top of page

Customer Segmentation with Data: RFM, Cluster Analysis, and Segments That Pay Off

Writer: Linda Orr
Linda Orr
Apr 3, 2025
7 min read

Updated: 5 days ago

Ever wonder why some marketing campaigns hit the bullseye while others miss the mark? Very often the answer is segmentation. A campaign built for "our customers" is built for an average customer who does not exist, while a campaign built for a well-defined group can say the right thing to the right people at the right cost.


Here is the short answer. Start with RFM (recency, frequency, monetary value) on your transaction data, because it is fast and tells you who your valuable customers are. Move to behavioral or needs-based segmentation when you need to know why customers buy, using cluster analysis such as k-means for behavioral data and latent class analysis for survey data. Choose the number of segments by combining statistics with practicality, prove the segments differ on profit, and only then turn them into targeting, messaging, and budget decisions. The numbers in the examples below are made up so you can follow the arithmetic.


Woman shopping in a store, representing distinct customer segments

1. What is customer segmentation actually for?


Segmentation is a decision tool. A segment earns its place only if it changes who you target, what you say, what you offer, or how much you spend. The simplest test is to ask, "Would we change our marketing approach for this group?" If the answer is no, the segment is a description, and descriptions do not grow revenue.


That test also tells you what data to use. If the decision is how much to spend on retention, you need value and behavior. If the decision is what to say in your ads or what product to build next, you need needs and attitudes, which usually means a survey. Demographics alone rarely change a decision, because two 40-year-old women in the same zip code can buy for completely different reasons.


For a healthcare organization, I segmented nurse managers so the team knew which products, and which basket of products, to bring to each group.


2. Why is RFM the right place to start?


RFM uses three numbers you already have in your order or billing system: how recently each customer bought, how often they buy, and how much they spend. Score each customer from 1 to 5 on each dimension by splitting customers into fifths, so the most recent 20% of buyers get a recency score of 5 and the least recent 20% get a 1. A customer scored 5-5-5 bought recently, buys often, and spends a lot. A customer scored 1-4-4 used to buy often and spend well but has gone quiet, and that is your at-risk group.


From there you can group the 125 possible score combinations into a handful of named segments, such as champions, loyal regulars, promising new customers, at-risk, and lapsed. That alone is enough to stop sending the same email to everyone, to give your best customers early access, and to run a win-back program for the at-risk group. RFM has two limits. It only describes past buying, so it cannot tell you why anyone buys. And it says nothing about margin, so a customer who spends a lot on heavily discounted products can look like a champion while making you very little money.


3. When should you use behavioral or needs-based segmentation?


Behavioral segmentation goes beyond RFM to look at what people buy and how: product mix, category breadth, discount use, channel, time between orders, returns, and for software or services, which features or services they use. It answers questions like whether you have a group that only buys on promotion, a group that replenishes on a schedule, or a group that only shows up for gifts. In B2B, the equivalent combines firmographics such as industry and company size with account behavior such as product usage, contract value, and expansion history.


Needs-based segmentation asks why. It comes from a survey that measures what customers are trying to accomplish, what they value, and what frustrates them. This is the segmentation that drives messaging, positioning, and product decisions, and it is the foundation good personas should sit on. If your personas were written in a workshop instead of built on data, my post on how to build customer personas that drive revenue explains how to fix that. The strongest programs link the two: segment on needs from the survey, then find a way to identify each needs segment in your behavioral data so you can actually target it.


4. How does cluster analysis find segments?


Cluster analysis groups customers who are similar to each other and different from everyone else, without you deciding the groups in advance. The most common method is k-means. In plain English: you pick a number of segments, k. The algorithm drops k starting points into your data, assigns every customer to the nearest point, moves each point to the average of the customers assigned to it, and repeats until nothing moves. Because the result depends on where it starts, you run it many times from different starting points and keep the best solution.


One step people skip ruins most homemade clusters: putting variables on the same scale. If annual spend runs from $0 to $5,000 and order count runs from 1 to 20, distance in spend swamps everything else and your clusters are really just spend bands. The fix is to standardize each variable by subtracting its average and dividing by its standard deviation. For example, if average spend is $400 with a standard deviation of $400, a customer who spends $1,200 scores (1,200 minus 400) ÷ 400 = 2.0. If average orders are 3 with a standard deviation of 1.5, a customer with 6 orders scores (6 minus 3) ÷ 1.5 = 2.0. Now both variables count equally.


