Pulse Forecasting FAQ
This FAQ is designed to answer frequently asked questions regarding Pulse forecasting.
Is there a commonly accepted time period for a new client to start using the Planner?
The forecasting engine uses a sample of actual user traffic (ad requests) from the previous two weeks when forecasting into the future so it needs two weeks of reasonably normal traffic data to get fully up and running. Reasonably normal traffic is the result of having a fully functional ad serving integration with Pulse for all traffic where you want to serve ads through Pulse during a period where you do not expect significant slumps or peaks in the traffic to your site.
Why just two weeks of historical user data?
This time span, along with other configuration parameters, was derived after researching various time spans and it outperformed the others in terms of accuracy. With longer periods, the forecaster reacts too slowly if traffic levels change, and with shorter periods it reacts too quickly. The research proved that using a simple moving average model based on the previous two weeks outperformed the other time spans in terms of accuracy.*
Why can I not forecast one year into the future?
Forecasting further than three months into the future does not make sense, because we have little information on what is going to happen then.
Do you make any changes to the simulation engine so it runs correctly for a particular client?
Yes, we change the sample rate (the number of users we capture, expressed as a ratio such as 1:100) to optimise between speed and accuracy, depending on the size of the client’s inventory.
How long does a simulation take?
We aim for simulations to take between 30 seconds and a few minutes, depending on the size of your query, where longer forecast periods take more time.
How accurate are the forecasts?
Forecasting the near future is more accurate than forecasting further into the future. 95% of the forecasts match the actual outcome when forecasting one or two days into the future. 85-90% of the forecasts match the actual outcome when forecasting one or two weeks into the future. We constantly work on improving accuracy.
How do I find all goals that are targeting a specific audience?
Why are disabled campaigns and its goals included in the forecasts?
We do not want surprises in available inventory when disabled campaigns are suddenly switched on. There is an “Exclude from forecast” setting available in the Campaign overview which enables you to exclude a campaign and all its goals from the forecasting results.
Why do I have less inventory in the campaign simulation than in the inventory simulation?
The inventory simulation report shows how much inventory there is in total, not considering campaign delivery rules and campaigns competing for the same inventory. Insertion policy settings limit available inventory for a specific campaign. For example, if there are three pre-rolls set for each pre-roll break, only one of the positions can be used by a campaign, but the inventory report would consider this as three possible impressions.
Why does the progress bar switch from green to yellow?
The projected delivery of a campaign or goal is embedded into the progress bar and updated often. When new information comes into the system, for example, changes in traffic or newly booked campaigns, the progress bar updates accordingly. For example, if a large high-priority campaign is booked, other campaigns which are running during the same period might not be able to deliver, and the progress bar turns yellow.
How is traffic seasonality accounted for?
You can use Simulation Adjustments to factor in significant anticipated or past traffic deviations in your forecasts.
Can I forecast on frequency capping?
Can I forecast on devices?
Yes. Select the ad format and device group you wish to target and Pulse simulates it for you.
Can I forecast on audience segments?
Yes. Select the audience segment you wish to target and Pulse simulates it for you.
Why can I not forecast on IP targeting?
Due to legal restrictions, we cannot store information about users' IP addresses. However, we can store geographic information down to city level, which can be used for geographic targeting.
Why can I not forecast on browser/OS targeting?
We currently do not store this information in our logs.