With Amazon Forecast, you pay only for what you use; there are no minimum fees and no upfront commitments. There are three different types of costs in Amazon Forecast:
Generated forecasts: A forecast is a prediction of future values for a single variable over any time horizon. For example, daily customer demand for a blue shirt at a single store location is one forecast whether you predict the next 10 days or the next 10 years. Amazon Forecast generates forecasts at three different quantiles: 10%, 50% and 90%.
Forecasts are billed in units of 1,000 (rounded up to the nearest thousand), and it’s likely that you will end up with multiple models that produce multiple forecasts. Additionally, forecasts are generated at multiple quantiles (10%, 50%, 90%) and your total forecasts will increase by a factor of 3. For example, the demand for every product a retailer sells is affected by a unique set of factors, so each product will typically have a unique model. It’s also typical to want each model to forecast product demand at the individual store level. Therefore, a retailer that sells 500 products in 20 stores will need to generate 10,000 forecasts (500 products x 20 stores = 10,000 total forecasts.) Since Amazon Forecast also generates forecasts at three quantiles, the example above will generate 30,000 unique forecasts.
Data storage: Costs for each GB of data stored and used to train your models.
Training hours: Costs for each hour of training required for a custom model based on data provided by customers.
As part of the AWS Free Tier, you can get started with Amazon Forecast for free today. For the first two months after first using Amazon Forecast, the Free Tier includes:
Generated forecasts: Up to 10K time series forecasts per month
Data storage: Up to 10GB per month
Training hours: Up to 10 hours per month
|Generated forecasts||$0.60 per 1,000 forecasts|
|Data storage||$0.088 per GB|
|Training hours||$0.24 per hour|
Pricing example 1 - Product Demand Forecasting
Let’s say you own a clothing company and have 2,000 items selling in 50 stores around the world. Each combination of an item and store location equates to one time series, so you’ll have 100k (2000 items x 50 stores) time series to forecast. Assume you need to upload 5 GB of data for this task, and a model with this dataset will take about 20 hours to train.
|Cost Type||Pricing||Usage Cost|
|100k time series forecasts||$0.60 per 1,000 forecasts||$180 (100 forecast units x 3 quantiles x $0.60 per forecast unit)|
|5 GBs of data stored||$0.088 per GB||$0.44 (5 GBs x $0.088 per GB)|
|20 training hours||$0.24 per hour||$4.80 (20 hrs x $0.24 per hr)|
|Total Cost = $185.24|
Pricing example 2 - Cash Flow Forecasting
Let’s say you own a financial consulting firm. Your customer owns 2,000 ice cream shops and wants to forecasts cash flow of each shop. Each combination of cash flow and shop location equates to one time series, so you’ll have 2,000 (1 cashflow x 2,000 shops) time series forecasts. Assume you need to upload 1 GB of data for this task, and a model with this dataset will take about 4 hours to train.
|Cost Type||Pricing||Usage Cost|
|2,000 time series forecasts||$0.60 per 1,000 forecasts||$3.60 (2 forecast units x 3 quantiles x $0.60 per forecast unit)|
|1 GBs of data stored||$0.088 per GB||$0.088 (1 GBs x $0.088 per GB)|
|4 training hours||$0.24 per hour||$0.96 (4 hrs x $0.24 per hr)|
|Total Cost = $4.648|