Introduction to Look-alike-Machine Learning Modelling
In Transactional & Marketing Email
Writing from personal experience, I previously consulted for a European company tasked with implementing a look-alike model for their systems. In this write-up, I aim to provide informative content with a touch of engineering, catering to beginners in the field who are eager to get started.
Join Semis today to network in AI & Bigtech
But first, let’s address a fundamental question:
Do emails have different types for business purposes?
The answer is yes. Businesses often engage with customers through various email types, each serving a particular purpose. These purposes can range from marketing, where businesses showcase their products, services, or offers to prospective or current customers, to transactional emails, which involve exchanges that may or may not include sales. It’s important to note that these email types operate under different modalities, which can have consequences if not properly managed. However, for the scope of this discussion, let’s focus on exploring machine learning and recommender systems in marketing.
In more detail, let’s take a look at the definitions of marketing and transactional emails based on my notes from last year. While I can’t recall the exact sources, the following quotes provide insightful explanations:
Sender: Marketing emails are often sent from a company or brand, whereas transactional emails are sent from a specific person or entity.
Recipient: Transactional emails are usually sent to an individual in response to a specific action they took, while marketing emails are sent to a larger audience.
Content: Marketing emails typically contain promotional content, while transactional emails contain information related to a specific transaction or action.
Subject: Marketing emails may have more attention-grabbing or sensationalist subject lines, while transactional emails tend to have more straightforward subject lines.
Frequency: Marketing emails are often sent on a regular basis, while transactional emails are typically sent on an as-needed basis.
Time: Transactional emails are often sent immediately after an action is taken, while marketing emails may be scheduled to send at a specific time.
Let’s say you sign up for a new web/mobile app, you get a welcome email, etc, those are transactional, the emails you receive compelling you or requiring you to take an action to purchase are marketing.
Become a mentor or mentee in big tech & AI, join Semis today!
Look Alike Machine Learning Models
So, where does machine learning come into the picture? Imagine a scenario where a business collects vast amounts of customer email data, encompassing both transactional and marketing interactions. But what if the goal is to go beyond simply sending emails and delve deeper into understanding customers — their preferences, spending habits, geographic locations with the highest spending patterns, and more?
Join Semis today to network in AI & Bigtech
Here’s where the concept of Look-alike Machine Learning Modelling (LAM) comes into play. LAM shares similarities with recommender systems in the space of ML/AI and offers a powerful solution for businesses seeking to enhance their understanding of their user base and improve customer engagement.
Consider this: by leveraging LAM, businesses can identify and match users with similar interests, tastes, and styles. This not only streamlines marketing efforts but also opens up new avenues for revenue generation, enhances customer satisfaction, and improves retention rates.
While there are various methods for achieving this goal, I’ll be focusing on LAM in this write-up due to its effectiveness and relevance to the context of email marketing and customer engagement.
In the following sections, we’ll go a bit further into LAM, exploring its applications, benefits, and practical implementation strategies for businesses looking to unlock the full potential of their customer data.
Become a mentor or mentee in big tech & AI, join Semis today!
Companies currently using LAM
Facebook (now Meta) stands out as a leader in Look-alike Machine Learning Modeling (LAM), particularly evident to those who have used their advertising tools on platforms like Facebook and Instagram. When creating ads on those platforms, the option to target look-alike customers is a common feature.
Join Semis today to network in AI & Bigtech
Besides Meta, other tech giants like Netflix, Amazon, and LinkedIn boast exceptional engineering teams dedicated to developing and refining recommender systems.
Based on this article, it is reported that LinkedIn has discontinued the look-alike audiences feature on its platform. Similar techniques being applied include; predictive audience, and audience expansion.
Approaches to LAM
In Look-alike Machine Learning Modeling (LAM), the approach involves analysing a new dataset containing features similar to an existing dataset, aiming to identify and capitalise on meaningful similarities.
For instance, consider Company Xy, which seeks to develop an LAM model to forecast the number of customers likely to purchase a new product. In this scenario, the customers targeted must exhibit a high likelihood of opening marketing emails sent by Company Xy. By leveraging LAM, Company Xy can identify and target potential customers who closely resemble their existing customer base, thereby maximising the effectiveness of their marketing efforts.
Requirements (engineering)
1. Seed data
2. Pool data
3. Data Pre-preprocessing
4. Feature engineering and selection
5. Ranking the customers in the pool and seed data (scoring and evaluating each customer based on their features/interests)
6. Selecting your preferred algorithm
7. Selecting your preferred algorithm Developing the model (choosing an algorithm)
Join Semis today to network in AI & Bigtech
Deploying your Model
See my post on machine learning deployments.
Support my work
This article has a prequel that is data pre-processing,
Best regards,
Tolulade
