What is RFM Analysis?
RFM analysis segments customers by Recency, Frequency, and Monetary Value. How to calculate the score, interpret the quadrants, and apply it in practice.
Many companies struggle to identify who their best customers are. They usually analyze only one metric and, as a result, end up summarizing customer behavior in a superficial way.
That’s where one of the main marketing analyses for companies that use data to make better decisions comes in: RFM analysis. By collecting three metrics, it is possible to segment a company’s customers into distinct groups and work with them one by one based on their specific characteristics.
This method is flexible—there is no single correct recipe for applying it. It is important to consider business requirements, adapting it for each situation and even evolving the method to match the company’s needs over time.
What will you learn here?
What RFM is and its purpose
How to evaluate the metrics used to calculate RFM
Calculate an RFM score for each customer
Segment customers according to their consumption profile.
What is it for? How does it work?
As mentioned above, RFM analysis is a method that uses three metrics to classify customers according to their consumption profile, with the goal of working each customer (promotions, win-back campaigns, cross-sell) according to their class. To do this, each customer will have their own RFM ‘score,’ in other words, a rating. We will evaluate the customer across the three metrics and assign a grade for each metric evaluated. We will use a scale from 1 to 4.
It is important to highlight that there are two parts to RFM scoring: in the first part, we only compute the metrics and assign them to RFM; in the second part, we translate the metrics into scores.
Therefore, we can work with RFM and RFM Score. In the end, we will have something like this:
Part 1: Metric assignment.
Part 2: Score assignment to classify the customer
Don’t worry if you didn’t understand the values in the example table—we’ll explain how to compute each metric and then score it. But here’s a hint: did you notice any relationship between the values assigned in the first table and the score in the second?
But after all, what exactly does RFM mean?
RFM is an acronym for English terms, and each term reflects one of the three main metrics we will evaluate. Therefore, each evaluated metric is directly tied to the term through the scores we assign to each of them.
R - Recency: In a literal translation, it means recency—that is, how recently the customer made their last purchase.
F - Frequency: Means frequency. How often does this customer buy? Are they an occasional customer or a loyal customer?
M - Monetary: This metric is usually understood as the customer’s total purchase value, but we will use average ticket to calculate it. Why?
We do this to create a clear distinction between value and frequency for each customer profile. Imagine the following scenario: one customer made 10 purchases with an average ticket of R$ 1,000, totaling R$ 10,000 in purchases.
A second customer made 2 purchases with an average ticket of R$ 5,000, totaling the same R$ 10,000 in purchases. If we used total amount spent, only frequency would differentiate them.
It is important to note that there is no single correct recipe when it comes to RFM Analysis. You can find different approaches, each with its own strengths and weaknesses.
The most important thing is to analyze whether the evaluated metric makes sense and whether this metric, together with the others, may benefit or harm specific situations.
Which metrics should we evaluate to score RFM?
This is the fundamental basis of RFM analysis. Which metrics should you choose? How should you work with these metrics and translate them into an RFM score? As stated earlier, there is no single correct recipe.
You can find different approaches, each with its own strengths and weaknesses. The most important thing is to analyze whether the evaluated metric makes sense and whether this metric together with the others will not benefit or harm specific situations. By the end of this article, you will understand and be able to evaluate which metrics make sense.
Recency:
We will use two metrics to calculate it: the current date and the customer’s last purchase date. It’s a simple subtraction—today’s date minus the customer’s last purchase date. In other words, the date difference will be your recency value.salin
Frequency:
Frequency is the number of events occurring within a given time interval. We invert numerator and denominator to make it more intuitive—after all, it is easier to interpret “3.66 days per purchase” than “0.28 purchases per day.”
In our case, events means orders, and the time interval is the period in which the customer has actually been a store customer. For this, we will use three metrics: current date, first purchase date, and number of distinct orders placed.
P: orders
d1: current date
d0: first purchase date
Practical example:
Monetary (average ticket):
Average ticket is already well known in the corporate environment. It is simply the average sales value per customer, i.e., that customer’s total revenue divided by the number of orders.
