When you run a traffic or inventory forecast, Ad Manager is working behind the scenes, running a model of the ad server to determine how contending line items will get delivered across the forecasted inventory. The projected values are estimates, reference a sample of the overall data, and are not guaranteed. Because of subtleties in how Ad Manager simulates the ad server, you may see some numbers that you don’t expect.
Explanation of sampling: Sampling is a technique used to collect data from a population. In the context of forecasting, sampling can be used to collect data from a small number of inventory segments and use that data to generate a forecast for the entire network. However, if the inventory segments that are sampled are not representative of the overall network, the forecast may not be accurate.
Following are some of the most frequently asked questions about forecasting.
NOTE: We’re constantly working to improve our forecasting system. If you have questions that aren’t answered below, contact support and include specific numbers, URLs, and so forth. We’ll analyze your issue and determine if the system is working as intended, or if there’s a way we can update the system to improve the numbers.
- Which line items does the forecasting model take into account?
- Is forecasting available for video?
- Why do two reserved line items with the same targets but different goals have different forecasting numbers?
- Why do new reservations impact existing reservations?
- Does forecasting take seasonal trends into account?
- Does forecasting take frequency capping into account?
- Does forecasting take user domain targeting into account?
- What happens to the forecasting numbers when you add new key-values or new ad units?
- Why aren't impressions forecasted to all special ad units?
- Why don’t forecasting numbers match the future sell-through report?
- I have two line items with the exact same targeting: one is sponsorship, and the other is standard. Why are the matching values the same?
- When is forecasting updated?
- Do inventory forecasting numbers affect the pace at which line items are delivered?
- Why isn’t a contending line item appearing in the contending line items list?
- Why is the "Expected to deliver" number different from "Likely to deliver" or "Available"?
- Is forecasting available for cross-sell networks?
Which line items does the forecasting model take into account?
The forecasting model mentioned above only takes standard and sponsorship line items (guaranteed line items) into account when determining which line item gets allocated for each forecasted impression. Note that guaranteed line items with a status of "Paused" or "Draft", as well as line items with no creatives, can also be allocated impressions.
Is forecasting available for video?
Yes, you can check available inventory for video ad units. Note that forecasting on optimized pods isn't supported.
Why do two reserved line items with the same targets but different goals have different forecasting numbers?
Even though the line items have the same targeting, the Available, Reserved by same-or-lower priority line items, and Unavailable numbers can be different. This is largely a result of the surpluses and shortages of impressions caused when lowering the priority of a line item for the first step of the forecast simulation. When that happens, other line items can claim the impressions that otherwise would have gone to the forecasted line item, thus leaving other impressions unused.
Why do new reservations impact existing reservations?
Example: I made a reservation in March, and it was fully booked. Later, I made a reservation in May, and Ad Manager told me I had to overbook. Why does the May reservation impact my reservation from March?
Our system’s goal is to optimize toward the completion of all existing reservations, so overbooking one reservation impacts all existing reservations that are at equal or lower priority.
Does forecasting take seasonal trends into account?
Yes, we consider holidays and other yearly patterns in our calculations.
- Holiday seasonality: The system predicts an increase or decrease in traffic during major holidays.
- Year-over-year seasonality: The system detects patterns that repeat annually. Note that we apply this only when there’s a strong signal of the pattern.
Does forecasting take frequency capping into account?
Yes, frequency capping is considered.
Does forecasting take user domain targeting into account?
Yes, but you first need to traffic at least one saved line item to a user domain target. After that first line item has had some time to traffic, the forecasting numbers will be up to date and will fully include user domain targeting.
What happens to the forecasting numbers when you add new key-values or new ad units?
The forecasting numbers will be smaller than expected until enough time has passed since adding new ad units, or since adding key-values to ad tags on your website. In the meantime, you can manually adjust the forecast in an attempt to get numbers that more closely reflect the new ad units.
Why aren't impressions forecasted to all special ad units?
