THIAGARAJAR SCHOOL OF

MANAGEMENT (TSM), MADURAI

CUSTOMER RELATIONSHIP MANAGEMENT- RFM

ANALYSIS

Dr. Nataraj, RFM Analysis Page 1

RFM Analysis

What is the RFM Model?

There are many models that can be used to segment customer data. If your
organization is not currently using a framework to drive segmentation, we
recommend starting with the RFM model because it’s simple and effective.

While RFM analysis can be helpful for B2B companies, it is typically most useful in
B2C businesses where customers make multiple purchases over time. The RFM
model was created in the mid-90s to help catalog marketers target specific customer
segments within their databases that had the greatest likelihood to purchase based
on previous behavior. The goal was to save on printing and shipping costs. Today,
the RFM model is used by digital marketers to send relevant marketing messages,
keep their customers engaged, and maximize the lifetime value of each customer.

What are RFM Scores?

RFM models can be customized based on your organization. However, in most RFM
models, each individual customer is given a score from 1- 5 for each RFM
component, with 1 being the lowest and 5 being the highest score. We then combine
those scores (not the sum) to create at a 3-digit RFM score.

RFM scores range from 111 to 555. An important takeaway from this post is there
are no bad RFM scores. Each RFM score should inform if and what messaging to
surface that customer. An RFM model that scores customer from 1 to 5 setups a
quintile so, when a customer scores a 1, that means they fall into the bottom 20% of
all customers. Similarly, if a customer scores a 5, it means they fall into the top 20%
of all customers. Note: this is not based on an absolute value but relative (compared
to all other customers).

• Recency – How long has it been since the customer’s last purchase?
• Frequency – How often has the customer made a purchase over a defined
period of time
• Monetary – How much money did the customer spend with us over a defined
period of time?

Dr. Nataraj, RFM Analysis Page 2

Together, these three individual scores give us an accurate snapshot of each
individual customer and help us determine what kind of offer, if any, we should
target them with. Let’s look at a retail example to understand how RFM scores work
in practice.

Despite the current love affair with predictive models, direct marketing’s three-
variable formula, Recency-Frequency-Monetary Value (RFM), still has a place in
modern database marketing. RFM is not a replacement for inferential statistics. But
in the real world, RFM can still be useful when models are not practical. RFM also
provides a management summary of customer behavior based on purchases and
plays a role in policing the black-box results of predictive models to ensure quality
before a campaign is implemented.

Segmentation of Customers
What comes to mind when you read these descriptions of customer segments?
Advocates
Repeat buyers
Gift givers
Too good to be true
o Habitual returners
o Fraud
Trial buyers
Once was enough
Dormant
Defected
o About to Defect
o Revolving Door

RFM Scores to Inform Customer Segments

Once you have set up your RFM model and have assigned customers RFM scores,
your team can use those RFM scores to divide customers into customer segments.
This enables a powerful framework to identify segments of customers and make
informed decisions around how to deliver a better experience.

Here are some standard customer segments:

• Fresh leads – RFM score = 511

• New and promising customers – RFM Score = 514
• Loyal customers – RFM score = 453 (the frequency is what makes them loyal)
• At risk customers – RFM score = 324
• Cannot lose – RFM score = 155, 245

Dr. Nataraj, RFM Analysis Page 3

s

Once customers are broken into segments, you can leverage these segments to power
your email marketing campaign strategies.

Don’t Just Focus on High Scores

High RFM scores will show you who your most promising customers are, which
enables your team to build experiences that keep them engaged. However, lower
scores can help you identify:

1. New customers that need to be nurtured.

2. Previously loyal customers that have since found a replacement for your
products or services and should be entered into a win-back campaign.
3. Customers who are buying low-value items on a regular basis and are prime
candidates to be moved up the chain to higher value items.
4. Lost customers who are not worth marketing to/wasting marketing resources
on.

How to Use RFM Scores for Smarter Remarketing

RFM scores can also be useful for powering paid advertising campaigns to deliver
integrated cross-channel customer experiences. Your team can trigger remarketing
campaigns automatically based on RFM scores using Salesforce MC’s Ad Studio and
Journey Builder. If you’re just getting started with remarketing using MC, you can

Dr. Nataraj, RFM Analysis Page 4

always export MC data to create ad audiences in your tool of choice, like Google
Ads. Once you’re leveraging RFM scores for smarter remarketing, the next step is to
leverage Einstein Engagement Scores (bit.ly/sfeinstein).

