Note: This is not an officially supported Google product. It is a reference implementation.

MoDeM

MoDeM (Model Deployment for Marketing) is a Google Cloud based ETL pipeline for advertisers, interested in ML-based audience retargeting. The pipeline extracts user data from BigQuery, runs it through the desired machine learning model (BigQueryML, scikit-learn, XGBoost, Tensorflow, AutoML), transforms the model predictions into an audience list, loading it into Google Analytics, for eventual activation in Google Ads, Display & Video 360 and Search Ads 360.

With marketers using increasingly sophisticated approaches in digital advertising, there has been an exponential increase in number of analysts, data scientists, and statisticians within marketing departments. While their mathematical modelling skills are second to none, the long-term success of ML projects hinge on making the jump from analysis to action. Often, analyst teams hack together a process, that can be extremely manual and error-prone with too many parameters, decoupled workflow dependencies and security vulnerabilities. In fact, an entire discipline called MLOps has emerged that focuses on operationaling machine learning workflows.

MoDeM hopes to provide the last-mile engineering infrastructure that enables analysts to quickly productionize & activate their models with the necessary operational and security rigor.

Prerequisites

1. Match Key for Data Import is a custom dimension in GA, eg: Client Id. If using Data Import - any model that is created using client id or user-id and the desire is to create remarketing audiences based off the output of the query you need to ensure that id is captured as a custom dimension in GA. The pipeline will point the clientId or user-id field from the query to the custom dimension in GA.

Instructions

1. BigQueryML models using Cloud Functions & Cloud Scheduler / Compute Engine - here
2. Python ML models with Google AI Platform (scikit-learn, XGBoost & Tensorflow) using Compute Engine - here
3. AutoML models using Compute Engine and/or AI Pipelines^*.

---

^*On the roadmap.