Adjust forecasts for upcoming traffic patterns

DoubleClick for Publishers currently uses 28 days of prior traffic for forecasting. As a result, forecasting works optimally after four weeks. If parts of your website have traffic patterns that aren't captured in the default 28-day sample because new content was created or you have traffic spikes during events that recur more than 28 days apart (for example, a yearly sports tournament or a semi-annual event), you can adjust forecasts either manually or historically to better reflect your traffic patterns.

Forecasting criteria fall into two categories: adjustable, which allow you to describe traffic volume per-criteria, and non-adjustable, which use the 28-day sample to determine impression distribution per traffic criteria.

Adjustable criteria

  • Ad unit
  • Ad size
  • Day of week

Examples of non-adjustable criteria

  • Key-values
  • Time of day
  • Number of ad units on a page
Although you can narrow down your adjustments to specific ad unit sizes, we recommend that you do so only when necessary. The more specific you get with your adjustments, the more moving parts you're introducing into the forecast, which slows down all queries for the network.

Manual adjustments

Manual adjustments are typically used for new ad units that do not yet have a history or for existing ad units that have special events that may affect traffic. You can, however, adjust the forecast for any ad unit, regardless of how new it is or whether there is a change in its traffic.

Example of using a manual adjustment for a new ad unit

To successfully sell a new ad unit, you need to know how many impressions it will receive. You know that the content of the web page on which the ad unit is placed closely resembles the content of another existing web page. You expect the new web page will have a very similar traffic pattern. You decide to manually adjust the forecast for the new ad unit for the first 28 days until DFP can build its 28-day sample.

To do this, you can run an ad unit report for the past 28 days on the existing ad units and use it to estimate how many impressions the new ad unit will need.

To create a manual adjustment:

  1. In the "Inventory" tab, click Forecast adjustments, then New adjustment.

  2. Enter a name for the adjustment, the ad unit you're adjusting, and the Future Date range to which the adjustment applies.

  3. In the "Type" field, select Manual.

  4. In the "Adjust" field, select either:

    • Date range total: Enter the number of impressions you want to spread out over the adjustment period.
      For example, if you enter 100,000 total over time for the month of January, then DFP will assume 100,000 total impressions for all of January and the forecasting sample will distribute that total between days.

    • Daily total: Enter the number of impressions you want to adjust for each day of the adjustment period.
      For example, if you enter 3,000 per day for the month of January, then DFP will assume 3,000 impressions per day, or a total of 93,000 impressions for the month (3,000 x 31 days in January).

  5. Enter the total amount of impressions by which you want to adjust the forecast. You can adjust all sizes for every tag allowed by the ad unit, a single size, or sets of sizes allowed by multiple-size ad tags. Here's how:

    • All sizes: Simply enter the total amount of impressions in the impressions total box.

    • Single size: Click Adjust specific sizes. Enter the single size you want to adjust. Enter the total amount of impressions by which you want to adjust the forecast for that size in the impressions total box.

    • Multiple-size ad tags: Click Add size set. For each set, enter all of the sizes the ad tag allows. Enter the total amount of impressions by which you want to adjust the forecast for the size set in the impressions total box. If you want to adjust additional sizes, either individually or as part of a size set, click Add size set again.

    If you don't enter any ad sizes, the forecasting simulation assumes that impressions are distributed among different ad sizes in the same proportions as they were served during the previous 28 days.

  6. Click Save. Forecasts will start reflecting the adjusted quantity in about 30 minutes.

Understanding multiple size ad tags and adjustments

Imagine you have a leaderboard on your site that's always 728x90, and a right-hand box on a multi-size ad tag that can be either 300x600 or 300x300. If you’re expecting an increase of 100,000 impressions for that page, you'd make two different adjustments:

  • one with the 728x90 on its own line with an adjustment of 100,000, and

  • one with the 300x600 and 300x300 in the same box, with an adjustment of 100,000.

When you adjust for multi-sized ad tags, you tell our forecasting system that you expect the traffic increase to come from some combination of the sizes you specify. That means that if you enter an adjustment for 100,000 for two sizes in a multiple-size ad unit, you're telling the forecasting system, "We expect 100,000 ad requests that simultaneously allow 300x600 or 300x300 sizes.”

If, after you’ve made the above adjustment, you run a forecast for a 300x300 ad size, the forecast returns a result of 100,000 available impressions; if you run a forecast for 300x600, you'll also get back 100,000. If you run a forecast for 300x600 OR 300x300 you still get 100,000, not 200,000—because the requests won't be double-counted.