We have our base in hand! Now, how should we score RFM scores?
With these data alone, you already have a great start. It is possible to create scatter plots and relate variables to try to understand your customers’ consumption behavior.
Number of customers by RFM macro group.
Relationship between average ticket and frequency across RFM groups. Well, let’s step back a bit and change examples. First, it helps to understand a bit of the statistics used for calculation. If you already know quartiles, you can skip this section and go straight to the next step!
When we have an ascending ordered sample (that is, from smaller values to larger values), we can divide it into equal parts—as many as we want. In our case, we will use quartiles, meaning we will divide our ordered sample into four equal parts, hence the name “quartile.”
A quartile is the “mark” or “value” used as a reference to divide the sample into equal parts, like a cutoff score in a selection process. Let’s look at a practical example to make this clearer. We will divide a group of students into quartiles. First, we have the list of students and their respective ages:
Now we must sort the sample in ascending order:
With the sample sorted and divided into four parts, we must define the quartile boundaries. To do this, we take the average of the boundary values.
After obtaining the boundary values for each quartile, we separate the data again to finally define the sample within quartiles.
Our quartile grouping is ready!
Now let’s execute these steps applied to RFM analysis:
Since we use RFM score following a natural and logical order—that is, the higher the score, the better—we must pay attention when scoring recency (R) and frequency (F). For the company, the more recent the customer purchase, the better.
Customers who haven’t purchased in a long time should have a lower recency score. As a reminder: the recency metric is calculated in days since last purchase, so more is less—we prefer customers with low recency.
Therefore, quartile calculation for ‘R’ must be reversed: just calculate quartiles ordering from highest to lowest or assign the highest scores to the lowest quartiles (this is the approach we will use).
The same applies to frequency calculation. In our method, frequency is the time period for a customer order to occur—that is, how long it takes for the customer to buy again from the company.
The higher this value, the less this customer buys. So following the same logic explained earlier, we prefer customers with lower frequency.
Let’s go back to the company customer example, but this time with simpler values.
Recency (R): Sort the sample and assign scores inversely to quartiles, i.e., the first quartile receives the highest score.
Frequency (F): Sort the sample and assign scores inversely, just like recency.
Average ticket (M): Sort the sample and assign scores normally—the higher the average ticket, the better! Now we should synthesize the scores of each metric for each customer, thus obtaining the final RFM score result.
Excellent, all our customers are segmented! Now we can assign a descriptive title to each segment to make understanding and quick actions easier.
At Erathos, to meet our client’s requirements, we created four major classes and then subclasses to complement them. You can manipulate segmentation and define a specific description for each RFM Score in whatever way best suits your needs. In our case, we suggest the following classes:
Champion
1. Champion: These are the best customers—they buy frequently, are active, and spend a lot!
(4–4–4) / (4–4–3) / (3–4–4)
2. Potential champion: These are potentially top customers who have not yet reached the best frequency and recency levels.
(4–3–4) / (4–3–3) / (3–4–3) / (3–3–4) / (3–3–3)
3. Hibernating champion: A customer with a champion profile who hasn’t purchased in a while.
(2–4–4) / (2–4–3) / (2–3–4) / (2–3–3)
4. Lost champion: A customer with a champion profile who hasn’t purchased for a very long time and is in the last recency quartile.
(1–4–4) / (1–4–3) / (1–3–4) / (1–3–3)
Loyal
1. Loyal: Good customers who buy frequently and are active, but do not spend as much as champions.
(4–4–2) / (4–4–1) / (4–3–2) / (4–3–1) / (3–4–2) / (3–4–1)
2. Potential loyal: Customers who underperform in frequency or recency compared to loyal customers. To be a potential loyal customer, recency and frequency should not both be high simultaneously, as that would characterize a loyal customer.