Einstein Engagement Scores in Salesforce is another great tool that can be combined
with RFM scores to help make better marketing decisions. Einstein Engagement
scores indicate how likely an individual customer is to open and engage with your
email, based on their history. Your team can layer these scores on top of your RFM
segments to determine how to best engage with each customer.

For example, if we know a user always engages with our emails, we may not want to
waste our ad budget on them and engage with them only via email. Or, if a customer
isn’t likely to open our emails, it might make more sense to target them with social
ads.

Trigger Automated Campaigns Based on RFM

With Journey Builder and Ad Studio, you can also leverage your RFM model to
trigger automated campaigns at every stage of the customer lifecycle across all
channels. Think about how your scoring model can help you automatically identify
users that should be entered into a Welcome series, Win-back, or Loyalty campaign.
Note: be sure to consider if and how you’ll exclude subscribers from existing in
multiple campaigns at the same time.

Dr. Nataraj, RFM Analysis Page 5

Wrap Up
Hopefully, this article provided you with actionable insights and answers to some of
your questions. If you enjoyed this post, you may want to check out our follow-up
post: How to Build an RFM Model for Customer Segmentation. In it, we tackle
everything your team needs to know to implement an RFM model, including:

• How to wrangle existing customer and transaction data to build your RFM
model
• The basic SQL skills necessary for this undertaking and how your team can
learn them with free and online resources
• How to calculate RFM scores
• How to sync Google BigQuery with your Salesforce Marketing Cloud
instance

Each of these customer segments is rooted in purchase behavior. Purchase behavior

is the best predictor of repeat purchasing and loyalty. While it is measured in
different ways, depending on industry and customer lifecycle, all database
marketers covet its empirical facts on how often buyers renew their
subscriptions/memberships, visit your site, and shop at your store. Further,
purchase behavior is about how much they spend, the products/services they buy,
and in what combination or sequence. Purchase behavior codifies both the tenure as
well as the recency of your relationship with your customers.

The good news for today’s marketers is that the most important purchase behavior is
already inyour customer database. What can you do with these data? You can:
• Identify groups of customers
• Target them for campaigns
• Promote repeat purchase and loyalty
• Defend against attrition/defection
• Acquire customers who resemble the best ones

Segmentation gives direct marketers a quantifiable way to distinguish between the

best and worst customers on file. Purchase behavior is the most powerful way
to segment your customers by historic value. You can also gain insight with
purchase behavior
from outsides sources of information such as co-op databases, which are a collection
of hundreds of direct marketing lists pooled together; individual response lists; and
any merge/purge processing itself (with its resulting intra- and inter- matches).
These data provide contextual dimensions and will help you realize there is more
going on in your customers’ lives than just their relationships with you.

Dr. Nataraj, RFM Analysis Page 6

Other kinds of segmentation bring you closer to why your customer buys from you,
particularly from primary research that adds their attitudes and experiences. This
information, along with demographics, explains their motivations and brings
customer segments to life. This area of research is the most interesting and strategic,
and has an impact on benefits customers seek, media consumption habits and
advertising strategy. It is critical input for decisions around what to say and how to
say it.

Recency – Frequency – Monetary Value (RFM)

Recency Frequency Monetary Value (RFM) is a quick, descriptive way to segment a
marketing database on purchasing behavior that direct marketers have used with
success since the 1930s. What is RFM? How do you use it? Is it superior to inferential
statistics such as predictive models and decision trees? Generally, there is no contest
against statistical modeling; RFM will lose just about every time. While this eBook
may have just come to a screeching halt, the balance will argue that RFM still has an
important place in today’s modern direct marketing as a complement to predictive
models, particularly for non-statisticians (Figure 1).

Dr. Nataraj, RFM Analysis Page 7

If You’re Forced to Choose, Choose ‘Both’

I choose to use both predictive models and RFM side by side. My rationale is that
predictive models do a superior job of predicting sales and my first priority is to
make money. Predictive Models have a better return on investment than
segmentations based on RFM.

However, RFM is applicable when modeling isn’t practical. For example, perhaps
your database is too small to warrant the investment in building the model. The fees
for building predictive models can run in the tens of thousands, while RFM is
essentially a do-it-yourself proposition; and, as a result, much cheaper. Finally,
building a model takes time, especially since you need to go back in time, build a
regression equation on a group of customers, validate the result on an equivalent
group, implement the model by scoring today’s database, and then pull the trigger.
RFM values, by contrast, can be applied to your database by the time you get to the
end of this paper. My favorite role for RFM is as a management tool. By predictor
variables, models gain predictive power, but lose the ability to explain the reasons
why it works. This is a black box for management, as well as a missed opportunity to
understand the key business drivers of the dependent variable (response, sales per

Dr. Nataraj, RFM Analysis Page 8

campaign, loyalty). It’s true; there have been admirable steps forward in pairing up
Regression models for predictive power with Principal Component Analysis and
treeanalyses to explain its building blocks. However, these are additional
investments in themselves that require time and money. My suggestion is to use
RFM for understanding, in concert with predictive models for campaign execution.

