We started using Amazon Forecast when it was released as a public beta and after just a few days we realized that this service can really allow us to build forecasting models with a very high accuracy and in a very short time. By applying Amazon Forecast to some different real-life scenarios, we were able to get accuracy levels higher than 90%, with heterogeneous time series gathered from different sources which were often dirty or incomplete.
As an example, we built predictors for energy consumption forecasting by combining energy usage and weather historical data and for detecting outages of IoT connected devices well in advance compared to the traditional monitoring methods.
Uses ML to automatically discover patterns among data which could be apparently unrelated.
Can be applied to time series from any industry (retail, finance, pharma, manufacturing, telco, etc.).
All data and contents processed are protected and encrypted by default.
Can achieve in hours, forecasting accuracy levels that usually require months.