RFM Analysis for Customer Segmentation

I wanted to refresh my data science skills with something practical, not just another tutorial notebook. So I built an end-to-end customer segmentation project using a real e-commerce behavioral dataset (millions of events from 2020–2021). What I did step by step: • Filtered raw clickstream data down to purchase events • Engineered RFM features (Recency, Frequency, Monetary) per customer • Scaled features and used K-Means clustering • Chose the number of clusters using elbow + silhouette analysis instead of just guessing • Profiled the clusters into high-value, at-risk, new/occasional, and low-value segments • Built a Tableau dashboard on top so a non-technical stakeholder can actually use it Why I like this project: • It starts from messy behavioral data, not a tiny toy dataset • It focuses on customer value and retention, not just “does the model run” • It ends with a dashboard that could realistically be handed to marketing / CRM teams GitHub repo (code + notebook + README): https://lnkd.in/gTQN7PSt Tableau dashboard: https://lnkd.in/geFsSv6P Next step: I want to extend this into churn / retention modelling and keep building a small portfolio of projects like this. If you work in data science, analytics, or marketing analytics and have ideas for what you’d add or change here, I’d love feedback. #datascience #analytics #machinelearning #tableau #customersuccess #segmentation

Account scoring can be notoriously difficult to build. RFM scoring is one of the most useful frameworks in RevOps and in many motions it can outperform ML models. But... it completely breaks down in enterprise selling Traditional RFM measures Recency, Frequency, and Monetary value of purchases. Works great in transactional B2B where customers buy often In enterprise? Customers purchase once every few years. Frequency is meaningless. Recency is a lagging indicator. By the time those metrics drop, you've already lost the renewal window Here's how to adapt it 𝗥 = 𝗥𝗲𝗰𝗲𝗻𝗰𝘆 𝗼𝗳 𝗠𝗲𝗮𝗻𝗶𝗻𝗴𝗳𝘂𝗹 𝗘𝗻𝗴𝗮𝗴𝗲𝗺𝗲𝗻𝘁 Stop measuring last purchase date. Measure the last time a qualified stakeholder took a high-intent action Your VP of Finance logging into the platform last week matters. An intern opening a marketing email does not reset the recency clock 𝗙 = 𝗙𝗿𝗲𝗾𝘂𝗲𝗻𝗰𝘆 𝗼𝗳 𝗠𝘂𝗹𝘁𝗶-𝗧𝗵𝗿𝗲𝗮𝗱𝗲𝗱 𝗘𝗻𝗴𝗮𝗴𝗲𝗺𝗲𝗻𝘁 Don't count total activities. Count breadth and depth across the account A single power user logging in daily is a frequency of one. Five people across three departments engaging monthly is far healthier Track the trend. An account going from 2 active contacts to 6 over a quarter is accelerating. Going from 6 to 2 is a churn signal no matter how active those remaining 2 are 𝗠 = 𝗠𝗼𝗻𝗲𝘁𝗮𝗿𝘆 𝗣𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹, 𝗡𝗼𝘁 𝗝𝘂𝘀𝘁 𝗖𝘂𝗿𝗿𝗲𝗻𝘁 𝗦𝗽𝗲𝗻𝗱 Current ARR matters but it's incomplete. Score current spend relative to total addressable wallet An account paying you $200K when they could spend $2M is a very different score than one paying $200K at full penetration The segments that matter most: → High R, High F, Low M = engaged but underleveraged. This is your expansion pipeline → Low R, Any F, High M = big accounts going quiet. Most dangerous segment in your book. Every CS team needs automated alerts here Traditional RFM asks "what has this customer done for us" Enterprise RFM asks "how healthy is this relationship and where is it heading" That directional shift is what makes scoring predictive instead of descriptive Good luck out there scoring accounts + see previous wallet share post (TAW) Go forth and operate 👋 (more to come in this weekend's Substack) P.S. the Substack is thriving and growing. Thank you for your support 

