The setup was created to more precisely target and report against people who were actively in the market for a new car, to obtain a better seed audience for twinning (also into “walled gardens”) and to create a better currency to value the performance of VW’s on-going programmatic media buying.
Volkswagen is (now!) the biggest Automotive Brand In Denmark and together with them we wanted to change the way we evaluate and buy on an ongoing basis.
Digital campaigns in the automotive sector are often measured against their ability to create the highest CTR and the lowest Cost Per Action (Conversion, Download brochure etc.). This especially holds true for Programmatic where optimization means relying on retargeting, using inferred non-competitive expensive 3rd party data sets and some contextual prospecting on ad cluttered automotive websites. From a reach perspective this is not working well as there is a limited supply, and many users are not really in the market for a car, they are often interested in cars (also see results section).
As a proxy demographic audiences are sometimes used for buying and measurement, but the vast majority of users are not in the market and they see the ads as irrelevant and not for them.
We wanted to go in-between and find the users who were planning to buy but did not yet behave like they were. Through surveys we found that around 14% of the population consider themselves likely to buy a new car within the next 12 months. We wanted to make this the audience that VW should concentrate their efforts on, obviously with a viewable message.
We also wanted to migrate the "programmatic" audience to a new channel, Facebook
Together with our online survey provider (measures demographic reach and frequency for VW Campaigns), we constructed a setup where they used their existing panel to ask a subset of their panelists (n= 19.426) the question if they were likely to buy a new car within the next 12 months and gave them the following response options;
Definitely (n=1.076), Likely (n= 1.751), No (n=14.497), Don't know (n=2.102)
We then grouped the panelists answering Definitely or Likely and set them as the definition of the in-market audience. Via the survey tool we were now able to extract the share of impressions served to the in-market group per media buy and Line-Item in the ongoing programmatic campaign.
By extracting the adserving & RTB logs and merging them with the panel measured media buys we could ingest cost and viewability data, creating a new way of measuring and optimizing the line-items (see further under execution).
For the in-market audience (answering Definitely or Likely) we had very specific knowledge of their demographic composition, interests and online behaviour (mainly collected as panellist background information), which enabled a very precise twinning construction going beyond observed online behaviour, that could be pushed to buying platforms (see further under execution).
This process was replicated for media buying in Facebook. Using those who answered they were in-market as seed audience for twinning.
We launched the setup in June 2017 and set out to first measure, then optimize, twin the audience and export the twinned in-market audience to the buying platforms.
The merging of the audience data, viewability data and the advserving logs allowed us to create a new currency to evaluate our buying on which was CPM-ii, cost per thousand users in-Market, in-view.
(Pls see Exhibit A for example)
The twinning audience was constructed using panelists who had answered Definitely or Likely to buy a new car only within the last 6 months, and the cut-off was set to an affinity of 400. The twinned audience ID’s was then exported to a DSP (demand side platform).
As a bonus VW’s newly twinned audience segments also proved highly effective with in-market impression shares reaching almost 45% in some cases.
For activation on Facebook the ID’s were transferred into a custom audience and from there the users were twinned and could now be activated on the VW Facebook ad account.
To ensure the validity of the in-market audience it is being updated on a monthly level (to remove panellists who are no longer in-market and to replenish “lost” ID’s).
The below results have not been skewed by doing a lesser more precise media investment as allocated monthly budget has increased since May and now indexing at 116 (may/December).