This is our story of developing a personalized ancillary recommendation system based on Google Analytics 360 Suite, Google BigQuery, and artificial intelligence (AI) platform.
The leading Central and Eastern European low-cost airline, as a value-driven carrier, focuses on innovation at every stage of the customer journey. Its goal is to provide affordable services with great user experience for its customers.
To successfully represent its core values, the must be able to cater to the needs of its customers and provide personalized recommendations for ancillary products, such as seats and priority boarding.
This is how Aliz came into the picture. Having previously worked with the airline, they knew Aliz would be able to help them realize this vision.
The challenge of ancillary recommendation
After some very informative meetings to discuss the client’s needs in detail, we decided to develop a personalized ancillary recommendation system on Google Analytics 360 Suite, Google BigQuery, and AI platform.
The client had been carrying out pre-departure email campaigns but as these campaigns hadn’t met their expectations, they wanted to improve them with personalized messages and recommendations.
They’d also been using Google Analytics 360 Suite (Analytics 360) to record user interactions on their online platforms; however, they hadn’t yet made use of the stored data.
The Problem: How to recommend the right offers like ancillary to the right customers
Based on our client’s business factors, the challenges we faced were many.
- We had to extract useful features from Google Analytics 360.
- We had to find the right offers (ancillaries) to the right customers.
- Then, we needed to identify customer segments to obtain effective target groups for marketing campaigns.
- We also had to design personalized messages by utilizing both the recommendations as well as the customer segment unique preferences was a must.
- Lastly, we had to build a data pipeline and integrate with their marketing platform.
The goals were to:
- Prove that personalized communication would deliver commercial uplift.
- Test Google Cloud for easier scalability/higher availability.
- Prove that Google Cloud could offer a better time-to-market.
We developed a personalized ancillary recommendation system on Google Analytics 360 Suite, Google BigQuery, and AI platform. The project took 3.5 months to complete with A/B testing, with an overall engineering effort of 77MDs.
We implemented a solution that utilizes the data collected in Google Analytics in Google BigQuery and built Machine Learning models on top of it.
Since Google Analytics 360 data already resided in Google BigQuery, the extract, transform, load (ETL) pipeline for obtaining useful features was also designed and implemented in BigQuery.
Because GA 360 data is not structured to have the features directly, it took several transformation steps to extract.
- Context data
- Historical user behavior (purchase and search)
- Historical flight data
As some of these steps are interdependent, we used Cloud Composer to orchestrate and schedule the tasks. Cloud Composer also allows users to easily monitor execution and alerts them in case of errors.
When it comes to user segmentation, it’s always important to come up with feasible target groups for a given business. With the extracted features, we were able to divide the customers into segments that were both targetable and meaningful.
We created and combined two different types of models:
- A recommendation engine consisting of four models for Seat/Priority and One-Way/Return combinations.
- Customer segments based on purchase and search history: Travel Lovers, Family Travelers, First Time Travelers, Business Travelers, and One-Way Travelers.
We then combined these two model types to A/B test the email campaigns and measure the effectiveness of personalization.
The key performance indicators used to determine effectiveness were as follows:
- E-mail open rate
- Ancillary sales
First, we created propensity scoring models for the two ancillary types: seat and priority boarding.
The recommendation engine consisted of four models for Seat/Priority and One-Way/Return combinations. We allowed a maximum of one recommendation for each user. Strong cooperation with our client’s teamthroughout the modeling process allowed us to incorporate their business knowledge into the models for maximum performance.
Then, we created customer segments based on purchase and search history.
We identified five different segments: Travel Lovers, Family Travelers, First Time Travelers, Business Travelers, and One-Way Travelers. Each of these has its own distinct features. For example, Travel Lovers are constantly searching for new trips to take and they travel every second month. They are a small segment but a high revenue contributor. Family travelers plan well ahead, travel with children, and stay for longer periods.
Then, we combined the results of the recommendation engine and the segmentation to deliver the final promotion messages.
We analyzed its effectiveness with A/B testing.
The Results and Business Value
By implementing their first personalized email recommendation pipeline, we helped our client demonstrate the feasibility of the proposed concept. The A/B testing showed a significant uplift in sales coming from the personalized campaigns compared to the control group. The project resulted in a 19.85% ROI.
For us, however, the most important value was the close collaboration between the client and us. Aliz has contributed to launching them on a path to acquiring a data-driven approach to serving their customers.
Interested in knowing more about our solutions?
Check out our offers and don’t hesitate to contact us at email@example.com.