Ever wonder why some brands seem to read your mind? It's RFM. Let me show you how. Recency, Frequency, Monetary value - the trifecta behind the curtain. By analyzing how recently and how often you engage with a brand, plus how much you spend, companies can predict your next move. Or try to persuade you to do something they want. 1️⃣𝗥𝗲𝗰𝗲𝗻𝗰𝘆: 𝗛𝗼𝘄 𝗿𝗲𝗰𝗲𝗻𝘁𝗹𝘆 𝗱𝗶𝗱 𝘁𝗵𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝗺𝗮𝗸𝗲 𝗮 𝗽𝘂𝗿𝗰𝗵𝗮𝘀𝗲? Imagine a customer who just purchased last week. They’re still excited about their new find. Capitalize on this enthusiasm with timely communications that thank them for their purchase or offer a complementary product as a follow-up. For instance, an online fashion retailer noticed a 30% higher email open rate from customers who had made purchases within the last month. ➟ Armed with this insight, they launched tailored email campaigns offering a "Welcome Back" discount to recent buyers and a "We Miss You" campaign to reactivate dormant shoppers. 2️⃣ 𝗙𝗿𝗲𝗾𝘂𝗲𝗻𝗰𝘆: 𝗛𝗼𝘄 𝗼𝗳𝘁𝗲𝗻 𝗱𝗼 𝘁𝗵𝗲𝘆 𝗯𝘂𝘆? Considered your regulars, the lifeblood of your business. A subscription-based meal delivery service found that customers who ordered more than twice a month were prime candidates for an upsell to a premium plan with more choices and exclusives. ➟ By targeting these frequent diners with personalized offers to enhance their plan, they not only boosted the average LTV but also reinforced customer loyalty. 3️⃣ 𝗠𝗼𝗻𝗲𝘁𝗮𝗿𝘆: 𝗛𝗼𝘄 𝗺𝘂𝗰𝗵 𝗱𝗼 𝘁𝗵𝗲𝘆 𝘀𝗽𝗲𝗻𝗱? High spenders are your VIPs. They expect—and deserve—a level of service commensurate with their expenditure. An online goods retailer used their data to identify customers spending over $500 per transaction. ➟ These high rollers were then offered access to an exclusive VIP program that included personal stylist consultations and early access to new products, enhancing their buying experience and encouraging even higher spends. By breaking down your customer base using these three metrics, you can tailor your marketing strategies to target specific groups more effectively. *************** I am Alvin Huang I'm an e-commerce veteran with over $189 million in sales, specializing in scalable growth and resilient leadership. I deliver no-nonsense, actionable insights for serious business growth. Follow me for real-world strategies and case studies that drive success. #RFMStrategies #customercentric #alwaysbeselling

I’m pleased to share my recent publication, “RFM-Based Customer Segmentation: A Pedagogical Case Study for Marketing Analytics Education”, in the Journal for Advancement of Marketing Education, with my colleague Henrique Carvalho. This paper provides a step-by-step guide to implementing customer segmentation using RFM analysis and Python, working with real transactional data from preprocessing through clustering and interpretation (Python code and dataset included!). For students and early-career analysts, this can serve as a portfolio project to demonstrate skills in data cleaning, feature engineering, segmentation, and clustering. For professors and instructors, the article offers a ready-to-use case study suitable for Marketing Analytics, Data Science, or Business Analytics courses. You can download the full article from here (it is open access!): https://lnkd.in/g_i3SHZM