If an ad unit is designated as special, no line item will serve that ad unit unless it’s been explicitly targeted. Therefore, when you check available inventory for a line item, the forecasting numbers will only include special ad units that have been explicitly targeted.
Why don’t forecasting numbers match the future sell-through report?
The "Unassigned" number in a future sell-through report is often larger than the "Available" number in a forecast. If two ad slots are on the same page, "Unassigned" includes both ad slots in its count. "Available" takes into account the fact that an ad can’t appear in two ad slots at once, so it only includes one ad slot in its count.
I have two line items with the exact same targeting: one is sponsorship, and the other is standard. Why are the matching values the same?
Whether an impression matches a line item depends on the targeting of the line item, and not its type (sponsorship vs. standard), priority, and so forth.
When is forecasting updated?
Forecasting is updated:
-
After creating a new line item: about 60 minutes.
-
After making changes to a line item: anywhere between 1 minute and about 60 minutes depending on edits.
-
After making network configuration changes (for example, Placements, Protections, Key-values, and so forth): up to 1 day.
Do inventory forecasting numbers affect the pace at which line items are delivered?
No. The ad server uses the steps described here to create a plan to pace the delivery of your line items. When determining the pace, the ad server doesn’t consider forecasts from the same system used to check available inventory.
If your network is enabled to use expected traffic shape to help determine pace, the ad server will analyze data to understand how traffic to the targeted inventory will look over the life of the line item. However, this is independent from the forecasts that are used to check available inventory. The main difference is that, with expected traffic shape, future line items are not considered.
Why isn’t a contending line item appearing in the contending line items list?
Any line item with fewer than 100 contending impressions isn't shown in the list of contending line items. In addition, a maximum of 3000 contending line items can appear on the page.
Why is the "Expected to deliver" number different from "Likely to deliver" or "Available"?
“Expected to deliver” is calculated differently from “Likely to deliver” and “Available”:
- “Likely to deliver” and “Available” are numbers that can appear when checking available inventory, either through the Forecasting button or Forecast tab. These numbers show the number of impressions your line item is predicted to win without impressions that would be taken away from contending line items. Therefore, these are usually the more conservative numbers, as they don’t take into account impressions that the line item could win when competing against contending line items.
- The “Expected to deliver” column appears on pages that list line items and is part of forecasting for underdelivery. This column portrays the actual number of impressions that this line item is predicted to win, including impressions that would be taken away from contending line items. This is usually the more aggressive number, as it does take into account impressions that the line item could win when competing against contending line items.
We believe it’s useful for publishers to see both of these numbers. If they want to know how many impressions they can book without adversely affecting other line items, they can use the “Available” or “Likely to deliver” number. If they want to know how many impressions the line item is actually going to win, even if it means taking away impressions from other line items, they can look at “Expected to deliver”.
We’re constantly evaluating our forecasting numbers and what they mean to you, to ensure that we’re giving you the most helpful numbers possible. Look for continued improvements to our forecasting numbers in future releases.
Is forecasting available for cross-sell networks?
Forecasting for cross-sell networks is supported and is based on filled ad impressions (Google and Partner-sold) and unfilled ad impressions—not on views or ad requests.
- Not every view results in an ad request to Ad Manager
-
Not every ad request counts towards historical traffic used in forecasting. Excluded requests include:
- No ad impression & VAST error (the line item is returned, but an error occurs)
- No ad impression & no VAST error (a user loads the page but leaves before the first frame loads)
When troubleshooting forecasting for cross-sell, compare the results with filled and unfilled impressions data (unified pricing rules reports or sell-through reports) to see if they align.
When forecasting, comparison with video views from YouTube Analytics should be avoided. There is no direct connection between these two metrics (forecasting is not based on views). The ratio between video views and forecasted impressions can be impacted by multiple reasons, such as video type (for example, shorts), monetization settings, length of content, number of monetized non-viewed playbacks (counted ad impression without a counted view due to user abandonment), or ad blockers.
