What is Forecasting?
A forecast is a prediction of future conditions based on past and present data. Pulse continually collects data about all ad requests and responses, the 'past' data, which is used for the forecasting functionality in the system. The 'present' data used in forecasting is the current state of all campaigns, goals, ads, and other settings on your account. In this context, Pulse forecasts future ad inventory based on historical user traffic, and current campaign and account settings.
The Simulation Engine
Pulse uses a simulation based forecasting approach, as opposed to a calculation or mathematical model based approach. This means that during a forecast, the forecasting engine simulates ad delivery by fast-forwarding the same ad decisioning engine, which will later deliver the ads, through a future inventory based on historical user data. It also includes the programmatic performance by looking at historical win rate and calculating the average settlement price for every programmatic marketplace competing in the forecast, and resolves the selection decision accordingly. The result is more realistic and accurate than a traditional calculation or mathematical model based approach.
- forecasting inventory,
- forecasting how much inventory is reachable for a new goal you want to run, and
- projecting delivery for running campaigns and goals.
The first two uses are available through the Planner and are discussed in the following sections. For the last use, see Progress Bar with Projected Delivery. For an overview of the Pulse distribution engine, see Pulse Distribution Engine.
Simulation and Engine Details
- The forecasting engine can project up to 100 days into the future. This means that forecast requests are limited to date ranges between tomorrow and 100 days from tomorrow. The further into the future the forecast looks, the more uncertainty exists in the results. We have determined that 100 days provides for an optimal balance between accuracy and performance for the forecasting system.
- The forecasting engine uses a sample of actual user traffic (ad requests) from the previous two weeks when forecasting into the future.
- Simulations and their results are kept for 14 days. Given how often both traffic patterns and account settings (campaigns, goals, and ads) change, results of forecasts more than a few days old tend to lose their meaning and relevance. For this reason, the length of time that results are kept is limited.
- Simulation Adjustments may affect your forecasts. It is recommended to see if any future events coincide with the date range of your forecast and if any past events occur within the sample window for forecasts.
- For campaign simulations, we assume the goal always 'fits' within a time-limited break. Depending on the length of ads used in an actual campaign, the forecast could appear to show more available impressions than the campaign can actually deliver in a time-limited break.
- You can use the Exclude from forecast option on any campaign to exclude a campaign and all its goals from the forecasting results and from projecting delivery, for example for filler and in-house campaigns. It allows you to see what inventory is available for sale and does not affect ad delivery. For more information, see Campaign Overview.
Campaign forecasts do not yield meaningful results if you have eCPM enabled on your account, because it is not possible to enter a price in the forecast. As a result the simulated campaign has zero as a price and only ever gets picked in the simulation forecast when none of the other campaigns with a price takes the spot. This skewed result will not match reality, when you were to enter and run the campaign in the system. See Campaign Settings for information on enabling and disabling eCPM.
Simulation adjustments enable you to factor in significant anticipated or past traffic deviations in your forecasts. For example, if you published videos that caused a significant increase in traffic compared to your normal traffic, then this extra traffic is going to influence all forecasts. To compensate for this traffic deviation, you mark the extra traffic in a simulation adjustment and the simulation engine then compensates for it.
For a list of frequently asked questions, see Pulse Forecasting FAQ.