🔍 Why Every Business Should Use RFM (Recency, Frequency, Monetary) — And How to Build It Step by Step Most companies talk about “knowing their customers,” but very few actually measure customer value in a simple, clear, and actionable way. That’s where RFM analysis comes in. It is still one of the most powerful customer analytics models — even in the age of AI. Let’s break it down in the simplest way 👇 --- 🔹 What is RFM? RFM is a method to score customers based on three behaviors: Recency: How recently did the customer buy? Frequency: How often do they buy? Monetary: How much do they spend? These three signals tell you: Who are your top customers Who is slipping away Who is ready for upsell Who needs reactivation And you can get all of this without complex AI or heavy data science. --- 🔹 How to Build an RFM Model Step by Step 1️⃣ Step 1 — Prepare the Data You only need three fields: Customer ID Historical transactions Transaction amount Clean the data, aggregate by customer, and calculate: Last purchase date Number of purchases Total spend --- 2️⃣ Step 2 — Create RFM Scores For each measure (R, F, M): Sort customers Divide them into 1–5 groups Assign a score (5 = best, 1 = lowest) Example: A customer who purchased yesterday = Recency score 5 A customer who didn’t purchase for 12 months = Recency score 1 --- 3️⃣ Step 3 — Build Customer Segments Using the Scores You can now group customers into business-ready segments: Champions (555) → Your VIPs Loyal (x5x) High-Value (xx5) At-Risk (1xx) Lost (111) Promising (x3x) These segments are simple to understand and very powerful for marketing. --- 4️⃣ Step 4 — Activate the Results This is where the value happens: Champions → Exclusive offers, new product launches, early access At-Risk → Win-back campaigns, discounts, reminders Good customers → Cross-sell and upsell Low-value → Automated nurture flows Dormant → Reactivation campaigns RFM is not a model for reporting. It is a model for action. --- 💡 Why RFM Still Drives Big Business Value? Because it gives you: Clear customer groups Quick activation opportunities Fast ROI No need for complicated models A perfect starting point before building predictive models like Churn or Next Best Offer Sometimes simple wins. --- 🧠 Summary & Takeaway RFM helps you understand customer behavior using three simple metrics. When activated properly, it becomes a strong engine for retention, cross-sell, upsell, and revenue growth. Every company—big or small—should start here before moving to advanced analytics. #CustomerAnalytics #DataStrategy #RFM #BusinessGrowth #AIForBusiness #DataAnalytics #CustomerValueManagement #MarketingAnalytics

My Favorite Analyses: the Recency-Frequency matrix. This simple yet powerful framework goes beyond traditional segmentation to provide actionable insights into customer behavior. By focusing on how recently and how often customers engage with your brand, you can tailor your strategies to maximize lifetime value. Why it works: - Recency: Customers who have purchased recently are more likely to purchase again. It's a strong indicator of engagement and future behavior. - Frequency: Customers who purchase more often demonstrate loyalty and satisfaction, leading to a higher customer value. Recency and Frequency are the most important indicators of customer value, exhibiting more correlation to CLV than Monetary Value which is the third component in traditional RFM analyses. The Recency-Frequency matrix helps you categorize your customers into segments based on behaviors instead of factors like demographics or psychographics that imply actions. The analysis reveals distinct customer segments that require unique marketing strategies, including your Champions, the customers who Need Attention, and those who have Already Churned. Implementing the Matrix: Depending on the size of your customer dataset, the Recency-Frequency matrix can be built in a spreadsheet or a more hefty tool like SQL or R. - Excel/Google Sheets: Use `MAXIFS`, `COUNT`, `PERCENTRANK`, and a pivot table to build the Recency-Frequency matrix, but watch out for row limits. - SQL: Leverage functions like `DATEDIFF` and `COUNT` to calculate metrics, and segment with `NTILE`. - R: The `RFM` package handles large datasets with ease, offering advanced segmentation and visualization. This approach isn’t just theory — it’s a data-backed method for ensuring your marketing dollars are spent where they’ll make the most impact. DM me if you'd like to learn more, including the marketing strategies that I most commonly recommend for each Recency-Frequency matrix customer segment. Art+Science Analytics Institute | University of Notre Dame | University of Notre Dame - Mendoza College of Business | University of Illinois Urbana-Champaign | University of Chicago | D'Amore-McKim School of Business at Northeastern University | ELVTR | Grow with Google - Data Analytics #Analytics #DataStorytelling #MyFavoriteAnalyses #ROI #MROI

