Near-Perfect Pacing & A New Standard for Maximizing Publisher Revenue

The AppNexus Publisher Adserving platform has set a new standard for pacing and yield maximization of a publisher’s digital advertising dollars.

For the first time, Yieldex forecasting is fully integrated with the delivery engine of the AppNexus Publisher Adserver. This is not talking about graphs in the UI; forecasting data is now piping directly into the real-time pacing system of the publisher adserver.

With line-item specific forecasting data, the AppNexus adserver now provides near-perfect pacing precision while maximizing publisher yield on an impression-by-impression basis like never before.

How it works

Forecast-Shaped Pacing (FSP) allocates a line item’s hourly delivery goals according to the specific inventory forecast of that guaranteed line item. Is there proportionally much less inventory at 2am? Stop trying to deliver so much in that hour. Huge volumes of inventory during the middle of the day? Deliver more during that time. Inventory starts to drop off sharply at the end of the day, right as the line item is approaching its daily delivery target? Don't save such a large % of delivery for those last few hours.

Why FSP is Different

Legacy adserving systems continue to use "linear" or "even" pacing, striving to deliver the same amount of impressions each hour of the day. This naïve approach ignores the realities of inventory availability, and does a relatively poor job at pacing to delivery goals and even worse job maximizing yield for a publisher.

Best-in-market solutions (including the AppNexus Publisher Adserver up until the release of FSP) use a slightly less naïve approach. In an effort to set pacing goals according to future inventory availability, these other platforms will either use:

  1. a universal supply shape based on an account-level average, which can be a dramatically different shape than a specific line item's actual future availability, or
  2. the line item's historical delivery pattern, which is a generally a weak indicator of future ability to deliver.

To highlight these drawbacks, we took a look at some example account-level average supply shapes and super-imposed them on line-item specific forecasts. The example below is illustrative:

As the graph above makes clear, the difference between an account’s overall supply shape and a given line item’s actual forecast can be dramatic, and choosing to pace against the account-level curve can cost publishers revenue as well as increase the risk of inaccurate pacing.

The Benefits

So the theory makes sense, using a line item's actual forecast should be a superior approach. But what does the data say? In our extensive production analysis, we consistently see the following two key benefits:

Benefit 1 - Increased Yield From RTB

FSP increases a publisher's yield through Open Dynamic Allocation. By pacing according to a forecasted supply curve, guarantees don't have to "try as hard" to deliver impressions when inventory is scarce (ex: late night/early morning). In other words, the predicted CPM (pCPM) of the guaranteed line item can be significantly lower during these hours when there is less supply. This allows the publisher to take advantage of high CPMs from RTB throughout all hours of the day.

In our production testing to date, this relaxed pressure during low-supply hours has meant guaranteed line items using FSP are able to bid a 25-35% lower pCPM on average. A lower pCPM means the opportunity cost of serving guaranteed is lower and publishers capture more of their most valuable RTB demand, which translates to 15-20% more RTB revenue.

To see this dynamic in action, the visual below shows the bidding & pacing behavior for a line item in the days leading up to and following the release of FSP.

Benefit 2 - Better Delivery Throughout The Entire Day

FSP not only improves how often guaranteed lines items deliver in full, but greatly improves how frequently they deliver through the last hour of the day. Why is that? Because FSP is better at staying on pace, we no longer need the same safeguards to protect delivery. Historically, up to 20% acceleration was applied to under-pacing guaranteed line items to account for the fact that the target delivery curve might not be a “good fit” for actual supply. With FSP, we have been able to remove this acceleration mechanism: not only do FSP guarantees still deliver budgets in full more frequently than before, but now 15-20% more guaranteed line items deliver through the last hour of the day (a jump from 75-80% line items delivering through end-hour to now 90-95%).

What is Coming Next

All guaranteed line items on the platform are now pacing according to FSP for their entire flights, with the exception of the last day of delivery. We are currently in a closed beta testing "last day" FSP. Because the lifetime delivery goal is at stake, we use a more aggressive pacing algorithm for the last day of a guaranteed line item's flight. We are planning to release Last Day FSP logic later this quarter.