Historical adjustments

Sometimes anticipated changes in traffic, such as for an event similar to one that happened a year ago, are not reflected in the 28-day sample. You can adjust the forecast of upcoming events by creating a historical adjustment. Historical adjustments combine the impression data from the past time period with ad request data distributed by ad size over the last 28 days, and apply that to the upcoming event.

Example of using an historical adjustment

Let's say that your website is about the travel industry and you're booking inventory for summer travel a few months in advance. You know that your traffic will skyrocket four weeks before the start of summer and you fear that if you only rely on the 28-day sample to book inventory, DFP may underestimate the amount available. You can use a historical adjustment to base this year's forecast on last year's summer travel season.

Additionally, you know that overall impression levels for your web site have been increasing since last year. You can enter a percentage of the historic total or trend over time in your adjustment to account for the increase. For instance, if your traffic has been increasing by 20%, you can adjust the forecast by 120%.

To create an historical adjustment:

  1. In the "Inventory" tab, click Forecast adjustments, then New adjustment.

  2. Enter a name for the adjustment, the ad unit you're adjusting.

  3. In the "Type" field, select Historical.

  4. Select the start date of the Future Date range to which the adjustment will apply. (When you select the historical date range, below, the end date of the future adjustment date is automatically selected.) The maximum historical time period is one year.

  5. Select the Historical Date range you want to use for the sample (up to 18 months in the past if your network has data going that far back), and click Apply. The future date range is updated to match the time period of the historical date range.
  6. In the "Adjust" field, select Weekly totals (trend).

  7. Enter the percentage of total impressions from the historical period that you want to spread out, on a week-by-week basis, over the adjustment period. (For non-adjustable criteria, forecasting uses the last 28 days of history to scale impression data.)

  8. For example, if you enter 110% total over time for the first four weeks of January based on the first four weeks of last January's volume, then DFP applies all impressions from those four weeks, plus 10%.

  9. Adjust either all sizes for every tag allowed by the ad unit, a single size, or sets of sizes allowed by multiple-size ad tags. Here's how:

    • All sizes: Enter either the percent of historic total or percent of historic trend in the impressions total box.

    • Single size: Click Adjust specific sizes. Enter the single size you want to adjust. Enter the percentage by which you want to adjust the forecast for that size in the impressions total box.

    • Multiple-size ad tags: Click Add size set. For each set, enter all of the sizes the ad tag allows. Enter the total amount of impressions by which you want to adjust the forecast for the size set in the impressions total box. If you want to adjust additional sizes, either individually or as part of a size set, click Add size set again.

    If you don't enter any ad sizes, the forecasting simulation assumes that impressions are distributed among different ad sizes in the same proportions as they were served during the previous 28 days.

  10. Click Save. Forecasts will start reflecting the adjusted quantity in about 30 minutes.

Additional factors

When you add or delete second-level ad units or ad sizes, associated forecasting adjustments might be affected.

Adding or removing second-level ad units

Forecast adjustments set on first-level ad units: Forecast adjustments set on a first-level ad unit are evenly distributed to child second-level ad units. If you then add or remove second-level ad unit, the adjustment you’ve set for the first-level ad unit remains unchanged, and its adjustment is redistributed evenly to account for the additional (or subtracted) second-level ad units.

Example: adding a second-level ad unit (Cricket) to an adjusted first-level ad unit.

Before: Sports (adjusted = 150k imps)

  • Baseball (apportioned = 75k imps)
  • Basketball (apportioned = 75k imps)

After: Sports (adjusted = 150k imps)

  • Baseball (apportioned = 50k imps)
  • Basketball (apportioned = 50k imps)
  • Cricket (created after adjustment = apportioned 50k imps)

Forecast adjustments set on second-level ad units: The sum of forecast adjustments set on second-level ad units equal their parent first-level ad unit’s adjustment. (Impressions for any non-adjusted second-level ad units are based on the standard 28-day sample.) Therefore, if you delete a second-level ad unit, its forecast adjustment is removed too, and so is no longer accounted for in its parent first-level ad unit.

Example: removing an adjusted second-level ad unit (Cricket) from an unadjusted first-level ad unit.