(4–2–2) / (4–2–1) / (3–3–2) / (3–3–1) / (3–2–2) / (3–2–1)
3. Hibernating loyal: A customer with a loyal profile who hasn’t purchased in a while.
(2–4–2) / (2–4–1) / (2–3–2) / (2–3–1)
4. Lost loyal: A customer with a loyal profile who hasn’t purchased for a very long time and is in the last recency quartile.
(1–4–2) / (1–4–1) / (1–3–2) / (1–3–1)
Occasional
1. Occasional: A customer profile that does not buy much but has a high average ticket.
(4–2–4) / (4–2–3) / (3–2–4) / (3–2–3) / (3–1–4) / (3–1–3)
2. Hibernating occasional: Occasional customers who haven’t purchased in a while, i.e., low recency score.
(2–2–4) / (2–2–3) / (2–1–4) / (2–1–3)
3. Lost occasional: Customers who, like occasional ones, have a high average ticket but haven’t purchased in a long time.
(1–2–4) / (1–2–3) / (1–1–4) / (1–1–3)
Regular
1. Regular: Customers who purchased recently (active customers), but do not buy frequently and do not spend much.
(4–1–2) / (4–1–1) / (3–1–2) / (3–1–1)
2. Hibernating regular: Customers with low overall scores. What makes a customer regular is activity despite low frequency and average ticket. When this customer becomes inactive and recency gets low, they become a hibernating regular customer.
(2–2–2) / (2–2–1) / (1–2–2) / (1–2–1)
3. Lost regular: Customer profile with the lowest overall scores. In addition to the regular customer characteristic of not buying frequently and not spending much, this profile is also the least active of all.
(1–2–2) / (1–2–1) / (1–1–2) / (1–1–1)
It is important to highlight that there are several ways to segment. If this model is not the best alternative for your company’s business requirements, discuss with your team how to adapt it to your context.
You can also get in touch with us and we will be happy to offer our solutions.
So, what did you think of the article?
We hope we were able to help you understand and execute an RFM analysis for your company. If you still have questions, or if your team does not have data science professionals to run this and other analyses, get in touch with us—we’ll be glad to help!
Follow us and stay up to date with what’s new in the data-driven world!
Many companies struggle to identify who their best customers are. They usually analyze only one metric and, as a result, end up summarizing customer behavior in a superficial way.
That’s where one of the main marketing analyses for companies that use data to make better decisions comes in: RFM analysis. By collecting three metrics, it is possible to segment a company’s customers into distinct groups and work with them one by one based on their specific characteristics.
This method is flexible—there is no single correct recipe for applying it. It is important to consider business requirements, adapting it for each situation and even evolving the method to match the company’s needs over time.
What will you learn here?
What RFM is and its purpose
How to evaluate the metrics used to calculate RFM
Calculate an RFM score for each customer
Segment customers according to their consumption profile.
What is it for? How does it work?
As mentioned above, RFM analysis is a method that uses three metrics to classify customers according to their consumption profile, with the goal of working each customer (promotions, win-back campaigns, cross-sell) according to their class. To do this, each customer will have their own RFM ‘score,’ in other words, a rating. We will evaluate the customer across the three metrics and assign a grade for each metric evaluated. We will use a scale from 1 to 4.
It is important to highlight that there are two parts to RFM scoring: in the first part, we only compute the metrics and assign them to RFM; in the second part, we translate the metrics into scores.
Therefore, we can work with RFM and RFM Score. In the end, we will have something like this:
Part 1: Metric assignment.
Part 2: Score assignment to classify the customer
Don’t worry if you didn’t understand the values in the example table—we’ll explain how to compute each metric and then score it. But here’s a hint: did you notice any relationship between the values assigned in the first table and the score in the second?
But after all, what exactly does RFM mean?
RFM is an acronym for English terms, and each term reflects one of the three main metrics we will evaluate. Therefore, each evaluated metric is directly tied to the term through the scores we assign to each of them.
R - Recency: In a literal translation, it means recency—that is, how recently the customer made their last purchase.
F - Frequency: Means frequency. How often does this customer buy? Are they an occasional customer or a loyal customer?
