Predictive marketing analytics using BigQuery ML machine learning templates
Enterprises are collecting and generating more data than ever—to better understand their business landscape, their market, and their customers. As a result, data scientists and analysts increasingly need to build robust machine learning models that can forecast business trajectories and help leaders plan for the future. However, current machine learning tools make it difficult to quickly and easily create ML models, delaying time to insights.
To address these challenges, we announced BigQuery ML, a capability inside BigQuery that allows data scientists and analysts to build and operationalize machine learning models in minutes on massive structured or semi-structured datasets. BigQuery ML democratizes predictive analytics so that users unfamiliar with programming languages like Python and Java can build machine learning models with basic SQL, and is generally available.
To make it even easier for anyone to get started with BigQuery ML, we have open-sourced a repository of SQL templates for common machine learning use cases. The first of these, tailored specifically for marketing, were built in collaboration with SpringML, a premier Google Cloud Platform partner that helps customers successfully deploy BigQuery and BigQuery ML. Each template is tutorial-like in nature, and includes a sample dataset for Google Analytics 360 and CRM along with SQL code for the following steps of machine learning modeling: data aggregation and transformation (for feature and label creation), machine learning model creation, and surfacing predictions from the model on a dashboard. Here’s more on the three templates:
Customer segmentation—By dividing a customer base into groups of individuals that are similar in specific ways, marketers can custom-tailor their content and media to unique audiences. With this template, users can implement a BigQuery ML k-means clustering model to build customer segmentations.
Customer Lifetime Value (LTV) prediction—Many organizations need to identify and prioritize customer segments that are most valuable to the company. To do this, LTV can be an important metric that measures the total revenue reasonably expected from a customer. This template implements a BigQuery ML multiclass logistic regression model to predict the LTV of a customer to be high, medium, or low.
Conversion or purchase prediction—There are many marketing use cases that can benefit from predicting the likelihood of a user converting, or making a purchase, for example ads retargeting, where the advertiser can bid higher for website visitors that have a higher purchase intent, or email campaigns, where emails are sent to a subset of customers based on their likelihood to click on content or purchase. This template implements a BigQuery ML binary logistic regression model to build conversion or purchase predictions.
To start using these open-source SQL templates and more, visit our repository—the code is licensed under Apache v2. We will also be adding templates for more use cases in the future. And to learn more about applying BigQuery ML for marketing analytics, watch this Google Cloud OnAir webinar.