RFM analysis used for customer segmentation based on SAP current/historical Sales Order Doc.

Using the current or/and historical Sales Order data extracted from SAP, it is important to perform data pre-processing or cleansing (ie: : Checking for data with duplicates, null values, negative values) before performing RFM analysis .

RFM Analysis is used to classify groups/clusters of customers (ie: BEST Customers, High-Spending New Customer, Lowest Spending Active Loyal, Loyal, Potential Churn) to generate differentiated marketing promotional campaign to increase customer retention, loyalty and customer lifetime value.

Apart from RFM analysis mentioned earlier, machine learning (ML) methods are applied to product association with association rule mining that are used to identify patterns across products and find correlations in customer purchase behaviour. Machine learning (ML) method can be achieved using Sales Order history data (ie: transactional dataset) extracted from SAP, provided that the Sales Order doc. comprises of multiple different products (ie: different Sku) for the machine learning algorithm to perform product association (ie: association rule mining).

Association rule mining are used across a wide range of applications including product recommendations as seen in Amazon’s e-commerce platform, digital media recommendations that have brought notoriety to companies like Netflix, and other areas such as politics and medical diagnosis to name a few. Any firm(s) can be drawn to association mining as it enable us to better understand customer purchasing behavior that can translate into product bundling or cross-selling purpose. Hence, it’s not only useful for marketer for campaign or strategic product planning, but also customer service specialist role useful for operational planning.

1. Exploratory Analysis of Historical/Current Sales Orders

1.1 — Product aspect analysis

• 1.1.1 Below illustrates frequency (count) of products_sku sold with respect to each SAP Sales Doc. & summation of Order_Qty.

Table below is sorted in descending order of freq_transaction_count , presenting insights on the most popular products and sales volume (order_qty).

• 1.1.2 Below is a time-series sales pattern/trend of different products (ie: nickname). Since exponential smoothing models are based on a description of the trend and seasonality in the data whilst ARIMA models aim to describe the autocorrelations in the data. From this time-series data visualization below, it provides insights whether exponential smoothing or ARIMA/ARMA (autoregressive moving average) can be used as the model for prediction.
• According to researchers, deep learning algorithm is capable or suitable in modelling data that has no seasonality or regression trend.

1.2 — Customer aspect analysis

Identify customer’s spending behaviour whether there’s often a One-Time purchase which is undesirable. With this chart, it helps any firm to identify the root cause and make correction action plans to improve the situation.

From the historical/current Sales Order (ie: Transactional Dataset), RFM Scoring Table can then be generated with each segments based on quartile being assigned to customers. Based on these RFM scoring, it classifies customers into specific segment types (ie: BEST Customers, High-Spending New Customer, Lowest Spending Active Loyal, Loyal, Potential Churn).

— The easiest way to split metrics into segments is by using quartile.

• This gives us a starting point for detailed analysis
• 4 segments are easy to understand and explain

RFM Table (Reference)

RFM Analysis based on the available dataset

— Lowest recency, highest frequency and monetary are our best customers

* For recency a good customer would be a part of the lowest quartile designated as ‘1’

* For frequency and monetary a good customer would be a part of the highest quartile here designated as ‘1’

1. Best customers

• These are the customers that bought recently, buy often and spend a lot. It’s likely that they will continue to do so.
• Since they already like you so much, consider marketing to them without price incentives to preserve your profit margin.
• Be sure to tell these customers about new products you carry, how to connect on social networks, and any loyalty programs or social media incentives you run.

2. Big Spenders

• Big spenders have spent a lot of money over their lifetime as your customer.
• They trust you enough to invest a lot in your products.
• Considering marketing your most expensive products and top of the line models to this group.

3. Loyal customers

Customers with a high frequency should be considered loyal. This doesn’t mean they have necessarily bought recently, or that they spent a lot, though you could define that with R and M factors.

4. Deadbeats

These customers spent very little, bought very few times, and last ordered quite a while ago. They are unlikely to be worth much time, so put them in your general house list and consider a re-opt-in campaign.