2022-11-14 | Article | Insights
This insights article describes how the Google Cloud AI platform Vertex AI can be used to automate the process needed in conducting marketing mix models and to extract insights for a data-driven optimised budget allocation.
Marketing managers know that MMMs will become more important in the future, but it is difficult to set up and train those models individually. It is well known that models simplify reality. However, it is always necessary to weigh things up: The simpler the model, the easier it is to understand, but at the same time we are more distant from the description of reality. Vertex AI allows to close this gap somewhat. The platform provides tools and automated machine learning algorithms to prepare data, train models and calculate predictions in an automated way. The results have a high level of sophistication and therefore come a bit closer to reality.
Vertex AI is the Google Cloud AI platform that enables to integrate all relevant steps of an MMM and generate AutoML or custom model results. The platform offers four core components, all of which are used within the framework of a marketing mix modeling:
(1) Tools
(2) Data
(3) Model development
(4) Deployment
Tools: Within the workbench tool user-managed notebooks are created to conduct the needed exploratory data analysis as well as data cleaning and preparation for the following training of the marketing mix model.
Data: The cleaned and prepared data to be used for training is created and provided in the datasets section. Here all datasets can be managed in a central location. As a data source CSV files can be uploaded from the computer or Cloud Storage as well as selected BigQuery tables or views can be used. Already at this point it is defined that tabular data are to be used with the aim of solving a regression problem.
Model development: To train the marketing mix model, training needs to be defined in detail in the model development section. Here, training pipelines can be used to create AutoML-trained as well as custom-trained models. One of the main outcome of the trained model is the feature importance per input variable, which indicates the role of each marketing channel with regard to the value contribution towards conversions. The Vertex AI out-of-the-box solution also provides metrics regarding evaluation and explainability and on top provides a direct possibility to deploy and use the model.
Deployment: For the marketing mix modeling it is sufficient to select batch prediction. When we provide the historical data, we receive as a result the estimated conversion values based on the model and can also calculate, for example, the isolated channel effects and the ROAS per marketing channel. In addition, batch predictions can be used to predict conversions for different future budget scenarios.
The MMM setup in Vertex AI can be customised. If desired, Vertex AI enables regular model updates through automation of the processes on a monthly or quarterly basis and thus always up-to-date conversion predictions and possible evaluations of budget scenarios. The outcome is also very flexible in the way it is presented. The model results can be easily viewed directly in the Google Cloud Platform interface, or individual Data Studio dashboards can be created.
Either way, with the marketing mix model results provided, it should be easy for relevant stakeholders to measure and evaluate the effect of their marketing activities. Based on this, the model updates simultaneously enable control and effect measurement of the adjusted budget distribution.