There are many methods that use data to analyze customer behavior. From a business point of view, a very important question that we need to answer is how much is the client worth to the company. This is known as customer scoring. Based on such information, we are able to determine the efficiency of marketing campaigns and decide whether a customer is important enough for us to maintain advertising spendings. The correct assessment of customer value is of the utmost importance for the business, as it translates into time and money savings: when it comes to communication, building loyalty, preventing churn and retaining clients.
This post is third in the series of articles on customer churn analysis. You can find my previous entry in this series here:
Why it is worth to group customers and estimate their value?
Customer value analysis (also known as CVA) is an analytic method for discovering customers’ characteristics and on this basis making further analysis of specific clients to extract useful knowledge from large data.
Enterprises apply value analysis methods in order to define which customers generate major profits for the company, and therefore apply this information in further communication with the most valuable segments. The RFM (stands for Recency, Frequency, Monetary) model is one of the most commonly used customer value analysis methods.
The main advantage of RFM analysis is that it provides meaningful information about customers while using fewer (three-dimensional) criterions as cluster attributes. This reduces the complexity of customer value analysis model without compromising its accuracy.
Moreover, when it comes to analyzing consumer behavior, the RFM model is once again a widely used method to measure the strength of the consumer-brand relationship. It is known that it costs way more to acquire a new customer than to retain an existing one. Hence, companies use RFM analysis to mine databases for valuable insights about customers who tend to spend more and therefore are the most valuable for them. This information allows to take special care of the most profitable clients and therefore prevents their retention.
What are the analytical methods used for RFM analysis and customer scoring?
When segmenting and evaluating the customer value, we can use different methods that are very often used to group clients and assess their worth to the company. Such actions can be performed using simple data mining techniques, e.g. the above-mentioned RFM analysis, or more advanced techniques that employ, among others, machine learning models to go even deeper and detect more unconventional segments based on patterns of high complexity extracted from the data.
In this article, I would like to discuss the approach to the preparation of the customer assessment process using RFM analysis, the k-means algorithm, and hierarchical clustering.
During the first phase, the data set is prepared for further modeling. Thus, we first delete the records with missing values and then transform the correct values for clustering.
The RFM indicator (Recency, Frequency, Monetary) is one of the most popular tools for valuing customers based on their previous purchases. It is especially used by marketers who deal with direct marketing. With RFM analysis, they can determine the value of the client (which is of considerable use in its subsequent application), but also predict his or her response to future marketing activities. This way, marketing campaigns can be planned with more accuracy and therefore become more cost-effective.
Components of RFM analysis:
R – Recency
How much time has passed since the last purchase? Basically, the shorter the time, the higher the value of the client.
So, the first step should be to divide the entire customer base into 5 equal segments.
- The value of 5 is attributed to 20% of customers who have recently made purchases.
- The value of 1 is attributed to 20% of customers whose last purchase took place a long time ago.
F – Frequency
How often or how many times has the customer made purchases in our store? The more often he has shopped, the higher his value.
If you already have a base divided into 5 segments based on the date of the last purchase, now it’s time to divide the recipients once again into 5 equal groups due to the number of purchases made by them in the entire history of the relationship with your brand. Similarly to the first segmentation, we assume that 5 is the highest score and is assigned to those who made the most purchases, while 1 is for those who didn’t buy much.
M – Monetary
How much money did the customer leave in our store? Of course, the higher the income, the higher the customer value.
Now the time has come for the last part of the analysis – determining how much money the customer has spent overall on your products. Just like in the previous segmentation, in this case, it’s recommended to use a scale from 1 to 5.
At this phase we can use various methodologies for customer clustering, for instance, the widely known k-means or hierarchical clustering. For both methods, we will take advantage of the characteristics determined previously during the RFM analysis phase.
K-means clustering is one of the simplest and most popular unsupervised machine learning algorithms.
To process the learning data through data mining, the k-means algorithm start with the first group of randomly selected centroids, which are used as the beginning points for every cluster. Then it performs iterative (repetitive) calculations to optimize the positions of the centroids.
The process of creating and optimizing clusters comes to an end when:
- The centroids have stabilized – there is no more change in their values because the clustering has been successful.
- The defined number of iterations has been achieved.
Another clustering method is hierarchical clustering. There are two types of this clustering technique:
Agglomerative Hierarchical Clustering Technique
In this technique, initially, each data point is considered an individual cluster. At each iteration, the similar clusters merge with other clusters until one cluster or K clusters are formed.
To put it simply, we can say that the Divisive Hierarchical clustering is exactly the opposite of the Agglomerative Hierarchical clustering. In Divisive Hierarchical clustering, we consider all the data points as a single cluster and in each iteration, we separate the data points from the cluster which are not similar. Each separated data point is considered as an individual cluster. In the end, we’ll have n clusters.
As we’re dividing the single clusters into n clusters, this method is known as Divisive Hierarchical clustering.
Based on the results obtained, we can prepare a loyalty classification.
How to combine customer scoring with churn analysis?
The combination of the two described activities: RFM analysis and customer clustering allows the organization to better target marketing campaigns. You can also merge it with churn analysis in order to predict customer churn and prevent the most valuable clients from leaving.
If a given customer is characterized by a high churn rate, there is a high possibility that this customer will leave soon. Applying the customer scoring, we can check whether this customer is valuable for the company. If not, the actions to retain this customer can be given up. The company can decide it’s not worth to try to keep a customer that made only one purchase a long time ago.
However, if the probability of losing this client is high and the RFM indicator is also high, meaning the client is of significant value to the company – then we can put much more effort and use non-standard measures to retain this client.
Reducing the effort to retain unprofitable customers can result in significant savings. Similarly, paying more attention to the customer of higher value can prevent him from churning and increase his loyalty.
The above-mentioned methods are just a few examples picked among other existing options. The same applies to the Machine Learning methods used for grouping. For each implementation, you need to select the set of algorithms individually, according to the type of data available for the specific project.