How line item delivery is paced
When a line item starts its campaign, DoubleClick for Publishers (DFP) creates an initial delivery plan that it then modifies based on changes in the system over time.
The ad server uses the following step-by-step methodology to create a plan to pace the delivery of your line items. Each step modifies the number of impressions the system sets for its daily and hourly delivery goals. Click through each step to read more about what calculations are made to serve your line items.1. Initial calculations
After a frontloaded or even delivery line item is booked, DoubleClick for Publishers (DFP) calculates how many impressions it should attempt to serve for the line item per day. This calculation is simply the number of impressions booked for the line item divided by the number of days in the line item's campaign. These goals represent truly even delivery, as shown in Chart 1 below.
Imagine a week-long campaign with 7,000 scheduled impressions. Perfectly even delivery would result in 1,000 impressions being served each day of the week.
Next, if your network is enabled to use expected traffic shape in pacing, DFP looks at the last 28 days of the network's average shape (derived from overall traffic to the network) to create an initial delivery plan. (Our systems use past delivery data that is independent from the online forecasting simulation.)
If the line item has been running for less than two weeks, we use your network data averaged by day of week to pace delivery.
- If the line item has been running for two weeks or more, we use the the past delivery pattern of the line item itself to pace delivery.
For example, if traffic to the targeted inventory is very high during the evenings and on weekends, DFP expects to serve more impressions at these times and fewer during slower times. This allows DFP to deliver impressions smoothly relative to the natural traffic fluctuations on your site, and prevents under-delivery when a campaign ends during a period of low traffic.
The graph below illustrates this point.
Applying past traffic shape in pacing uses past traffic patterns to your websites to predict trends and pace ad delivery. For example, imagine a website whose traffic is highest between 4:00 PM and 1:00 AM each day. The ad server will serve fewer impressions during the low traffic hours in the early morning and will be more aggressive during the high-traffic evening hours.
Next DFP applies a frontloading factor to the line item goals. As described above for frontloaded line items, this factor can be as high as 40% at any one time, but averages to 25% over the course of the first half of the campaign. This compares to about 5% above goal during the first half of a campaign for even delivery. These frontloading factors are applied on an iterative basis, which means that an even delivery line item on the first day will serve 5% higher than the delivery goal. On the second day the line item will serve 5% higher than the goal for the second day, where the second day goal is slightly lower than the first day goal due to the 5% addition on day one.
This continues through the life of the line item such that, all other factors being equal, there is a small decrease in the impressions served each day. For line items set to "Evenly," this decrease will be very small in most cases. For line items set to "Frontloaded," it will be more apparent.
When a line item is set to deliver evenly, the ad server is only allowed to exceed the daily goal by 5%. For frontloaded delivery, the ad server can exceed the goal by as much as 40% at any one time, but averages to 25% over the first half of a campaign, and returns to evenly delivered during the second half. This results in more impressions being served toward the beginning of the campaign than would serve with perfectly even delivery, and fewer being served at the end. This helps ensure that delivery doesn’t fall short of the line item's goal due to lack of inventory at the close of the campaign.
Applying this frontloading factor to our running example we get the serving goals illustrated in Chart 4.
When ad serving combines traffic shape (if enabled for your network) with frontloading, delivery will align with traffic patterns but will exceed the goal briefly by as much as 40%, and 25% during the first half of a campaign. Again, the result is that more impressions are served toward the beginning of the campaign to avoid falling short at the last moment.
Changes over time
Over time DFP has to adjust the goals for your line item based on the actual number of impressions it has served. The originally calculated goals, as outlined above, are only valid when the server delivers exactly to the goal every hour. Click through each step below to see which calculations are updated and modified throughout delivery.1. Satisfaction Index
DFP calculates a satisfaction index (SI) to measure how far ahead or behind schedule a line item is. Specifically SI measures how well a line item has performed over the last 24 hours, relative to the goals set for it by the ad server. The SI is then used during the line item selection process so that line items that are further behind are selected more frequently. If you are not generally familiar with SI and how it is used in the line item selection process, please see the ad selection whitepaper in this help center for more information.
Rather than selecting a line item directly based on SI, DFP instead sets a different threshold for each line item. The SI for the line item must exceed this threshold in order for the ad server to select it amongst all of the line items that accepted the ad server's request. This threshold is also calculated once an hour and is controlled by the Proportional Integral Derivative (PID) controller.
The PID controller attempts to make a line item react more gradually to changes in its environment such as unexpected changes in your site’s traffic levels. Instead of dramatically decreasing the threshold every time the SI starts to fall, the PID controller instead takes a more gradual approach. When setting the threshold it considers both how far away the line item is from a perfect SI and how quickly the SI of the line item is changing. The net effect is that your line items will deliver more evenly than they did in the DoubleClick ad server.
Since the PID controller’s adjustments are gradual, it will take your line item several hours to fully correct if it gets ahead or behind. Although this may seem odd in the short term, it significantly improves delivery over the life of the line item.
DFP regularly recalculates the daily and hourly delivery goals. This follows the same process outlined for a new line item above, but uses updated information. Impressions are allocated based on the total remaining impressions and the total remaining time in the line item's campaign. For line items that have been live less than 24 hours, these delivery goals are updated every hour. After the first day the delivery goals are updated every eight hours. The goals are updated more frequently for new line items to allow the server to calibrate itself based on how the new line item is delivering.
If a line item has fallen behind schedule or has been paused for a period of time, then begins delivering again, DFP tries to make up the missing impressions during the next 24 hours, then deliver the remaining impressions evenly over the remaining days (assuming that the line item is set to deliver evenly).
For example, a line item has an impression goal of 100,000 over 10 days. The line item delivers its goal of 10,000 impression on each of the first two days, then is paused for four days. On the seventh day, the line item is restarted; it has 80,000 impressions still to deliver. DFP will try to deliver approximately 50,000 impressions on the seventh day: 10,000 impressions for each day that the line item was paused, plus 10,000 impressions for the current day. (Because of the 5% frontloading that's applied to line items set to spread evenly, the exact impression goal for each day would vary.)
If your network is enabled to use expected traffic shape in pacing (see the section above), the ad server also takes hints from past delivery patterns when determining how many times to serve a line item in a given time window.
For networks that are enabled to use expected traffic shape in pacing, OSI will take the traffic-shaped ad server goals into account and will represent a more accurate measure of delivery status. Still, always be sure to check the forecast for a line item before making any delivery changes.
During the final six hours of a fixed-goal line item that hasn't yet reached its goals, DFP begins adjusting the PID controller to ensure that the goal is met. In the last hour, a line item that still needs to deliver impressions will function as if it were set to deliver as fast as possible.
If a campaign is behind schedule, it may even serve ahead of higher-priority campaigns that have recently started delivering and are expected to deliver in full.