During the PoC stage, we developed a prototype and designed the architecture for the functionality that would automate the marketing feature within the client’s e-commerce platform. We created a WBS (Work Breakdown Structure) to effectively organize, plan, and execute all project activities, focusing on template management, email sending, data layer, and statistics. Then, based on the PoC results, we decided to build an AWS-based solution using the cloud-native approach and serverless best practices. This approach reduces maintenance costs and makes the platform flexible for integration with a Machine Learning system. Additionally, our team applied MLOps (Machine Learning Operations) best practices for automating email campaigns to enhance the reproducibility and efficiency of the campaign process. Implementation: setting up a serverless workflow and Machine Learning infrastructure Sparknet team has helped implement a multi-tenant Machine Learning-based solution within the client’s e-commerce platform. It enables accurate subscription churn prediction and helps target subscribers more effectively. We have set up a serverless workflow for email campaigns utilizing AWS Lambda and AWS Step Functions. AWS Step Functions define the sequence of steps needed for email campaigns. We integrated AWS Lambda to generate personalized email content, send emails, track user interactions, and update campaign analytics. As a result of the migration to AWS, customers can configure their email campaigns through APIs. Every customer can reuse the campaigns by setting different configurations, such as lists of receivers, schedules for sending emails, and segments of churn probabilities. Additionally, we have assisted in setting up Machine Learning infrastructure. We began by gathering data about end-user behavior during an email workflow. Then, our AI engineers analyzed this data and calculated churn probabilities, which indicated the likelihood of a customer discontinuing their subscription. We used AWS SageMaker to develop predictive models that determine the likelihood of user actions, such as subscription churn or engagement with email content. Sparknet team enabled the automatic sending of customized emails based on the churn probability data. For instance, in cases of high churn probabilities, specific users receive tailored email campaigns designed to motivate them to continue with their subscriptions. On the other hand, more loyal users receive regular marketing emails. Additionally, our engineers have used churn probabilities to create dashboards with calculated revenue for clients. These dashboards provide a comprehensive view of user interactions with email campaigns. Finally, we have used React to develop the front-end part for the template manager, where customers can move ready-made blocks to build their email campaigns. This feature significantly reduced the time required to compose emails since users no longer have to do it manually.