🚀 New Project Drop – Azure Data Lakehouse (Bronze–Silver–Gold) for RFM Analysis in Power BI 🚀 Thrilled to share my latest end-to-end Microsoft Azure Data Engineering + Power BI project, where I designed and implemented a modern data lakehouse architecture to perform RFM (Recency, Frequency, Monetary) Analysis — a core customer segmentation method for business growth. 💡 What’s Inside: • Architecture: Bronze → Silver → Gold data layers in Azure Data Lakehouse • ETL/ELT: Data transformation pipelines to clean, standardize, and enrich datasets • Analytics: RFM scoring model to segment customers into actionable groups • Visualization: Interactive Power BI dashboard with drill-throughs, filters, and KPIs for instant insights 🔗 Full Project & Step-by-Step Blog (with visuals):Maven Showcase → https://lnkd.in/g4dzMBEq 💻 Code Repository: GitHub → https://lnkd.in/gNCr3CNu This project brought together my expertise in Azure Data Lakehouse architecture, data modeling, and BI storytelling, proving how cloud-native pipelines + business intelligence can create data products that truly drive decisions. I’d love your feedback — What’s one advanced metric you’d add to an RFM model for even deeper customer insights? #Azure #DataLakehouse #PowerBI #DataEngineering #RFMAnalysis #BusinessIntelligence #CustomerSegmentation #MavenShowcase #DataStorytelling

RFM analysis is one of the oldest and most effective segmentation techniques. In case you don't know, RFM stands for: 1️⃣ Recency: how long ago? 2️⃣ Frequency: how often? 3️⃣ Monetary value: how much $, or was a given goal achieved? Despite its proven effectiveness, RFM analysis remains under the radar for many professionals who mistakenly believe it’s too complex or requires a team of data scientists to execute. The truth? It’s a straightforward process that can yield incredibly valuable insights with just a few steps. Here’s how I leveraged RFM analysis for a client: • Data cleansing: A client provided a dataset of 65,000 transactions. I ensured all personal information, like customer names and emails, was anonymized, focusing only on transaction data. • Initial analysis: Using ChatGPT, I conducted an RFM analysis on a sample of the data. The output included the RFM values themselves, but also quintiles "bins" (grouped by slices of 20%). • Customized segmentation: I further refined the analysis by creating original segment names tailored to the client’s industry, complete with descriptions and targeted marketing tactics. • Visual enhancements: To make the insights more actionable, I added visualizations directly into the Excel output file, making the data easier to interpret and apply. • Automated efficiency: Finally, I asked ChatGPT to generate the complete Python code for the analysis and applied it to the entire dataset of 65,000 transactions—all in just a few seconds. RFM analysis isn’t just a relic of the past—it’s a practical, powerful tool that can be executed quickly and effectively, and powerful tools like ChatGPT makes it even easier! What could have taken many hours, if not days, was done in about an hour. RFM analysis isn’t limited to sales data—it can also be applied to behavioral data, provided you have a user or customer ID. In the days of Universal Analytics, marketers had easy access to metrics like the number of days since the last visit and visit frequency. With GA4, these insights aren’t as readily available unless you implement custom tracking or utilize BigQuery.

I spent 13 years sending millions of emails. Only 3 variables consistently made us money. At The Motley Fool, we built a monster company off of tremendous investing, and database marketing. The crazy part? A simple three-letter framework outperformed every “personalized” email anyone swore would be the next big thing. 🎯 Core Tenets of Database Marketing: RFM. -Recency (last click, last purchase, last email open) -Frequency (# of trades per year, # of logins, # of emails opened) -Monetization  ($ spent per year) That was the whole game. Personalization is great - once you’re “in” the product. But, what really drives revenue is RFM. Anything outside of RFM had to be tested first before we added it to our segmentation plan. 1/ The Metals System We created a simple tier system: Platinum, Gold, Silver, Bronze. Based entirely on RFM behavior. Then we made sure our best people were in our best products. Simple and highly profitable. 2/ The Testing Rule "Personalization" was all the rage at one point. Email marketers were segmenting by geography, age, industry, favorite color - you name it. But failed often to ask one very important question: Does this actually increase the response rate? Most didn't. Some did (investable assets was one that mattered). But age, Geography? Didn't move the needle. We only added variables that passed the test. 3/ Steal This: Before you segment by job title, company size, or industry vertical, ask: Have you tested if it actually improves conversion? Or are you just making your life harder? RFM first. Everything else gets tested. P.S. What's one segmentation variable you're using that you've never actually tested?