Google's new AI model predicts the future from sales data, weather, and discount schedules
Google's TimesFM-3 fills all future time points in one pass—cutting compute time and compounding errors in forecasting.
Why it matters
A new foundation model for time-series prediction that handles multimodal context (sales, weather, promotions) in a single forward pass. Practitioners in supply chain, retail, and demand planning would adopt this for faster, more accurate forecasts; enthusiasts tracking Google's model releases see a targeted capability play against specialized forecasting vendors.
The key facts
11 to knowModel: TimesFM-3 (Google Research)
Parameter count: 330M
Approach: single-pass forecasting instead of step-by-step (reduces compounding error)
Inputs: time series + external context (weather, promotions, discount schedules)
Output: all future time points in one inference
Model: TimesFM-3, 330M parameters
Approach: parallel decoding (single-pass future fill) vs. sequential step-by-step prediction
Benefits: reduced compute time, lower compounding error
Input modalities: time series + exogenous features (weather, promotions, known future events)
Use case domains: sales forecasting, demand planning
Source: Google Research
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Google Research has released TimesFM-3, a forecasting model that analyzes time series alongside related data and known future events like sales promotions or weather forecasts. Instead of predicting the future step by step, the 330-million-parameter model fills in all future time points in a single…