ad
ad

AI-Driven Personalized Email Marketing

Science & Technology

Introduction

Hello everyone, I'm Xi Yang from Endeavor Drinks, and I'm very excited to bring to you our recent work on AI-driven personalized email marketing. We started this project last November, leveraging Databricks to empower all aspects of our personalized marketing design and implementation.

Introduction to Endeavor Group Limited

Endeavor Drinks, as we call ourselves within the company, is Australia's largest drink retailer with over 2,000 venues across various brands, including popular names like Dan Murphy's and BWS. Our data science team, in collaboration with a cross-functional team, has been focused on building a personalized marketing engine. This engine performs segmentation to target specific groups with customized messages. With around 4.5 million customers, this initiative is vital for our strategy, aiming to significantly boost revenue and enhance customer experience.

Problems Solved by Personalized Marketing

In personalized marketing, we tackle four critical problems:

  1. Theme of the Email: Determining whether the email is recommending a new product, an existing product, a new catalog, or a promotion.
  2. Product Selection: Deciding which products should be featured in the email.
  3. Target Audience: Identifying which customers should receive the email.
  4. Timing: Figuring out the best time to send the email based on customer preferences.

Addressing Each Problem

  1. Theme of the Email:

    • We use a multi-arm bandit algorithm to determine the email theme.
    • The algorithm considers two factors: the likelihood of a customer clicking a product and the likelihood of a purchase following the click.

  2. Product Selection:

    • We utilize an XGBoost model to rank the likelihood of customers purchasing different products based on past behavior, demographics, and engagement.

  3. Target Audience:

    • This involves advanced customer segmentation and targeting, ensuring the email is sent to the right individuals.

  4. Timing:

    • Currently, we send emails randomly between 3 to 5 time slots but plan to optimize this as the next step in our project.

Solution Components

Our personalization engine consists of three major components:

  1. Action Incubator:

    • Generates email content, including message copy and images, using Airtable.

  2. Personalization Engine:

    • The brain of the operation, determining time, products, and which emails to send based on sophisticated business logic.

  3. One-to-One Channel Allocation:

    • Ensures the right content is sent through proper channels, like email.

Multi-Arm Bandit Problem

  • The core challenge is to balance exploration (trying out new email options) and exploitation (using known effective emails).
  • We use the Thomson algorithm to strike this balance effectively.

Addressing New Challenges

  • Cold Start vs. Warm Start: For new users, we opted for a cold start approach, sending random emails to gather more information over time.
  • Exploration and Experimentation: Continuously test and improve the relevance and effectiveness of different email templates.
  • Business Overrides: Implement business logic for seasonal or important announcements that need to reach specific customer demographics.

Business Problems & Customer Experience

  • Email Frequency: Testing different email frequencies to avoid annoyance and unsubscription.
  • Anti-Repetition Rule: Preventing the same or similar emails from being sent repeatedly.
  • Customer Journey: Ensuring a pleasant and consistent customer journey across different stages and channels.
  • Cross-Channel Consistency: Maintaining consistent messages across various marketing channels.
  • Introducing New Emails: Balancing the introduction of new email templates while maintaining customer engagement.

Future Plans

We have several exciting plans in the pipeline, such as:

  • Personalized landing pages, search results, customer reviews, and offers.
  • Basket building to pre-populate products in customer carts.
  • Enhancing customer retention and journey mapping.
  • Implementing app push notifications for timely offers.

Keywords

  • AI-driven personalized marketing
  • Databricks
  • Customer segmentation
  • Multi-arm bandit
  • XGBoost model
  • Target audience
  • Email frequency
  • Anti-repetition rule
  • Customer journey
  • Cross-channel consistency

FAQ

Q: What is Endeavor Drinks? A: Endeavor Drinks is Australia's largest drink retailer, with over 2,000 venues and popular brands like Dan Murphy's and BWS.

Q: What major problems does personalized marketing solve? A: It solves problems related to email themes, product selection, target audience identification, and timing.

Q: What algorithms are used in your personalization engine? A: We use multi-arm bandit algorithms and XGBoost models for decision-making.

Q: How do you balance exploration and exploitation in email campaigns? A: We use the Thomson algorithm to find the optimal balance between trying new email options and using known effective emails.

Q: What are some upcoming features in your AI-driven marketing strategy? A: Future plans include personalized landing pages, basket building, app push notifications, and enhanced customer journey mapping.

Feel free to ask any more questions!