Audi, PHD and Semasio Prove Privacy-First Semantic Context Conversion Efficiency

Semasio, which is also deployed in the U.S. and many other countries, uses its unique form of AI content scanning to deconstruct the semantics of each webpage, and of privacy-first IDs. Two tests were conducted by PHD for Audi, one in which personae of known converters were included (specifically high income, aged 50+ and with an affinity for business content) and other demographics excluded (specifically low income, under 24 years old, with no affinity for business content). This was done both on a privacy-first ID basis as well as on a website contextual basis (based on a webpage's dominant visitor profile). We call this the demo test.

The other test used semantic lookalikes of known converters -- people who went to the same types of sites/pages in terms of semantic content -- and again did this both on an ID and a site/page basis.

Results of the top-down test:

1. 70% of the conversions came from the test cell, which used both IDs and contexts to exclude <24 low income. 30% came from the >50 high income cell. This suggests that it is better for a premium car manufacturer to exclude the highly unlikely demos than it is to target the demos most likely to buy. It makes one think this principle might also apply more widely. My own work with Next Century Media addressable commercials in the '90s was focused more on exclusion and so I am particularly interested in this finding and its possible generalizability.
2. Looking at the universe of non-available IDs where page-level Semasio demographics were used, the exclusion principle again showed itself as stronger than the inclusion approach: 48% of total demo test conversions came from the page-based exclusion cell and 17% from the inclusion cell -- all based on the same demos as in finding No. 1 above.
3. The page-level strategy accounted for 65% of the conversions vs. 35% for the ID-based strategy.

In other words, context beat IDs! The likely explanation for this is threefold:

1. By taking a different approach to other premium car manufacturers in this market characterized by a low level of user-level targetability (35% as opposed to around 65% in the U.S.), Audi was able to circumvent the micro-economic dynamics of contracting supply with constant demand. This enabled them to buy the right users and impressions more cost-effectively, thus driving up ROAS.
2. Persona-based targeting has an element of subjectivity to it through the assumption of the personae as the "best audiences."
3. Probably an even larger factor is the degree of resonance between any car ad and the environments that are frequented by known Audi converters, which adds to the impact of an ad and is known as context effect (aka priming effect, aka alignment effect). Probably this effect would be heightened even more if Semasio's semantic system were used (as RMT uses it) to detect semantic resonance between the specific Audi ad and the page/site.

In the second test in Denmark, semantic lookalikes for known Audi converters were used, on both an ID and page basis. I call this the bottom-up test.

Results of the bottom-up test:

1. On an ID basis, the semantic lookalikes of the converters reduced cost per conversion -59% versus the baseline.
2. Projecting the converters onto the contexts in which they were greatly overrepresented, produced a CPA reduction of 81%.

Again, context beat IDs.Here the explanations from the first campaign appear to be applicable again: Doing something different to everybody else allowed Audi to discover "pockets of value" in the form of net new prospective buyers and net new contexts without any semantic connection to premium cars but rather selected for their overrepresentation of the Audi converters.

Our conclusions from this landmark study:

1. These semantic and contextual approaches will be what replaces cookie-based targeting in the privacy-first era we are all headed for despite Google's stay of deprecation for the 3^rd-party cookie. Why would people use it here? (a) Because it works well (b) Because it mitigates non-judicial privacy risk (bad buzz social media etc.).
2. The one additional layer we recommend adding to this formula is the Resonance of the ad with the user and with the site/page. Semasio does this already using RMT Motivations to match IDs and sites/pages to specific ads -- such as a specific Audi ad (alas these are not yet available in Danish and thus could not be leveraged for this study). This has been proven to increase ROI and branding effects, and would allow campaigns practicing exclusion, converter semantic lookalikes and contextual targeting to also leverage the psychological resonance of individual ads -- reaching people who have the Motivations the ad subconsciously communicates -- through contexts that echo those motivations, creating holistic harmony. This would also enable the campaign to reach larger audiences who may not qualify as converter lookalikes but who mirror the motivational content of the specific campaign ad(s).

The Semasio semantic twins method is a superior means vs. demographic lookalikes of detecting audiences and contexts that reach the real purchaser target. RMT/Semasio Motivational Resonance between audiences/contexts and the specific creative is not about targeting at all, but about maximizing the impact of the specific creative used. The two go together like Astaire and Rogers.

Click the social buttons to share this story with colleagues and friends.
The opinions expressed here are the author's views and do not necessarily represent the views of MediaVillage.com/MyersBizNet.