For survey data, latent class analysis is usually the better tool. Survey answers are categories and rating scales, which k-means handles poorly. Latent class analysis is a statistical model that assumes there are a few hidden groups in the population, each with its own pattern of answers, and estimates both the groups and each respondent's probability of belonging to each one. It comes with proper fit statistics for choosing the number of groups, and it lets you build a short "typing tool," a handful of questions that assign a new customer or prospect to a segment, which is how a needs-based segmentation gets used in sales conversations and onboarding.


5. How do you choose the number of segments?


There is no single correct number, so use several checks together. On the statistical side, for k-means you look at how much tighter the clusters get as you add one more, and stop where the improvement levels off (the "elbow"), along with a silhouette score that measures how cleanly customers separate. For latent class analysis you compare models with a fit statistic such as BIC, where lower is better.


Then apply the practical tests. Is every segment big enough to justify its own campaign or offer? Is the solution stable, meaning that if you rerun it on a random half of your customers, the same segments come back? Can you describe each segment in one plain sentence? And does each one lead to a different action? The right number is the smallest one where every segment passes all of those. A seven-segment solution that the team cannot remember will lose to a four-segment solution everyone uses.


6. How do you prove your segments differ on profit?


This is the step that separates useful segmentation from a pretty slide. Profile each segment on outcomes you did not use to build it: gross margin, retention, return rate, cost to acquire, and response to past offers. If the segments do not differ on those, they will not change your results, no matter how distinct they looked on the inputs.


Here is an example for a DTC brand with 10,000 customers and four segments built from behavioral data:


  • Loyal replenishers: 2,000 customers, $480 annual revenue each, 55% gross margin, $264 gross profit each, $528,000 in total

  • Deal seekers: 3,500 customers, $220 annual revenue each, 30% gross margin, $66 gross profit each, $231,000 in total

  • Gift buyers: 2,500 customers, $180 annual revenue each, 50% gross margin, $90 gross profit each, $225,000 in total

  • One-and-done: 2,000 customers, $60 annual revenue each, 40% gross margin, $24 gross profit each, $48,000 in total


Total revenue is $2.3 million and total gross profit is $1,032,000. Loyal replenishers are 20% of customers and 51% of gross profit ($528,000 ÷ $1,032,000). Deal seekers bring in $770,000, or 33% of revenue, but only 22% of gross profit ($231,000 ÷ $1,032,000). If you had ranked segments on revenue, you would have pointed your budget at the wrong group. Before acting, also confirm the differences are real and not noise, using confidence intervals or a simple statistical test, especially for small segments. Retention belongs in this check too, and my guide to calculating and modeling customer churn shows how to measure it by cohort.


7. How do segments turn into targeting, messaging, and budget?


Targeting comes first. In the example, you would build lookalike audiences from loyal replenishers, not from all buyers, and you would stop paying to retarget one-and-done customers. Messaging follows the reason each group buys: replenishment reminders and subscription offers for loyal replenishers, value bundles for deal seekers instead of ever-deeper discounts, and gifting occasions for gift buyers, timed to the seasons when they buy.


Budget should follow expected profit per acquired customer, not revenue. Suppose it costs $90 to acquire a customer who looks like a loyal replenisher and $40 to acquire one who looks like a deal seeker. First-year gross profit per dollar of acquisition cost is $264 ÷ $90, about 2.9, for the replenisher and $66 ÷ $40, about 1.65, for the deal seeker. The cheaper customer is the worse buy. To go further, measure how each channel performs for each segment, which is where a marketing mix model earns its keep, because it tells you what each channel actually contributes instead of what the platforms claim.


8. When do you need an expert for segmentation?


You can build an RFM segmentation in a spreadsheet, and you should start there. Bring in help when you need a needs-based segmentation from a survey, because questionnaire design, latent class models, and a working typing tool take real statistical training. The same goes for combining data from several systems, segments that look different but do not differ on profit, healthcare data with privacy rules, or a segmentation that will drive a big decision such as a new product line, a new market, or a major budget shift. My market research services cover exactly this work, from survey design through segments your team can use.


If you want segments that change where your money goes, you can book a free 30-minute introductory call here.


Related service: brand strategy and positioning consulting, including brand audits, positioning and repositioning, buyer personas, and messaging.


