Advertisers are facing a paradox. Increasingly, consumers expect advertising to reflect the seamless personalization and relevance exhibited by the digital products and experiences that have come to play major roles in our lives. At the same time, users are increasingly concerned about how their personal data is being used.
Businesses globally are realizing this: The New York Times recently stopped selling behavioral segments as part of their ad proposition outside the US and saw revenue increase. Facebook can no longer share between Instagram and Whatsapp and is banned from collecting 3rd party user data. With GDPR in full force, historically large audience segments are decreasing in size.
With relevance crucial to engaging consumers online, but privacy more important to consumers than ever before, Google and Essence decided it was time to reject the tradeoff between personalization and privacy and set a new standard in digital advertising.
Our challenge was to create a new approach to advertising capable of delivering ads that are more contextually relevant, meaningful, and helpful to consumers while using no personal data and still meeting and exceeding our high standards for performance.
Our answer to this challenge was Project Pegasus.
We theorized that - using the best of Google Cloud's AI suite - we could create a way to produce personalized advertising that didn’t rely on harvesting personal data or employing questionable models inferred from browsing behaviors. Instead, our new approach would leverage the Google Marketing and Cloud Platforms in imaginative ways to tailor ads for readers based on the context of the article they were reading.
To put our theory to the test, we built a prototype we named ‘Pegasus’ which identified a new way of reading publisher data more deeply than before and could automate the production of creative tailored to individual articles on the Guardian's website. We then built out a Vision Machine Learning model (with Google Cloud's Natural Language API), to read article context using both text and images.
We used this vision machine learning model to holistically evaluate the context of almost 3,000 articles on The Guardian’s website and automated the production of thousands of ads, each customized for contextual relevance on individual article pages. Finally, Pegasus employed machine learning to ensure all of these ads met our rigorous standards for brand safety.
With initial tests proving fruitful, we began our first Google Home campaign with Pegasus in March, serving more than 300,000 dynamic impressions contextually harmonized with articles to demonstrate how Google’s smart speakers can be used in environments and use cases directly related to content on TheGuardian.com. For example, against a travel article about Madrid, the ad would show how Google Home can be used to learn Spanish, search for restaurants in the city, or a local weather report.