Berge & Meer Uses Conversion Probability Predictions to Increase Marketing Efficiency in Search Ads 360 Campaigns
2023-01-09 | Case study | Insights
Digitl was able to help us identify opportunities to focus our budget on valuable campaigns and users, and the use of their expertise across the GMP tech stack was key to our success. The results speak for themselves, with our branded search campaigns returning a ROAS well above our target.
Johannes Grandmontagne | Head of Online Marketing
Understanding conversion behavior is essential to increasing marketing efficiency
Understanding conversion behavior is essential to increasing marketing efficiency
Located in Rengsdorf, Berge & Meer Touristik pioneered the travel industry by offering exceptional all-inclusive travel experiences around the world. The company operates primarily in Germany and Austria, to a lesser extent. It sells almost exclusively directly to consumers. After the Coronavirus outbreaks in the last few years, the business shifted to a digital-first strategy to ensure business continuity and sought ways to focus its budget on valuable campaigns and users. After Digitl’s in-depth assessment, a clear opportunity was identified that could use onsite conversion behavior data to define, create and target valuable audience lists aligned with Berge & Meer's marketing strategies.
Return on ad spending in search branded campaigns
Using Google Vertex AI to train predictive models
Using Google Vertex AI to train predictive models
Users' previous activity and behavior data can be used to develop individual conversion probability predictions (CPP). Using exploratory data analysis (EDA), the team extracted Google Analytics raw data via BigQuery and exported it to a Vertex AI workbench. A pre-selection of input features was done in order to avoid overloading the model and by correlation analysis, the most relevant features were identified. For evaluation purposes of the model performance, Digitl applied unseen test data to the model and measured the outcome with a confusion matrix. As the model demonstrated very positive performance, it was deployed. Digitl used the Vertex AI pipeline to manage and monitor the whole deployment workflow.
Activation of CPP audiences
Activation of CPP audiences
Digitl automated the analysis & data import of these predictions, from BigQuery to Google Analytics 360. The import enhanced user data in Google Analytics by storing predictions in a custom dimension. As a result, specific prediction scores were formed into audiences that reflect Berge & Meers's user journey stages: Awareness, Interest, Consideration, and Intent. These audiences were synced with Google Ads and monitored in the new Search Ads 360 audience manager. The Search Ads 360 (SA360) campaigns used a Revenue bid strategy focused on optimizing ROAS using custom floodlight columns and a Data-Driven Attribution model.
The results and business impact
The results and business impact
The automated conversion probability approach provided Berge & Meer with an always-adopting data-driven solution to audience building. In addition, the user funnel stage categorization helped align it with marketing strategies and made it easily applicable. These audiences had an outstanding impact on search branded campaigns, returning a return on ad spending (ROAS) above 3500% in various scenarios, well above the target ROAS target of 1000%. To date, the CPP audiences have been applied to several search campaigns, contributing to overall revenue performance and continuing to target valuable users.
Devinson Peña
Team Lead Ad Technology
Isabelle Hanisch
Senior Consultant Digital Analytics
The Goals
- Harness first-party data to predict customer behavior.
- Reach customers that are likely to book a trip.
- Shift budgets towards valuable campaigns and users.
- Optimize branded search campaigns with SA360.
- Automate the audience-building process.
The Approach
- Used Analytics 360 data to create a conversion probability prediction model in BigQuery.
- Connected SA360 with Google Analytics 360 to engage custom segments of likely-to-convert visitors.
- Compared branded vs. non-branded campaigns to set the future activation strategy.
The Results
- For all branded search campaigns, a ROAS above 3500% was achieved using the CPP audiences created by Digitl.