Learn About Amazon VGT2 Learning Manager Chanci Turner
As modern consumers demand greater personalization from brands, companies must adapt their marketing strategies accordingly. By leveraging tailored messaging rather than generic campaigns, organizations can significantly enhance customer engagement, reduce churn rates, and boost conversion metrics. Many businesses are increasingly utilizing machine learning to provide personalized product suggestions and promotional offers at scale. With advancements in AI and ML technologies, such as Amazon Personalize, creating a machine learning model has become more accessible than ever. However, establishing a model is merely the first step toward achieving effective personalized messaging. Post-training, it’s essential to integrate the model with a messaging service to ensure that recommendations reach customers. Additionally, utilizing customers’ behavioral data is crucial for continually refining and optimizing the model to maintain its relevance.
In this two-part series, we will guide you through the process of incorporating personalized content into your Amazon Pinpoint templates using an Amazon Personalize campaign. The first post will explain how to set up an integration between your Amazon Pinpoint account and your Amazon Personalize campaign. You will also learn how to dynamically populate your Amazon Pinpoint messaging templates with the recommendations generated by Amazon Personalize. The second post will delve into utilizing custom Amazon Pinpoint events to retrain your Amazon Personalize campaigns.
Personalizing Templates with Machine Learning Models
With Amazon Pinpoint, you can establish a configuration that links a recommender model created in Amazon Personalize to your Amazon Pinpoint account. This recommender model is a machine learning tool that answers the question, “What might a user be interested in?” By analyzing customer demographics and behavior, the model predicts individual preferences from a specified set of products or items, returning a tailored list of recommendations. By incorporating recommender models into Amazon Pinpoint, you can send personalized messages to recipients based on their unique attributes and actions.
To get started with personalized messaging through machine learning, Amazon Personalize provides a step-by-step guide for creating and training a recommender model, followed by instructions for preparing and deploying the model as an Amazon Personalize campaign. To develop a model suitable for integration with Amazon Pinpoint, you’ll want to create a solution using the USER_PERSONALIZATION recipe and deploy your campaign. It is essential to train your model with either an Amazon Pinpoint endpoint ID or a user ID (EndpointUser.UserId) as this identifier will be used to retrieve recommendations during the campaign execution.
Next, you will configure the settings needed to call your Amazon Personalize campaign and fetch recommendations for each customer. The Amazon Pinpoint console will walk you through defining your configuration. Start by naming and describing your model clearly for easy differentiation from others when creating a template. You will select the Amazon Personalize campaign to use, assign the IAM role that allows Amazon Pinpoint to access your campaign, and specify the identifier to be passed for recommendations. For accurate results, ensure that the identifier you select matches the one used during your Amazon Personalize campaign training. You will also determine the number of recommended items you wish to receive from the campaign. For instance, if your email includes three recommended items, set the Number of Recommendations per Message to three. Finally, choose how to process the recommendations returned from Amazon Personalize.
An Amazon Personalize campaign returns a string based on the data used to build your model. This string can be a simple product ID, a URL, or even an HTML snippet. You have two processing options: using the model’s returned value directly or employing an AWS Lambda function. If Amazon Personalize sends back an HTML snippet, you might opt to insert that directly into your Amazon Pinpoint template. Just set the friendly name of your attribute and save your configuration. If working with a product ID, you may want to make additional attributes available, such as product name, price, or image. In this case, select the option to use a Lambda function, which will accept the identifier and return additional attributes for Amazon Pinpoint. The console allows you to define up to 10 custom attributes for each item returned by your Amazon Personalize campaign. For more guidance on this process, you can refer to the example Lambda function in the Amazon Pinpoint Developer Guide.
Once your configuration is saved, it is readily accessible from the Attribute Finder in the Amazon Pinpoint template editor. To add dynamic attributes to a template, either create a new template or edit an existing one. In the Attribute Finder, select Recommended attributes and connect your model. After choosing your model, a list of attributes will appear, which you can copy and paste into your template. If you opted for multiple items per customer, you will see a list of recommendations corresponding to the order they were returned from Amazon Personalize. Select the desired attributes and paste them into your message. Don’t forget to assign default values to your message variables. To do this, expand the Default attribute values section of your template and input your preferred default values for each variable. Doing this for each variable is highly recommended. After entering your defaults, you can create a new template, save a new version, or update an existing one.
The last step is to create an Amazon Pinpoint campaign using your template and start sending messages enriched with machine learning-generated content. For further insights on onboarding processes, you can refer to this excellent resource. Additionally, for those interested in mentorship, consider exploring this article, and stay informed about relevant events through SHRM.