M - Monetary: This metric is usually understood as the customer’s total purchase value, but we will use average ticket to calculate it. Why?
We do this to create a clear distinction between value and frequency for each customer profile. Imagine the following scenario: one customer made 10 purchases with an average ticket of R$ 1,000, totaling R$ 10,000 in purchases.
A second customer made 2 purchases with an average ticket of R$ 5,000, totaling the same R$ 10,000 in purchases. If we used total amount spent, only frequency would differentiate them.
It is important to note that there is no single correct recipe when it comes to RFM Analysis. You can find different approaches, each with its own strengths and weaknesses.
The most important thing is to analyze whether the evaluated metric makes sense and whether this metric, together with the others, may benefit or harm specific situations.
Which metrics should we evaluate to score RFM?
This is the fundamental basis of RFM analysis. Which metrics should you choose? How should you work with these metrics and translate them into an RFM score? As stated earlier, there is no single correct recipe.
You can find different approaches, each with its own strengths and weaknesses. The most important thing is to analyze whether the evaluated metric makes sense and whether this metric together with the others will not benefit or harm specific situations. By the end of this article, you will understand and be able to evaluate which metrics make sense.
Recency:
We will use two metrics to calculate it: the current date and the customer’s last purchase date. It’s a simple subtraction—today’s date minus the customer’s last purchase date. In other words, the date difference will be your recency value.salin
Frequency:
Frequency is the number of events occurring within a given time interval. We invert numerator and denominator to make it more intuitive—after all, it is easier to interpret “3.66 days per purchase” than “0.28 purchases per day.”
In our case, events means orders, and the time interval is the period in which the customer has actually been a store customer. For this, we will use three metrics: current date, first purchase date, and number of distinct orders placed.
P: orders
d1: current date
d0: first purchase date
Practical example:
Monetary (average ticket):
Average ticket is already well known in the corporate environment. It is simply the average sales value per customer, i.e., that customer’s total revenue divided by the number of orders.
We have our base in hand! Now, how should we score RFM scores?
With these data alone, you already have a great start. It is possible to create scatter plots and relate variables to try to understand your customers’ consumption behavior.
Number of customers by RFM macro group.
Relationship between average ticket and frequency across RFM groups. Well, let’s step back a bit and change examples. First, it helps to understand a bit of the statistics used for calculation. If you already know quartiles, you can skip this section and go straight to the next step!
When we have an ascending ordered sample (that is, from smaller values to larger values), we can divide it into equal parts—as many as we want. In our case, we will use quartiles, meaning we will divide our ordered sample into four equal parts, hence the name “quartile.”
A quartile is the “mark” or “value” used as a reference to divide the sample into equal parts, like a cutoff score in a selection process. Let’s look at a practical example to make this clearer. We will divide a group of students into quartiles. First, we have the list of students and their respective ages:
Now we must sort the sample in ascending order:
With the sample sorted and divided into four parts, we must define the quartile boundaries. To do this, we take the average of the boundary values.
After obtaining the boundary values for each quartile, we separate the data again to finally define the sample within quartiles.
Our quartile grouping is ready!
Now let’s execute these steps applied to RFM analysis:
Since we use RFM score following a natural and logical order—that is, the higher the score, the better—we must pay attention when scoring recency (R) and frequency (F). For the company, the more recent the customer purchase, the better.
Customers who haven’t purchased in a long time should have a lower recency score. As a reminder: the recency metric is calculated in days since last purchase, so more is less—we prefer customers with low recency.
Therefore, quartile calculation for ‘R’ must be reversed: just calculate quartiles ordering from highest to lowest or assign the highest scores to the lowest quartiles (this is the approach we will use).
The same applies to frequency calculation. In our method, frequency is the time period for a customer order to occur—that is, how long it takes for the customer to buy again from the company.
The higher this value, the less this customer buys. So following the same logic explained earlier, we prefer customers with lower frequency.
Let’s go back to the company customer example, but this time with simpler values.