Before: Sports (sum of adjustments = 200k imps)

  • Baseball (adjusted = 75k imps)
  • Basketball (adjusted = 75k imps)
  • Cricket (adjusted = 50k imps)

After: Sports (sum of adjustments = 150k imps)

  • Baseball (adjusted = 75k imps)
  • Basketball (adjusted = 75k imps)
  • Cricket (removed after the adjustment = 50k imps removed from first-level ad unit)

Adding or removing ad sizes

For multi-size ad units, you need to explicitly enter the adjusted impressions per eligible ad size. For example, if the ad tag has sz=300x250, 300x600, then enter adjustments for each size. Failure to do so can inflate forecast results because the simulation adds natively forecast impressions to any sizes you don’t explicitly adjust. In this example, that means if you enter 1000 impressions only for the 300x250 ad size, the simulation would add an adjustment according to the 28-day sample file for the 300x600 size. If the sample called for 2000 impressions, the total would be 3000 for that ad unit.

Forecast adjustments set on a per-size basis: When you adjust a multi-size ad unit on a per-size basis, forecasts for ad sizes added after the adjustment use the standard 28-day sample, and existing per-size adjustments are not affected by the addition of the new ad size. Likewise, if an adjusted ad size is removed, its adjustment is removed with it.

Example: ad size added to an adjusted ad unit that’s also adjusted by size.

Sports (adjusted ad unit)
  • 728x90 (is set to 50% of total)
  • 300x250 (is set to 50% of total)
  • 160x600 (created after the adjustment -- forecast is based on the standard 28-day sample)

Forecast adjustments set only on the ad unit level: If no per-size adjustment are made, then the standard 28-day sample is used to derive impression data for additional ad sizes.

Example: ad size added to an adjusted ad unit that’s not adjusted by size.

Sports (adjusted ad unit)
  • 728x90 (determined by 28-day sample)
  • 300x250 (determined by 28-day sample)
  • 160x600 (created after the adjustment -- determined by 28-day sample)

Counteracting adjustments

You'll generally make a manual adjustment because you're expecting a change in traffic volume due to a specific event. Keep in mind that the actual spike (or dip) in traffic will be included in the 28-day sample file, and so you might need to make an additional manual adjustment to counteract that spike's (or dip's) effect on the sample's rolling average.

Here's a scenario: you create an adjustment for a week-long sporting event, which you surmise will bring an extra 10,000 impressions per day to your site, in addition to the average of 5,000 per day, for a total of 15,000 impressions per day (the adjustment). The event comes to pass and brings an average of 8,000 extra impressions per day (for a total of 13,000 per day) for that week. If you were to run a forecast for the following week, the 28-day sample on which the forecast is based includes those extra 8,000 impressions per day. This would cause the forecast results to inflate the amount of inventory that's available in the coming weeks. To remedy this, create a manual adjustment that counteracts the effect the event has on the sample file.

In this scenario, four weeks of daily data would be:

5,000 + 5,000 + 5,000 + 13,000 = 28,000 / 4 = 7,000
impressions per day on average. So you'd want to create a manual adjustment of 5,000 impressions per day if you're not expecting any other events.

Maximum number of adjustments

The number of adjustments that your network can store at any one time is constrained by a maximum number of ad unit days, adjustment count, and impressions. Once you’ve reached the maximum number of adjustments, the system does not allow you to save new adjustments. As a remedy, you can delete existing adjustments, or lower the number of impressions or days for existing adjustments. Note that the limits below apply to the sum of historical and manual adjustments.

The system won’t let you enter forecast adjustments once any of the following limits have been reached in your network:

  • Max number of adjustments: 250,000
  • Max number of future days x adjusted ad units: 250,000
  • On any given day, the number of target impressions for the adjustments that are in effect exceeds the cap. The cap is calculated by: 10 x (over the past 28 days, the max number of ad requests the network as a whole had on a single day)

Multiplier: each size override multiplies the days and count for that adjustment. The multiplier is applied prior to the calculation of the max single-day impression count.

Example of breaching the max number of daily impressions

  • The highest daily requests for your network during the last 28 days was 1,000.
  • You create an adjustment for 9,000 requests for April 15th.
  • You create a second adjustment of 1,100 requests per day for the month of April, which the system rejects.

Reason for rejection:

  • The max number of daily adjustment requests is 10 x (max ad requests in the last 28 days) = 10 x 1000 = 10,000 impressions per day.
  • The adjustment of 9,000 impressions for April 15th allows for an additional adjustment of up to 1,000 impressions for that day.
  • Therefore, the attempted adjustment of 1,100 per day is 100 impressions too many for April 15th.

Solution: Change the adjustments so that the total for April 15th is less than or equal to 10,000.

If the above example were an historical basis rule, the adjustment impression count would be determined by the historical basis of the adjustment. If the basis time period had more than 10,000 impressions per day, the system would reject the historical basis rule. You would then need to select a different basis period with fewer impressions.