Validate a Predictive Model with RFM before You Contact Customers

How can you validate that a model is correct before you pull the trigger? What are some basic
diagnostics you can run before you put the offers into the marketplace?

When brings you very attractive F-Statistics and R2 compelling you to roll the dice
on your next campaign, you should verify their work as your first step. First, you’ll
be reviewing the decisions they made with regards to sample sizes, statistical and
sampling techniques, treatment of outliers, treatment of incomplete information
(blanks, nulls, garbage), transformation of variables and so on. But you should also
ask for a crosstab comparing their “score” (outcome of the model) versus RFM.
This crosstab should contain not only population counts, but also ratios such as LTD
Dollars/Buyer, Average Order Size, Average Days since Last Purchase, and LTD
Orders/Buyer. You’ll be looking for a correlation between the best scores and the
best ratios. If there are attractive ratios in the basement of your predictive model,
you may be missing out on opportunities with good customers.

A Modern Take on Recency

It is part of Green Bay Packer folklore that coach Vince Lombardi starting a training
camp, attended by professional athletes, snorted “This, is a football!” The implication
was to start from scratch and re-learn the game (his way). I’ve always related to that
sentiment in a more positive light and interpreted it to mean learn everything you
can about a particular topic, whether that’s a pulling guard sweep to the weak side
(football jargon) or customer segmentation. As business people who leverage
databases to be better direct marketers, what can Recency, the R of the RFM model,
teach us about our customers and marketing programs? First, here’s a definition:
Recency (R) is defined as the time since last purchase, or meaningful transaction, that
your customer makes. The more recent the last action, the higher the likelihood your
customer will respond to the next e-mail, phone solicitation, direct mail campaign,
etc. It is operationalized as the number of months since the last purchase, but the
unit of measure can easily be changed to weeks or days for online businesses. You
should always use the update date or the high date1 on the database, not “today’s”
date. What is the Date of Last Purchase that makes sense for your business?

Recency has important business applications beyond segmentation. It is a key

business dimension and can be triangulated (same customer - two recency dates) to
see customers with a deeper understanding. (See Figure 2.)

Dr. Nataraj, RFM Analysis Page 9

Show Me the Data
A report card on a season’s worth of campaigns can illustrate the power of Recency
on its ability to segment customers on response rate and sales. This simulation is a
powerful way to understand the range of quality of your customers (See Figure 3).
I’m presenting this as a management summary, not as a suggestion on how to
execute your campaigns.

Dr. Nataraj, RFM Analysis Page 10

It is clear to see that “sales per piece” and its first cousin “response rate” march to
the beat set by Last Purchase Date, the row in this report. As managers, we’re
looking for ways to distinguish between good and bad customers. Clearly, the recent
customers (0-3 months since their last purchase) are superior to the dormant
customers (no purchase in 2 years). This can be charted on dimensions including
revenue per buyer as well as contribution margin per buyer (especially as it relates
to a threshold for breaking even). See Figure 4.

Dr. Nataraj, RFM Analysis Page 11

A Modern Take on Frequency

Frequency (F) is defined as how often a buyer has a dollar earning transaction. It is
typically a “Life to Date” field, and thus would be the accumulation of all
transactions from an original date to the update date. This value includes the
original transaction (e.g., enrollment event, first purchase, etc).
The modern take on Frequency is that it is more interesting to understand the
difference between 1x and 2x, versus 10x and 11x. That may seem odd, especially
since the 10 time buyers are your loyal advocates, and they are disproportionately
important to the health of your business. That’s true, however, when planning
campaigns the trial buyers need all the help they can get from your customer
database and beyond. (Using RFM, I’m defining 0-3M 1x as a trial customer, whereas
19M + 1x is a goner for most businesses). What else can we use for 1x buyers? I’ve
found several important clues as to which one-time trial buyers will blossom into

Dr. Nataraj, RFM Analysis Page 12

advocates and which won’t. Here are some idea starters: Depth: # items Breadth: #
of departments (distinct “aisles” in the store) Price points
Demographics - Geography (zip code, clusters) is more important for 1x
buyers (and critical for 0x buyers)
Contextual information
From co-op databases
1x for you; 10x for your competitors
1x for you; new to catalog shopping
From the merge/purge
Match to rentals (1x with lots of matches is far more promising
than 1x with no matches)
“Multi-Buyers”: 1x multi buyer is a customer who did not match another list
in a merge/purge
2x multi matched 1 other list
3x multi matched 2 other lists
From the a b2b site
1x buyer in a loyal, high-revenue site is more promising (and may already be
a specifier-decision maker) than a 1x buyer in a new 1-buyer site.
Source of the customer
Self-selection (did they come to you, or did you bribe their first purchase with
a $-off coupon?)

