gPS Phoebe

Real-time AI-based value based bidding with Google Ads products, that keeps sensitive data away from determined end users.

Use Server Side Tag Manager with Vertex AI, to generate real time predictions based on ML models, and use them with Google Ads products and other services to improve your targeting and bidding processes.

Description

With server-side Tag Manager we are able to bid to a secret conversion value by leveraging the built-in Firestore lookup (see gPS Soteria). One of the limitations with this approach is that the conversion value needs to sit in Firestore before the conversion happens. This prevents more advanced bidding use-cases that use AI to determine this value.

gPS Phoebe allows users to leverage server-side Tag Manager to get predictions directly from a Vertex AI model in real time, and use the predicted values as secret conversion values to be used in value based bidding.

The image below provides a high level overview of the core components involved, and outlines the flow.

1. A GTM web container is used to set up tagging on the site, and is configured with a “purchase” event.
2. The client’s website is set up to have an e-commerce purchase event in the data layer: this contains the revenue data. When a purchase is made, the event fires, sending the payload to a GTM server container.
3. A custom variable is attached to a tag, triggered by “purchase” events, which queries the Vertex AI to get a prediction based on the provided data, and replaces the revenue conversion value with the predicted value. This request passes through a proxy app to enable unauthenticated calls to the Vertex AI endpoint.
4. The updated event (with the predicted conversion value) is sent to Google Analytics, Google Ads or Floodlight.

Demo Implementation

If you are interested in deploying an entire demo environment to test the solution, along with a demo web store and sample tags in GTM, you can follow the guide here. Otherwise, skip to the next section for the implementation guide.

Implementation Guide

Requirements

• A Google Cloud project with billing enabled
• A GTM server container up and running, and receiving events from a GTM web container
• A Vertex AI endpoint with the model you want to us
• It is recommended to have all the components in the same cloud region to reduce network latency

Prepare the environment

Clone the repository to a local folder in your current environment (we recommend using cloud shell for the deployment as it provides all required utilities out of the box), using this command:

git clone https://github.com/google/gps-phoebe

If you are not using cloud shell, you must install the gcloud CLI in your environment.

Deploy the proxy application

You need to use a proxy application to call a Vertex AI endpoint, as currently server-side Tag Manager does not support outbound authenticated requests.

Before proceeding, you need to select the project the Google Cloud project with the following command (substitute PROJECT_ID with the correct project id):

gcloud config set project PROJECT_ID

Execute the following command to deploy the proxy application to Cloud Run:

cd proxy_app/cloud_run
gcloud run deploy phoebe-proxy-app --source proxy_app/cloud_run

Configure Tag Manager

Import the template

Download the template file here.

From the Google Tag Manager UI, select your server container, go to the Templates section, and click on New:

Then, click on the top right options button, and select Import:

Select the file you downloaded and continue. Then, click on Save and close the template using the top left close button.

Create the variable

Now, from the Google Tag Manager UI, go to the Variables section, and click on New:

Click on Variable Configurationto select a variable type:

And select Vertex Prediction:

Then, fill in the details of your environment in the configuration screen:

Remember to append the Proxy App URL with the /predict suffix. Example: https://my-proxy-ap.ew.r.appspot.com/predict. You can find the URL to use in your Cloud Run console.

Apart from the required fields, you need to provide the request data that will be sent to the Vertex AI endpoint through the Proxy App. This data would depend on what your AI Model requires. This is an example for our demo model:

Fields will be included in the requests with string type. If your model requires a parameter to be integer or number type, you can add a suffix to indicate it to the system, _intfor integer and num for number.

Remember to give it a name, and click Save when finished.

Finally, to add the predicted value coming from Vertex to the purchase or conversion event, open the appropriate Tag and replace the conversion value (or equivalent) field with the variable you just created.

For example, in Google Ads Conversion Tracking, you can use the Conversion Value field:

Disclaimers

This is not an officially supported Google product.

Copyright 2023 Google LLC. This solution, including any related sample code or data, is made available on an “as is,” “as available,” and “with all faults” basis, solely for illustrative purposes, and without warranty or representation of any kind. This solution is experimental, unsupported and provided solely for your convenience. Your use of it is subject to your agreements with Google, as applicable, and may constitute a beta feature as defined under those agreements. To the extent that you make any data available to Google in connection with your use of the solution, you represent and warrant that you have all necessary and appropriate rights, consents and permissions to permit Google to use and process that data. By using any portion of this solution, you acknowledge, assume and accept all risks, known and unknown, associated with its usage, including with respect to your deployment of any portion of this solution in your systems, or usage in connection with your business, if at all.