DoubleClick for Publishers forecasts use a sample of the last 28 days of your network’s impression activity as a basis to model future traffic patterns. The forecasting simulation pits the prospective line item for which you’re running a forecast against booked line items, using the 28-day sample as the expectation of how much and in what ways inventory will be available.What data is included in the sample file?
Samples are collected on an anonymized, per-user basis (for ads served on the web, users are identified by their unique cookie ID). This allows you to create forecast estimates that include frequency caps. Keywords and key-values are also captured as part of the sample, so you can run a forecast against combinations of targeting criteria--for example, keywords, geography, and site targeting. (Learn more about targeting criteria and forecasting.)
Because forecasts are based on sample data, the narrower the slice of inventory for which you create a forecast, the greater the variance in results. In other words, forecasts tend to be less reliable as targeting criteria get more specific. This is especially true for networks and sites that get less traffic.
Forecasting counts an unfilled impression as belonging to the ad unit size to which it served. For example, if an unfilled impression serves to a 728x90 ad unit and the impression is captured in the sample file, then forecasting counts the impression against that size.
The following scenarios describe whether or not delayed impressions are counted toward forecasting:
- Counted: A regular ad request (not a delayed impression).
- Counted: A delayed impression request that does not return an ad (i.e., it's an unfilled impression).
- Counted: A delayed impression request that returns an ad and also has its impression pinged (i.e., it's a filled impression).
- NOT Counted: A delayed impression request that returns an ad but does NOT have its impression pinged. This is also called an unviewed impression.