Dr. Nataraj, RFM Analysis Page 13

Breaking down Recency ranges by Frequency ranges will add to your understanding
of your customers. Not all 1x buyers are bad (particularly the more recent ones). Not
all 3x+ buyers are loyal (particularly the less recent ones).

Again, segments with better RF scores garner more attractive results (sales per piece)
from a 6-month campaign.

Sales per piece now range from a high of $24 (for 0-3M 3x+) to a low of $1.34 (for
19M+ 1x).

Dr. Nataraj, RFM Analysis Page 14

Charting margin per buyer gives marketing managers more information to make
campaign decisions.

Viewing the chart with RF scores on its x-axis now illustrates that the 1x portion of
the 13-18M segment is below breakeven, but repeat buyers in the same recency
ranges are above.

Dr. Nataraj, RFM Analysis Page 15

A Modern Take on Monetary Value

Monetary Value (M) is the sum of all revenue earned. A judgment is used to decide
between “Life to Date” dollars, “Average” dollars, or some dollar amount over time
(“0-12 month” dollars). Adding the third leg of the RFM triumvirate makes our
charts resemble what’s called a screeplot. Scree plots borrow their name from the
junk that falls off very high mountains. The RFM chart now looks like giant peaks of
high customer value compared with run off of lower value segments. See if you
agree: compared to your best customers, everybody else is just scree.

Some might call this the proverbial “80/20” rule. My colleague Chris Pickering calls
it “Pickering’s Law of Disproportionate Value,” which is only fair given that the
definitive book on RFM is called: Libey and Pickering on RFM and Beyond,
published by MeritDirect Press.

RFM Model
The biggest complaint about RFM models is the use of cells and the unwieldy
number some RFM models produce. A large number of cells defeats the purpose of
ease of use for management understanding.

Dr. Nataraj, RFM Analysis Page 16

Rather than create a cell-based approach to RFM (where R would have month ranges
such as 0-3, 4-6, 7-12 …, F would have ranges like 1x, 2x, 3x+, and so on), the
breakthrough thinking is to create a continuous RFM score for each customer on
your database. This score is one value for each customer and lends itself to sortation
of the customers by value, and the decile summaries

Dr. Nataraj, RFM Analysis Page 17

Best Practices for using RFM

Use common sense. With durable goods and large ticket items (Cars, furniture,
etc) RFM may actually work in the reverse (where less recent is better than more
recent). This may also be true with seasonal businesses. Also, be flexible in how you
operationalize the values within the RFM model. For example, web marketing and e-
commerce have their own RFM. Recency may be measured in days, not months.
Frequency may turn into how frequently a visitor returns to your site, what they
purchase or view and where they click. M might be extended to include what’s in
the shopping cart (before abandons).

Seasonality makes defining the R ranges a challenge. Does your 0-12 M range
include “Back to School” , Valentines Day, Holiday Shopping, etc.,

Augment RFM where you can, if it helps you understand your business dynamics,.
Bob Kestnbaum pioneered the concept of RFMP, adding a product dimension to
RFM. Kestnbaum called it “FRAC” where “Frequency” was the first variable since
he thought it was most predictive, “A” was “Average Dollars,” and “C” was
“Category of Purchase.” His rationale is sound: The best predictor of future Product
A purchases is past Product A purchases. This addresses the challenge of
seasonality. Your key business levers are also viable candidates to extend the simple
RFM summary. For b2c direct marketing, this might be RFMIncome. For b2b, this
might be RFMIndustry. Both might see benefits in an RFMChannel segmentation
scheme.

Extend RFM for acquisition marketing. RFM can also be put to work to acquire new
customers. There is some current thinking on calculating RFM for each zip code on
file to target prospects based on observed RFM for customers that live in the
neighborhood.

RFM is not a replacement for inferential statistics. It should add value in its unique
ability to add management understanding, ensure data quality, and substitute for
predictive models when models are not practical. RFM has a role in modern direct
marketing.

Dr. Nataraj, RFM Analysis Page 18