As depicted in the figure below, there are two data types required for executing accurate forecasts: log files ans campaign information.
Historical log files are required for forecasting, which should initially be provided dating at least one month back in time. Daily server activity should be provided on a daily basis to a common storage. Log files should be sampled at the user level and aggregate information from multiple event channels, with ad requests, ad opportunities, RTB pricing and any other relevant information provided in the same log row.
In order to keep the forecast results accurate, all active campaigns in the client’s system must be synchronized with the Ad Forecaster. This is done using the Campaigns API. These campaigns should also be periodically updated with their delivered impressions.