Overview

Ghost Bidding is a testing framework designed to measure the incremental impact of media buys without incurring the cost of control group impressions. Within this framework, certain bids are excluded from auctions and designated as "ghost bids." These ghost bids do not win auctions, but are still tracked for testing purposes. Buyer Cloud predicts whether these ghost bids would have won the auction if they had been included. If the system forecasts that a ghost bid would have won, it generates a corresponding "ghost impression." These ghost impressions effectively create a control group consisting of users who are treated as though they had been exposed to the ad, even though they did not actually see it.

Using this control group, Buyer Cloud compares the conversion rates of users who saw the ad with those who did not. This allows for an analysis of the incremental lift, which reveals how much of the observed conversion behavior can be attributed to the ad exposure itself, as opposed to other factors. The results of this analysis are available in the Incrementality Report, which can be accessed through Reporting. For those interested in more granular control, ghost logs are also available. These logs provide transparency into the ghost bidding process, offering detailed data for those who wish to experiment with or refine elements of the methodology.

How Ghost Bidding Works

Treatment and Control Group Assignments

• Ghost bidding assigns 80% of users to a treatment group (real bids) and 20% of users to a control group (ghost bids).
  
  
• Bid requests get assigned a user index (1 to 1000) using a hash based on the User ID type. Requests with a user index from 1 to 200 are put in the control group; requests with an index between 201 and 1000 are put in the treatment group.
  
  
• Standard IDs, IP Addresses, or FreeWheel IDs are used for identifying the ghost bidding User ID Type.

• • For Standard Ghost Bidding User ID Type, Buyer Cloud uses cookies.
•  

  • For the IP Address Ghost Bidding User ID Type, Buyer Cloud uses both IPv4 (IP version 4) and IPv6 (IP version 6). However, if both types of IP addresses are present in a bid request, IPv6 will take priority over IPv4.

    Additionally, if a bid request contains obfuscated IP addresses (where the IP is hidden or masked), that request will not be eligible for line items using IP Address as the Ghost Bidding User ID Type. In other words, only bid requests with visible, non-obfuscated IP addresses can be used for Ghost Bidding based on IP address.

•  
  • For FreeWheel ID Network Ghost Bidding User ID Types, Buyer Cloud uses the user ID as determined by the FreeWheel Identity Network. For more information on how the FreeWheel Identity Network identifies users, see Attribution Powered by FreeWheel Identity Network.

Ghost Bidding

• All requests go through targeting, creative matching, pacing, and frequency capping to see if a line item bids on it.
  
  
• If a line item bids and the user has been assigned a user index between 1 to 200, then the bid will be dropped (not sent to the exchange) and logged as a ghost bid. 

Win Prediction

• Buyer Cloud predicts whether a ghost bid will win and become a ghost impression. To make these predictions, Buyer Cloud trains win rate models daily for each line item on the previous day's data. For the model to perform well, it needs sufficient data. The model requires at least 5,000 impressions from the previous day to be effectively trained.
  
  

• The model training process runs daily, therefore, ghost bid logging will begin only after the model has been trained and deployed for the line item. Ghost bid logs are typically viewable the day after enabling the feature.
  
  

Treatment vs Control Group Conversion Attribution

• When a conversion event is received, it runs attribution methodology across both the treatment and control groups' impressions to determine which group was responsible for driving the conversion.

  This is not supported for in-app/postback/SKAd attribution methods because these attribution processes are run outside of Buyer Cloud.

Activate Ghost Bidding on a Line Item

Ghost bidding is enabled at the line item level and can be configured during creation or when editing the line item. To enable ghost bidding on a line item:

1. Navigate to a line item's General screen (either during or after creation of the line item).
   
   
2. Scroll down to the Ghost Bidding section and expand it.
   
   
3. Set the Enable Ghost Bidding toggle to "YES," then choose the appropriate Ghost Bidding User ID Type—either Standard, IP Address, FreeWheel ID Network Person, or FreeWheel ID Network Household—which will be used to categorize users. See below for more details on each ID type.
   
   
4. Click Save.

Ghost Bidding ID Types

Ghost bidding can be configured to use one of the two different ID types:

When ghost bidding is enabled on a line item, it implicitly targets bid requests with a specific user ID type. This means Buyer Cloud will bid on requests that include the selected user ID type and filter out those that do not. For example, if the user ID type is an IP address, Buyer Cloud will filter out bid requests where no IP address is present or where an anonymous IP address is used

Reporting & Logs

Incrementality Report

In reporting, the Incrementality Report calculates the incremental lift and other key metrics across both treatment and control groups, including impressions, conversion, conversion value, and conversion rate. Incremental Lift is calculated using the following equation: (Treatment group conversion rate – Control group conversion rate) / Control group conversion rate.

Ghost Logs

Ghost win logs are offered for users who would like to run their own attribution methods, and ghost attribution logs for users to run their own incrementality analysis.