Customer Segmentation

Not All Customers Are the Same

You already know this instinctively. The uncle who comes to your store every morning for milk and bread is not the same customer as the college student who shows up once a month for a big snack haul. The working professional who orders groceries on BigBasket every Sunday is not the same as the homemaker who browses Meesho for 30 minutes before adding one item to the cart.

Yet most retailers treat all customers the same — same discounts, same messages, same offers. This is like a doctor prescribing the same medicine to every patient. It wastes money on customers who would buy anyway, and fails to reach customers who need a specific nudge.

AI-powered customer segmentation groups your customers by how they actually behave — not just who they are. This chapter teaches you the most practical segmentation framework in retail (RFM analysis), how AI enhances it, and how to use segments to personalize at scale.

RFM Analysis: The Retailer's X-Ray

RFM stands for Recency, Frequency, and Monetary value. It is the simplest and most powerful way to understand your customer base.

Each customer gets a score (typically 1-5) on each dimension. A customer who bought yesterday (R=5), buys weekly (F=5), and spends heavily (M=5) is your best customer. A customer who last bought 8 months ago (R=1), bought only once (F=1), and spent Rs 200 (M=1) is nearly lost.

The Five Customer Segments

Based on RFM scores, customers naturally fall into these groups:

Open data/customer-transactions.csv in the code panel. This dataset contains 12 months of transaction data for 2,000 customers of an online fashion retailer — customer ID, purchase date, order value, items purchased, payment method, and city. This is the raw material for RFM analysis.

How AI Enhances Basic RFM

Manual RFM gives you 5 segments. AI-powered segmentation goes further:

Beyond Three Dimensions

AI can incorporate dozens of behavioral signals beyond RFM:

UPI vs. COD: A Behavioral Signal Unique to India

In India, how a customer pays tells you a lot about them:

A customer who switches from COD to UPI on their third order is showing increasing trust — AI flags this as a signal to offer loyalty benefits. A customer who insists on COD after 10 orders might be avoiding digital trails for tax reasons or simply prefers the familiarity.

Tier-1/2/3 City Buying Patterns

India's tiered city structure creates distinct customer profiles that AI must account for:

Tier-1 Cities (Mumbai, Delhi, Bangalore, etc.)

Tier-2 Cities (Jaipur, Lucknow, Kochi, etc.)

Tier-3+ Cities (Smaller towns, rural areas)

AI segmentation that ignores city tier will misclassify customers. A "low-value" customer in Tier-3 spending Rs 500/month might be proportionally more loyal than a Tier-1 customer spending Rs 2,000/month.

Open data/rfm-analysis.json to see pre-computed RFM scores for the 2,000-customer dataset, including segment assignments, city tier classifications, and payment method breakdowns.

Personalization at Scale

Once you have segments, you can personalize every customer interaction:

Email and WhatsApp Campaigns

Product Recommendations

AI can recommend products based on segment behavior:

Pricing by Segment

This is sensitive territory (see Chapter 3 on ethics), but some approaches are accepted:

Building Your First Segmentation

Using AI Chat (No Tools Required)

Here is my customer purchase data for the last 6 months:
[paste: customer_id, purchase_date, order_value, items, city]

Please:
1. Calculate RFM scores (1-5) for each customer
2. Assign them to segments (Champions, Loyal, Potential
   Loyalists, At Risk, Lost)
3. Tell me the size of each segment
4. Suggest one specific action for each segment
5. Highlight any customers who recently moved from Loyal to
   At Risk (they need immediate attention)

What to Do With the Results

Key Takeaways