Automated Customer Segmentation Prediction Pipeline

IYKRA Data Fellowship 12 capstone project for Group 2: Automated Customer Lifetime Value Prediction with Machine Learning.

Driven Output

• Exploration of the importance of CLV in strategic decision-making within banks.
• Highlighting the construction of a scalable data pipeline for continuous CLV model training and prediction.
• Description of the methodology used to collect, preprocess, and analyze customer data for CLV prediction.
• Insights into the machine learning algorithms tested, selection rationale, and performance metrics.
• Case studies or examples of how CLV predictions have informed marketing strategies and product offerings.
• Guidelines for banks to implement and leverage CLV prediction models.
• Design suggestions for the utilization of GenAI for related use case

Background

There’s a Pareto principle that comes from an Italian economist saying that for many outcomes, roughly 80% of consequences comes from 20% of causes. Similarly, 80% of companies business comes from 20% customers. That’s why companies need to identify those top-valued customers and maintain the relationship with them to ensure continues revenue. In order to maintain a long-term relationship with customers, company need to schedule loyalty schemes such as the discount, offers, coupons, bonus point, gifts, etc.

Customer Lifetime Value (CLTV) helps business identify the high-potential customers. By analyzing past customer behaviour and revenue, CLTV can predicts the total value a customer can bring over time. This allow companies to prioritize efforts toward attracting and retaining customers with the greatest future profit potential.

Problem Statement

An e-commerce company has a large number of customers to serve and wants to effectively manage personalized marketing to improve customer loyalty and decrease churning rate.

Goal

Create a data pipeline from existing data that can be utilized to segment the current customers’ value and predict the segment for the next period.

Objective

1. Predict Customer Lifetime Value: Develop a machine learning model that can predict the future value a customer will bring to the company.
2. Automate the Prediction Process: Create an end-to-end automated pipeline that handles data collection, preprocessing, model training, and prediction.
3. Dashboard for Reporting: Build a dashboard to improve data visualization and analysis for better marketing decisions.

Proposed Architecture

Data Pipeline

Machine Learning Pipeline

Data Warehouse

Data Modeling

Data Lineage

Tools

• Data Warehouse: BigQuery
• Compute Engine: GCP VM Instance
• Container: Docker
• Workflow Orchestrator: Astronomer for Airflow
• Data Transformation: data build tool (dbt)
• Data Governance: Soda
• Exploratory Data Analysis: Vertex Colab Notebook
• Data Visualization: Looker Data Studio
• API: FastAPI

Data Visualization

Resource

• Dataset: UCI Machine Learning Repository - Online Retail
• Deck: here
• Whitepaper: here
• API: FastAPI
• Dashboard: Looker Data Studio

Team Member

• Alivia Rahma Sakina (Github, Linkedin)
• Alya Mutiara Firdausyi (Github, Linkedin)
• Fathurrahman Hernanda Khasan (Github, Linkedin)
• Muhammad Zakie Arfiansyah (Github, Linkedin)
• Ridho Alfayet Umar (Github, Linkedin)