Dr. Linda Orr, PhD, is the founder of Orr Consulting, a fractional CMO and marketing analytics practice serving brands from $5M to $500M in revenue. She built Upwork's first enterprise marketing mix model, covering an $88M annual media portfolio, and spent about 20 years as a marketing professor before moving fully into consulting.



Featured In Leading Media & Journals

Quoted in national media, published in peer-reviewed journals, and the author of five business books.

Forbes logo, linking to the Forbes article quoting Dr. Linda Orr on solo marketers
Entrepreneur logo, linking to Dr. Linda Orr’s articles on Entrepreneur.com
Yahoo logo
The Daily Upside logo, linking to the Daily Upside article quoting Dr. Linda Orr on AI marketing and brand risk
business.com logo, linking to the business.com article featuring Dr. Linda Orr on offline marketing tactics
Glossy logo, linking to the Glossy Fashion Briefing article quoting Dr. Linda Orr
Journal of the Academy of Marketing Science logo, linking to Dr. Linda Orr’s article in the journal
Journal of Business Research logo, linking to Dr. Linda Orr’s article in the journal
Psychology & Marketing logo, linking to Dr. Linda Orr’s article in the journal
Journal of Business Ethics logo, linking to Dr. Linda Orr’s article in the journal
Marketing Management Journal logo, linking to Dr. Linda Orr’s article A Re-Inquiry of Hofstede’s Cultural Dimensions

Why Orr Consulting

You work directly with a PhD-level strategist.

When you work with Orr Consulting, you work directly with me. I set the strategy, decide where the budget goes, and stay accountable for the results.

 

I also lead the people you already have: your internal team, agencies, developers, analysts, creatives, and specialists. You get senior direction without replacing the team you have built.

 

When you need more hands, my team does the implementation: tracking and CRM setup, web pages, email marketing, Meta and paid social, SEO and AEO, and video and image editing. I manage Google Ads myself and review everything the team delivers, so you have one point of accountability.

 

My background combines 25+ years of marketing leadership, two decades as a marketing professor, and a PhD in Marketing with depth in statistics and psychology. That work includes building Upwork's first marketing mix model across an $88M annual media portfolio.

 

Dr. Linda Orr, PhD, Founder

Dr. Linda Orr, founder of Orr Consulting

SELECTED CLIENTS AND COMPANIES

Selected Orr Consulting clients: PepsiCo, Upwork, McKesson Foster and Thrive, Midmark, FedEx Custom Critical, Arm and Hammer, TheraBreath, Mavrik Dental Systems, medmix, Elos, GFOA, Indeed, Magenta Inc, PreVu Dental, SMRTFT, Sparks, The Critical Thinking Institute, Tahoe Modern, Mayberry Law, Ollie Pediatrics, Therapevo, Mind On Photography, Artemis, Disc It
Selected Orr Consulting clients: PepsiCo, Upwork, McKesson Foster and Thrive, Midmark, FedEx Custom Critical, Arm and Hammer, TheraBreath, Mavrik Dental Systems, medmix, Elos, GFOA, Indeed, Magenta Inc, PreVu Dental, SMRTFT, Sparks, The Critical Thinking Institute, Tahoe Modern, Mayberry Law, Ollie Pediatrics, Therapevo, Mind On Photography, Artemis, Disc It (continued)

What Clients Say

Most engagements are under NDA. Here is what clients say, plus our reviews on Google.

★★★★★

“Dr. Orr's exceptional leadership and advanced knowledge in sales and marketing were invaluable. She provided practical tools our team could use immediately to perform with confidence.”

B. MacDonald

B2B Sales Enablement · Sales Training

★★★★★

“Linda was a pleasure to work with and extremely thorough. She led a market research project for our 3D dental imaging launch, and the insights directly shaped our positioning and go-to-market strategy.”

Midmark Dental

Healthcare · Market Research · Launch Strategy

★★★★★

“Working with Linda was an absolute pleasure, and the results she delivered were beyond our expectations! ... She interviewed over 20 key players in the yachting industry, ranging from yacht charter company owners to marketing managers, extracting valuable insights that we wouldn't have been able to gather on our own... The final report she presented was top-notch. It was thorough, well-structured, and packed with actionable recommendations that demonstrated her deep understanding of the industry.”

Founder, Yacht Charter Marketing Agency

Market Research · Market Entry Strategy

Prefer to Schedule Directly?

Pick a time that works for you, or start with our free guide to getting found in AI search.

Contact

Thanks for submitting!

bottom of page