Ghost Ads: for a transparent and robust programmatic experiment

Isolate the impact of each marketing action in the buying process

The Challenge

With the multiplication of contact points in the conversion path, Sofinco questions the profitability of its investments on Ad Exchanges: how many people would have naturally turned to Sofinco and converted without exposure to the brand’s messages?

The objective is to precisely measure the real contribution of this marketing stimulus in the conversion path to ultimately optimize the budget allocation on the most efficient channels and mechanisms.

The solution

To find out how effective its marketing investments really are, Sofinco began a large-scale experiment with Numberly.

The steps to set up such a project are as follows:

  • Mapping of existing populations and marketing treatments
  • Identification of the issues and indicators to be monitored
  • Construction of the protocol
  • Monitoring and reporting

This path led us to define a test within this framework that leverages a feature of Xandr, the “Ghost Ads“.
Concretely, a test population is exposed to Sofinco “auto journey” banners while the control population is created from simulated exposures: we collect information that allows us to know that a cookie could have been exposed without really exposing this person, this is done without paying the cost of “false exposure”.

The Results

We were all very pleased with the results. This method allows us to isolate a reliable and robust control group, so that we can accurately measure the impact of the campaign on conversions. We measure a 15% increment in conversion rate between the test group and the control group.


Relative difference in conversion rate between test group and unexposed control group of 15%.

Numberly awarded

This collaboration with Sofinco was rewarded with a Gold Trophy in the Insights category at the Trophées Marketing 2021.