Integrating Bidding-Based Auction With Existing Waterfall

The purpose of a transparent and successful bidding process is to find the highest paying ad among all demand sources in real time. This requires two things: 1) Giving each demand source equal opportunity to bid for every impression and 2) Allowing for an increased level of competition between demand sources. If these requirements are not met, the result can be lost revenue for the publisher and lost bid opportunities for the advertiser.

An all-bidding integration is ideal for maximizing publisher revenue. However, since some demand sources do not yet have bidders, it is often necessary to combine bidding and waterfall systems. This document provides one example of doing such an integration in a correct way. It also provides one counter example of doing such an integration in an incorrect way that can result in revenue loss. Keep in mind that an all-bidding system will outperform a mixed waterfall and bidding solution so you should strive to move completely to bidding as soon as possible.

Correct Integration Example

Key Characteristics

  • Running bidding in parallel with waterfall.
  • Granular price floors for each demand source in waterfall.

An optimal integration allows every bidder to have access to every impression opportunity and makes the existing waterfall more competitive by giving each demand source a better chance to compete for different CPM price levels. Both waterfall and auction run in parallel to reduce latency.

The implications of this setup are as follows:

  • By giving bidders access to every impression opportunity, the ad demand sources are making better informed decisions an provide their best bid every time. This improves auction results and can increase publisher revenue.
  • Granular price floors in the waterfall allow demand sources that don't yet have bidders to simulate real-time bidding albeit at a few preset price points. Instead of using one average CPM per demand source, each demand source now has different price points, as determined by the price floors, enabling better competition between waterfall demand sources and bidders for each impression opportunity.

    For example: DS1 will compete with both DS2 and DS3 for impressions with CPM values above $20. If none of them responds with a fill, they will compete again for a CPM value above $10 and so on. This way, each impression opportunity has a better chance of fetching the highest price from the waterfall. Then that price from the waterfall can compete with the winner of the auction to determine the overall winner across waterfall and auction.

    Using multiple price floors per demand for non-bidders is required because demand sources are given more information about the target price of that particular impression and can make a more informed decision on whether they can serve an ad with the respective minimum value. This more closely simulates a true bidding environment even with waterfall-based demand sources.

Correct Usage of Price Floors

  • Price floors cannot be applied (will have no effect) to the Facebook Audience Network bidder. The bidding price will be based on the impression/user evaluation strictly and will not take into account any minimum threshold set by the publisher.
  • Static price floor must not be applied specifically to the auction as a way to merge it with the waterfall. Both the waterfall and auction need to compete equally for each impression and neither must start with a disadvantage. The only correct price floor that can be set to the auction prior to the merge with the waterfall is the winning price of the waterfall (highest price for which a demand source provides a fill).
  • Price floors can be used as an opportunity cost to only allow profitable ads into the user experience (for example the user is playing a game in the app and showing an ad for which the CPM is 10 cents will disrupt the user experience without bringing significant revenue to the publisher). In the case of the above model, this price floor can be applied to the two winning prices (winner from the waterfall and winner from the auction) to enforce this opportunity cost.

Incorrect Integration Example

Key Characteristics

  • Fewer than 3 price floors for waterfall demand sources.
  • Using the auction as the first demand source in the waterfall.

One way of wrongly integrating bidding into a waterfall setup is by adding the auction winner as the first demand source in the waterfall and using CPM averages for other demand sources.

Although this setup correctly gives all bidder demand sources access to each impression opportunity, it has several negative results:

  • The demand sources below auction demand source are at a disadvantage because they are not called unless the auction cannot fill. This can result in revenue loss as waterfall demand sources might be able to pay higher CPMs for some impression opportunities but they won't be able to if the auction fills them.
  • Other demand sources are classified based on one average CPM value per demand source. This also has a negative revenue impact since a demand source average does not accurately reflect the price it might have paid for that impression opportunity. Also, a demand source might decide to run a temporary campaign with higher prices for specific impressions and being positioned in the waterfall based on historical average CPM may not allow the demand source to see higher-value impressions that are filled higher up in the waterfall. A better way of using demand sources within a waterfall is by having granular price floors as per above model.

The graph below shows how the average is calculated: by taking the number of impressions for each CPM value into account and aggregating all different prices into one static value.

For this example the average of $1.5 CPM is average based on total revenue divided by total impressions. In case this is the reference in a waterfall, all potential CPMs valued at more than $2 (220 impressions for an average CPM of $6.5) will be lost. Moreover, since the CPMs that are higher than $2 will probably no longer fit into the revenue matrix, the average CPM will also decrease in time.

In conclusion, integrating bidders with this setup does not maximize publisher revenue and does not reflect the benefits of bidding.