Posted On: May 21, 2021

Today we are excited to announce that Amazon Forecast can now generate predictions for 5X more items using 3X more historic data points, increasing the number of items and the amount of historic information that your forecasting models can learn from and predict.

Amazon Forecast uses machine learning (ML) to generate more accurate demand forecasts, without requiring any prior ML experience. Amazon Forecast brings the same technology used at to developers as a fully managed service, removing the need to manage resources or rebuild your systems.

Until today, customers were limited to generating forecasts for 1 million unique items based on the prior limits imposed by the CreatePredictor API. In some use cases, customers with a large number of items have either had to limit their model to a subset of items or generate separate models for specific items, categories, or locations. Both approaches require additional data preparation and may not fully leverage the inherent advantages of training deep learning algorithms with large data sets.

Today, we are raising the limit on the number of unique items that you can forecast from 1 million to 5 million for forecasting frequencies of daily, weekly, monthly and yearly. Additionally, we have also increased the number of maximum number of historical observations or rows that can be supported in the target time series dataset from 1 billion to 3 billion.

To review system limits see the Guidelines and Quotas page. You can use these new limits in all Regions where Amazon Forecast is publicly available. For more information about Region availability, see AWS Regional Services.