Olist E-Commerce — Customer Segmentation

This project segments 96,475 Olist Brazilian e-commerce customers into 10 distinct behavioural groups using RFM Analysis, a proven customer analytics framework used by marketing and commercial teams worldwide.

The analysis answers one central business question:

"Who are our customers, what are they worth, and what should we do about each group?"

This project is the second part of a two-project Olist analysis series:

• Project 1: Olist Funnel Performance Analysis — SQL & Tableau — What happened in the marketplace?
• Project 2 (this project): Customer Segmentation — Python — Who are the customers and which ones matter most?

RFM stands for Recency, Frequency, and Monetary value, three dimensions that together describe a customer's relationship with a business:

Each customer receives a score of 1–4 on each dimension. The combined score (e.g. 441, 112) determines which of 10 behavioural segments they belong to.

R$3,076,197, 19.9% of all platform revenue, belongs to customers who have left.

These 11,625 customers spent an average of R$264.62 per order, the highest average order value of any segment, yet they have not returned. They are not low-value customers who drifted away. They are the platform's best customers, now shopping elsewhere.

This is the most urgent commercial finding in the analysis. A win-back campaign targeting even 20% of this segment would recover over R$600,000 in revenue.

The combined Needs Attention, At Risk — High Value, and At Risk — Low Value segments account for:

• 24,188 customers (25.1% of the customer base)
• R$3,853,728 in combined revenue (24.9% of total)

These customers are still in the data, they have not fully lapsed, but their recency scores indicate declining engagement. This is the retention window. Once they cross into Lost territory, re-engagement becomes significantly more expensive.

Champions (12.3% of customers) and High Value New (6.5%) together represent fewer than 1 in 5 customers, yet generate R$3,591,399, 23.2% of total revenue. These segments punch well above their weight and warrant priority treatment in retention strategy.

The average order value chart makes the premium clear:

Nearly all customers across every segment have a Frequency score of 1. This is a structural characteristic of Brazilian e-commerce, the marketplace model attracts one-time purchasers rather than repeat buyers.

This single finding defines the platform's biggest growth opportunity. Repeat purchase rate is the lever. Converting even a fraction of one-time buyers into two-time buyers would materially shift revenue without any new customer acquisition cost.

The RFM scatter plot maps every customer by recency and total spend simultaneously, revealing the full customer lifecycle in one view:

• Left cluster (green/blue), recent customers, still engaged, varying spend levels
• Centre band (orange), customers drifting toward inactivity
• Right mass (dark red), Lost High Value customers, 400–600 days inactive
• Far right cluster, earliest 2016 customers, completely lapsed

The visual separation between active and lapsed segments confirms that churn on this platform is not gradual, customers tend to either stay engaged or disappear entirely.

import pandas as pd          # Data manipulation and aggregation
import matplotlib.pyplot as plt  # Chart creation
import seaborn as sns        # Statistical visualisation
from datetime import datetime    # Date handling

# Recency: lower days = better = higher score
rfm['R_Score'] = pd.qcut(rfm['Recency'], q=4, labels=[4, 3, 2, 1])

# Frequency: higher count = better = higher score
rfm['F_Score'] = pd.qcut(rfm['Frequency'].rank(method='first'), 
                          q=4, labels=[1, 2, 3, 4])

# Monetary: higher spend = better = higher score
rfm['M_Score'] = pd.qcut(rfm['Monetary'], q=4, labels=[1, 2, 3, 4])

# Combined score
rfm['RFM_Score'] = (rfm['R_Score'].astype(str) +
                    rfm['F_Score'].astype(str) +
                    rfm['M_Score'].astype(str))

Customers were assigned to segments based on their R, F, and M score combinations:

I am a data analyst with a Master's degree in Accounting and a Google Data Analytics Certificate (Coursera), actively transitioning into a full-time data analytics role.

This project is part of a two-project Olist series demonstrating end-to-end analytical capability — SQL cleaning and funnel analysis in Project 1, and Python-based customer segmentation in Project 2.