With Amazon Forecast, you pay only for what you use; there are no minimum fees and no upfront commitments. There are four different types of costs to consider when using Amazon Forecast:

- Imported data: Cost for each GB of data imported into Amazon Forecast for training and forecasting.
- Training a predictor: Cost for each hour of infrastructure use required for building a custom predictor based on your input data or for monitoring predictor performance. Training time includes time taken to clean your data, train multiple algorithms in parallel, find the best combination of algorithms, calculate accuracy metrics, generate explainability insights, monitor predictor performance, and infrastructure use of forecast creation.
**Note that costs are based on the number of instance hours used, not the actual clock time it takes to train a predictor.**Because Amazon Forecast deploys multiple instances in parallel to train a predictor, the number of hours used will exceed the actual clock time observed. - Generated forecast data points: Cost for number of unique forecast values generated across all time series (items and dimensions) combinations. Forecast data points are the combination of number of unique time series (e.g., SKU x stores), number of quantiles and the time points within the forecast horizon. Forecasted data points include those created by generating forecasts, and those produced through what-if analyses.
- Forecast explanations: Cost for explaining the impact of attributes or related data on your forecasts for each item and time point. Explainability helps you better understand how the attributes in your datasets impact your forecast values. The cost is based on the number of forecast data points and the number of attributes (e.g., price, holidays, weather index) being explained.

## Free Tier

For the first two months of using forecast, customers receive up to 100,000 forecast data points per month; up to 10 GB of data storage per month; and up to 10 hours of training per month.

## Pricing tables

Cost Type |
Pricing |
Details |

Imported data | $0.088 per GB | For each GB of data imported into Amazon Forecast. |

Training a predictor | $0.24 per hour | For each hour taken to clean your data, train multiple algorithms in parallel, find the best combination of algorithms, calculate accuracy metrics, generate explainability impact scores, monitor predictor performance, and create forecasts. Amazon Forecast deploys multiple instances in parallel to train a predictor, so the number of hours used will exceed the actual clock time observed. |

Generated forecast data points | *See tiered pricing Table 1 below | For every 1,000 forecast data points at each quantile to generate forecasts, including what-if analyses. Forecast data points are rounded up to the nearest thousand. |

Forecast Explanations | **See tiered pricing Table 2 below | For every 1,000 explanations – forecast data points multiplied by the number of attributes (such as Price or Holidays). Explanations are rounded up to the nearest thousand. Each explainability job has a limit of 50 time series and 500 time points. |

***Table 1**: Generated Forecasts Data Points tiered pricing table

Generated forecast data points per month |
Price per 1000 forecast data points |

First 100K forecast data points | $2.00 |

Next 900K forecast data points | $0.80 |

Next 49M forecast data points | $0.20 |

Over 50M forecast data points | $0.02 |

*Note: Customers generating forecasts using a predictor which has been trained with the legacy CreatePredictor API will continue to get charged $0.60 per 1,000 time series, which is the combination of items and dimensions, for reach quantile. Forecasts are rounded up to the nearest thousand.*

*** *Table 2:** Forecast Explanations tiered pricing table

Forecast Explanations per month |
Price per 1000 explanations |

First 50K explanations | $2.00 |

Next 950K explanations | $0.80 |

Next 9.9M explanations | $0.25 |

Over 10M explanations | $0.15 |

## Pricing examples

### Pricing example 1 - Product Demand Forecasting

Let’s say you own a clothing company and have 1,000 items selling in 50 stores around the world and are forecasting for product demand for the next 7 days at 1 quantile. Each combination of an item and store location equates to one time series, so you’ll have 50K (1000 items x 50 stores) time series to forecast. Since you are forecasting at 1 quantile, you are forecasting for a total of 50K forecasts (50K time series x 1 quantile). At 7 days ahead forecasts with a weekly forecasting frequency, you are forecasting for 1 data point in the future with a total forecast data points of 50K (50K forecasts x 1 data point).

Cost Type |
Pricing |
Usage Cost |

5 GB of data imported | $0.088 per GB | 5 GB x $0.088 per GB = $0.44 |

3 training hours | $0.24 per hour | 3 hrs x $0.24 per hr = $0.72 |

50K forecast data points | $2 per 1000 forecast data points for the first 100K forecast data points | 50K forecasts x $2 per 1000 forecasts = $100 |

Total Cost = $101.16 |

Now let’s assume the following change: You are now forecasting 7 days ahead forecasts with a daily forecasting frequency. This translates to forecasting for 7 data points in the future with a total forecast data points of 350K (50K forecasts x 7 data points).

Cost Type |
Pricing |
Usage Cost |

5 GB of data imported | $0.088 per GB | 5 GB x $0.088 per GB = $0.44 |

3 training hours | $0.24 per hour | 3 hrs x $0.24 per hr = $0.72 |

350K forecast data points | $2 per 1000 forecast data points for the first 100K forecast data points $0.80 per 1000 forecast data points for the next 900K forecast data point |
100K x $2 per 1000 forecast points = $200 Total = $200 + $200 = $400 |

Total Cost = $401.16 |

*Pricing example above is based on a single forecasting job in a month*

### Pricing example 2 - Capacity Planning

Let’s say you own an energy company. You have 5K resident customers who use both gas and electricity. Each combination of resident customer and types of energy equates to one time series, so you’ll have 10K (2 energy types x 5K resident customers) time series. Let’s assume you need to plan 24 hours ahead with an hourly forecast at 1 quantile, so you are forecasting a total data points of 240K forecast data points (10K time series X 1 quantile x 24 hours).

You are adding a Price attribute and have selected to add the Holidays and the Amazon Forecast Weather Index built-in datasets for predictor training. Let’s say that you are interested in learning what attributes are driving forecasts for your top 100 gas customers. The cost for forecast explainability will be as follows.

Number of explainability jobs | 100 customer time series / 50 time series maximum per explainability job = 2 |

Number of forecast data points being explained per explainability job | 50 resident customers x 1 energy type x 1 quantile x 24 hours = 1200 |

Number of attributes being explained | Price + Holiday + Weather Index = 3 |

Total number of explanations in one month | 1200 x 3 x 2 = 8000 (rounded up to nearest thousandth) |

Total cost |
$2/1000 explanations x 8000 explanations = $16 |

*Pricing example above is based on a single forecasting job in a month*

## Additional pricing resources

Easily calculate your monthly costs with AWS

Contact AWS specialists to get a personalized quote

Refer to developer guide for instructions on using Amazon Forecast.

Instantly get access to the AWS Free Tier.

Get started building with Amazon Forecast in the AWS console.