With this self-service approach and customer model in place, not only is our tech team freed to focus on higher value items like building a robust system architecture for optimizing ads in real-time, but marketing analytics also has the opportunity to go deeper in areas that are key to understanding our business. Since Tableau lets our business users ask and answer many of their own questions, or make their own data manipulations, our team is relieved of the daily back and forth of report updates or simple data change requests. Now, we can use the zulily data platform for developing advanced machine learning models that help the business drive customer acquisition and better understand the customer experience.
With all of zulily’s data now in BigQuery, we are able to build a robust machine learning model to predict customer lifetime value (LTV) using a variety of customer behaviors as inputs and tie the results to specific marketing campaigns to measure long-term performance.
To accomplish this, we assign a historical lifetime value to find existing high value customers. We then use gradient boosting to consider over 1,000 transactional and behavioral variables and test hundreds of models. We eventually reduce to around 30 key features that are contributing factors for predicting the lifetime value of a customer.
The end result is a model that can predict with very high accuracy the likelihood that a new zulily shopper will have a high lifetime value. We are then able to tie these predictions to marketing datasets living in BigQuery and provide the combined marketing performance data in highly dynamic and customizable dashboards, internally dubbed the Channel Metrics Dashboard (CMD), on our Tableau server. The CMD allows marketing channel manager and specialists to easily generate their own insights using historical and predicted data by selecting the metrics, date ranges, and even granularity they need by creating custom graphs and reports right on the server. This solution enables the zulily marketing team to quickly make decisions that optimize ads, emails, and offers towards customers who best respond to zulily engagements.
When we moved over to this new data platform, we had a lot of help and support from both our zulily tech team as well as Google Cloud engineers and Tableau customer engineers.
In part two of this series, this combined team will share tips and tricks about integrating BiqQuery with Tableau—stay tuned!