Mastering Buyer Segmentation utilizing Credit score Card Transaction Information | by Sadrach Pierre, Ph.D. | Jul, 2023

Utilizing RFM Scores to Construct Buyer Segments

Buyer segmentation is the method of figuring out buyer segments primarily based on historic buying patterns. For instance, it might contain figuring out repeat/loyal prospects, excessive spending prospects, prospects that make one time or rare purchases and far more. Segments may be created utilizing data like frequency of purchases, transaction quantities, buy dates and extra. All of those traits can be utilized to generate nicely outlined clusters with straightforward to interpret traits.

The traits of those segments can present quite a lot of data and enterprise worth for firms. For instance, firms can use these segments to extend income by means of focused promotional messaging, buyer retention campaigns, loyalty applications, product cross promoting and extra. Firms can goal buyer segments with tailor-made messaging that’s related to every section. Additional, these segments can present details about which channels prospects are most conscious of, whether or not it’s electronic mail, social media or exterior purposes. Firms may carry out upselling or cross-selling utilizing client segments. For instance, they will supply excessive priced choices for ceaselessly bought objects or complementary merchandise to beforehand bought objects. All of those techniques can be utilized to extend income and buyer retention.

There are all kinds of strategies used for buyer segmentation. One standard approach for producing buyer segments is recency frequency and financial worth (RFM) scores.

Recency, Frequency and Financial

1. Recency is the variety of days between the final buy made by a buyer and a reference date, often the present or max date accessible within the information.
2. Frequency is the variety of purchases between the date of their final buy and the present or max date accessible within the information.
3. Financial is the overall amount of cash spent between the date of their first and final purchases.

You need to use these values to assemble RFM Scores which can be utilized to section and establish excessive and low worth prospects. These scores can be utilized for all kinds of of enterprise use instances included customized advertising and marketing, churn evaluation, value optimization and extra.

Right here we are going to see tips on how to calculate RFM scores utilizing a bank card transaction dataset. For our functions we might be working with the Artificial Credit score Card Transaction information accessible on DataFabrica. The info comprises artificial bank card transaction quantities, bank card data, transaction IDs and extra. The free tier is free to obtain, modify, and share beneath the Apache 2.0 license.

For this work, I might be writing code in Deepnote, which is a collaborative information science pocket book that makes working reproducible experiments very straightforward.

Exploring the Information

To start out, let’s navigate to Deepnote and create a brand new challenge (you may sign-up totally free when you don’t have already got an account).

Let’s set up needed packages:

And import the packages we might be utilizing:

Subsequent, let’s learn our information right into a pandas dataframe and show the primary 5 rows of information:

Subsequent let’s filter our dataframe to solely embody prospects that bought Chick-fil-A:

Subsequent we are able to take a look at the quantity for purchasers by state. To do that we have to map the merchant_state to state abbreviations, which is able to enable us to plot the client counts for every state in Plotly:

Subsequent let’s map the state abbreviations and outline our state_count desk. We do that by performing a groupby nunique()operation for every cardholder for every state:

df['merchant_state_abbr'] = df['merchant_state'].map(state_abbreviations)
state_counts = df.groupby('merchant_state_abbr')['cardholder_name'].nunique().reset_index()
state_counts.columns = ['State', 'Customer_Count']

Subsequent we are able to use the Plotly Specific chloropleth technique to generate a geoplot of buyer counts for every state:

fig = px.choropleth(state_counts, places='State', locationmode='USA-states',
shade='Customer_Count', scope='usa',
color_continuous_scale='Blues',
title='Variety of Clients by State')
fig.present()

The total logic is:

Producing RFM Scores

Now let’s outline the logic for creating our RFM scores. Let’s begin by changing our transaction_date to a pandas datetime object and a NOW variable that’s the most transaction_date:

df['transaction_date'] = pd.to_datetime(df['transaction_date'])
NOW = df['transaction_date'].max()

Subsequent let’s carry out a groupby aggregation operation that enables us to calculate recency, frequency and financial worth.

1. Recency — max total date minus max buyer date: df.groupby('cardholder_name').agg({'transaction_date': lambda x: (NOW — x.max()).days})
2. Frequency — variety of transaction IDs for every buyer: df.groupby('cardholder_name').agg({transaction_id': lambda x: len(x)})
3. Financial worth — sum of transaction quantity for every buyer: df.groupby('cardholder_name').agg({transaction_amount': lambda x: x.sum()})

We’ll additionally convert transaction_date , which is remodeled to recency, to an integer:

rfmTable = df.groupby('cardholder_name').agg({'transaction_date': lambda x: (NOW - x.max()).days, 'transaction_id': lambda x: len(x), 'transaction_amount': lambda x: x.sum()})
rfmTable['transaction_date'] = rfmTable['transaction_date'].astype(int)

Subsequent, let’s rename our columns appropriately.

