is Time Series foundation models. So similar to we have foundation models for text. We've now got them for time series, but time series modeling has been around for decades now.

Why do we need foundation models for it? 

Rami Krispin: I think the short answer is scaling. When I started to work on time series data. Back years ago I use R and then there were the common Time series object was the Ts. For those who remember, this is a very notorious one that you can, it was built for a time that people use the monthly or quarterly dataset.

And since then we. Started to collect data in an exponential manner and think about it like your phone or any electronic device collecting logs. And so now when we are working with time service data, we have, this is probably the most common format of data, and we have more than we can actually process.

So trying to handle it with the traditional methods that are not very scalable, it's very challenging, and I think that was in recent years, we moved from the statistical models that are really great to what it was designed, very structural time series and to more like machine learning approach.

And now we are seeing the models that are based on foundation models that are taking it to the next lev

Topics

Related

podcast

[AI and the Modern Data Stack] Adding AI to the Data Warehouse with Sridhar Ramaswamy, CEO at Snowflake