Recency (R): Sort the sample and assign scores inversely to quartiles, i.e., the first quartile receives the highest score.
Frequency (F): Sort the sample and assign scores inversely, just like recency.
Average ticket (M): Sort the sample and assign scores normally—the higher the average ticket, the better! Now we should synthesize the scores of each metric for each customer, thus obtaining the final RFM score result.
Excellent, all our customers are segmented! Now we can assign a descriptive title to each segment to make understanding and quick actions easier.
At Erathos, to meet our client’s requirements, we created four major classes and then subclasses to complement them. You can manipulate segmentation and define a specific description for each RFM Score in whatever way best suits your needs. In our case, we suggest the following classes:
Champion
1. Champion: These are the best customers—they buy frequently, are active, and spend a lot!
(4–4–4) / (4–4–3) / (3–4–4)
2. Potential champion: These are potentially top customers who have not yet reached the best frequency and recency levels.
(4–3–4) / (4–3–3) / (3–4–3) / (3–3–4) / (3–3–3)
3. Hibernating champion: A customer with a champion profile who hasn’t purchased in a while.
(2–4–4) / (2–4–3) / (2–3–4) / (2–3–3)
4. Lost champion: A customer with a champion profile who hasn’t purchased for a very long time and is in the last recency quartile.
(1–4–4) / (1–4–3) / (1–3–4) / (1–3–3)
Loyal
1. Loyal: Good customers who buy frequently and are active, but do not spend as much as champions.
(4–4–2) / (4–4–1) / (4–3–2) / (4–3–1) / (3–4–2) / (3–4–1)
2. Potential loyal: Customers who underperform in frequency or recency compared to loyal customers. To be a potential loyal customer, recency and frequency should not both be high simultaneously, as that would characterize a loyal customer.
(4–2–2) / (4–2–1) / (3–3–2) / (3–3–1) / (3–2–2) / (3–2–1)
3. Hibernating loyal: A customer with a loyal profile who hasn’t purchased in a while.
(2–4–2) / (2–4–1) / (2–3–2) / (2–3–1)
4. Lost loyal: A customer with a loyal profile who hasn’t purchased for a very long time and is in the last recency quartile.
(1–4–2) / (1–4–1) / (1–3–2) / (1–3–1)
Occasional
1. Occasional: A customer profile that does not buy much but has a high average ticket.
(4–2–4) / (4–2–3) / (3–2–4) / (3–2–3) / (3–1–4) / (3–1–3)
2. Hibernating occasional: Occasional customers who haven’t purchased in a while, i.e., low recency score.
(2–2–4) / (2–2–3) / (2–1–4) / (2–1–3)
3. Lost occasional: Customers who, like occasional ones, have a high average ticket but haven’t purchased in a long time.
(1–2–4) / (1–2–3) / (1–1–4) / (1–1–3)
Regular
1. Regular: Customers who purchased recently (active customers), but do not buy frequently and do not spend much.
(4–1–2) / (4–1–1) / (3–1–2) / (3–1–1)
2. Hibernating regular: Customers with low overall scores. What makes a customer regular is activity despite low frequency and average ticket. When this customer becomes inactive and recency gets low, they become a hibernating regular customer.
(2–2–2) / (2–2–1) / (1–2–2) / (1–2–1)
3. Lost regular: Customer profile with the lowest overall scores. In addition to the regular customer characteristic of not buying frequently and not spending much, this profile is also the least active of all.
(1–2–2) / (1–2–1) / (1–1–2) / (1–1–1)
It is important to highlight that there are several ways to segment. If this model is not the best alternative for your company’s business requirements, discuss with your team how to adapt it to your context.
You can also get in touch with us and we will be happy to offer our solutions.
So, what did you think of the article?
We hope we were able to help you understand and execute an RFM analysis for your company. If you still have questions, or if your team does not have data science professionals to run this and other analyses, get in touch with us—we’ll be glad to help!
Follow us and stay up to date with what’s new in the data-driven world!