1. transaction_date turns into recency
2. transaction_id turns into frequency
3. transaction_amount turn out to be monetary_value

rfmTable.rename(columns={'transaction_date': 'recency', 
'transaction_id': 'frequency',
'transaction_amount': 'monetary_value'}, inplace=True)
rfmTable = rfmTable.reset_index()

The total logic is:

We will take a look at the distribution in recency:

Frequency:

And financial worth:

Subsequent we are able to calculate quartiles utilizing the Pandas qcut technique for recency, frequency and financial worth:

rfmTable['r_quartile'] = pd.qcut(rfmTable['recency'], q=4, labels=vary(1,5), duplicates='increase')
rfmTable['f_quartile'] = pd.qcut(rfmTable['frequency'], q=4, labels=vary(1,5), duplicates='drop')
rfmTable['m_quartile'] = pd.qcut(rfmTable['monetary_value'], q=4, labels=vary(1,5), duplicates='drop')
rfm_data = rfmTable.reset_index()

Subsequent we are able to visualize recency/frequency heatmap, the place every cell shows the share of shoppers with the corresponding recency and frequency values. First let’s calculate the chances:

heatmap_data = rfm_data.groupby(['r_quartile', 'f_quartile']).dimension().reset_index(title='Share')
heatmap_data['Percentage'] = heatmap_data['Percentage'] / heatmap_data['Percentage'].sum() * 100

Subsequent let’s generate our heatmap matrix:

heatmap_matrix = heatmap_data.pivot('r_quartile', 'f_quartile', 'Share')

Generate the map and label/title our heatmap utilizing Seaborn and Matplotlib:

sns.set()
sns.heatmap(heatmap_matrix, annot=True, fmt=".2f", cmap="YlGnBu")
plt.title("Buyer Segmentation Heatmap")
plt.xlabel("Frequency Quartile")
plt.ylabel("Recency Quartile")
plt.present()

The total logic is:

We will see the next insights from our heatmap:

1. 16.21% of shoppers bought not too long ago however sometimes.
2. 3.45% of shoppers are frequent and up to date prospects.
3. 10% of shoppers bought ceaselessly however not for a very long time.
4. 5.86% of our buyer haven’t bought not too long ago and don’t buy ceaselessly.

We’re solely contemplating the one service provider Chick-fil-A, however I encourage you to repeat this evaluation for among the different informal and high quality eating retailers.

Subsequent we are able to generate our RFM Scores by concatenating the quartiles in recency, frequency and financial worth:

And we are able to visualize the distribution in our RFM scores:

Right here we see that the most typical RFM rating is ‘411’ which corresponds to a latest buyer who spends sometimes and little or no.

Producing and Visualizing Buyer Segments utilizing RFM scores

Subsequent we are able to generate our buyer segments. This step is a bit subjective, however we outline out segments as follows:

1. Premium Buyer: r, f, and m all ≥ 3
2. Repeat Buyer: f >= 3 and r or m >=3
3. Prime Spender: m >= 3 and f or r >=3
4. At-Danger Buyer: two or extra of r,f and m <=2
5. Inactive Buyer: two or extra = 1
6. Different: anything

Subsequent we are able to see the distribution in segments:

Right here we see that the biggest buyer section is inactive prospects, adopted by at-risk, high spenders, repeat prospects and premium prospects.

And visualize the distribution as nicely:

We will additionally take a look at the common financial worth for every buyer section:

We see that the common financial worth is highest for premium prospects repeat prospects, and high spenders. Additional, inactive and at-risk buyer have low financial values.

The typical frequency worth for every buyer section:

We see that premium prospects, repeat prospects, and high spenders are probably the most frequency whereas at-risk and inactive prospects have low frequency.

Lastly common recency worth for every buyer section:

We see that the ‘different’ class have the latest purchases. This This can be section for “new” prospects whose buying patterns are inconclusive.

Utilizing Buyer Segments for Customized Advertising and marketing

You need to use these buyer segments to generate custom-made/personalised advertising and marketing messages related to every section. For instance, you may reward premium prospects with particular promotions and loyalty offers. It’s also possible to promote different merchandise that they could be fascinated with buying primarily based on their buy historical past. For repeat/loyal prospects you may develop automated electronic mail campaigns to maintain their loyalty. At-risk prospects are sometimes disengaged. We will develop campaigns primarily based on these buyer to reengage them and get them to start out buying once more. The same marketing campaign may be developed for inactive prospects. Lastly, for high spenders we are able to supply particular promotions and offers on extremely value merchandise that they’re prone to buy once more.

Utilizing this information, we are able to take our evaluation a step additional and use objects and costs fields to craft these custom-made advertising and marketing campaigns for these segments.

I encourage you to obtain the code from GitHub and repeat this evaluation by yourself for another restaurant retailers.

Conclusions

Right here we mentioned tips on how to carry out buyer segmentation on artificial bank card transaction information. We began by performing some easy information exploration the place we appeared on the depend of shoppers in every state for the service provider Chick-fil-A. Subsequent we calculated the columns wanted for producing RFM scores, recency, frequency and financial worth. We then confirmed tips on how to generate quartiles in recency, frequency and financial worth for every buyer, which we then used to assemble RFM scores. We then used the RFM rating to generate buyer segments “Premium Buyer”, “Repeat Buyer”, “Prime Spender”, “At-risk Buyer” and “Inactive Buyer”. Subsequent we generated some visualizations that allowed us to investigate the distribution in buyer segments throughout the info.

Utilizing RFM scores to generate insightful buyer segments is a useful talent for any information scientist that wishes to ship enterprise values. Establishing interpretable segments and pulling insights from these segments may help companies design advertising and marketing marketing campaign methods that may improve income and buyer retention. Understanding the buying habits of shoppers permits companies to tailor promotional gives to acceptable buyer bases. This text supplies the basics wanted to get began!

The free model of the artificial bank card information is right here. The total information set may be discovered right here.