An Analytics Blog by Naga Krothapalli

It is often easier to sell your products or services to existing customers than to acquire new customers. How should we cross sell? You can try to sell every product to every single customer. This is certainly a strategy, but not necessarily an effective one. You can even risk customer attrition from poor customer experience. Instead, we can learn from previous transactions of all customers, and propose right set of products that will be well received and purchased by customers. A couple of companies that do this well are  Amazon in product recommendations and Netflix in movies recommendations. I am sure there are several other good examples. Now, how can we do the same? How can you promote products that are likely to be purchased by a customer? This post attempts to answer that question.

Market Basket Analysis

You have probably heard the story where a retailer found that diapers and beer are purchased together by men on Fridays. Whether this is true or an urban legend, valuable cross sell insights can be found  from customer purchase patterns.  Association rules are often used to identify such patterns from customer transaction data, which is also known as market basket analysis. I will use a publicly available groceries dataset in R-arules package to identify customer insights.

Customer Transactions

You need to first collect information about your customer transactions and create a transaction set. Following is a sample of transactions from the groceries dataset. Your dataset can be limited to a particular time period, a store/region, or set of targeted customers based on your focus.

1 {citrus fruit, semi-finished bread, margarine, ready soups}            
2 {tropical fruit, yogurt, coffee}                 
3 {whole milk}             
4 {pip fruit, yogurt, cream cheese, meat spreads}           
5 {other vegetables, whole milk, condensed milk, long life bakery product}

Cross Sell Insights

I have used Apriori algorithm to identify association rules from the groceries data. These rules can be interpreted for different purposes.

• Common Cross Sell Patterns: Identify most common purchase patterns and make changes to product promotions to take advantage of that information. In the following rules, you will notice that whole milk is closely associated with a few products such as tropical fruit, root vegetables, yogurt, pip fruit, other vegetables, butter, and whipped/sour cream. This is an opportunity to place these products in close proximity so that we can take advantage of these purchase patterns. You can apply same principles to your product set. In addition to physical proximity, you can package promotions of products together.

  lhs                     rhs              support confidence     lift
1 {tropical fruit,                                                   
   root vegetables,                                                  
   yogurt}             => {whole milk} 0.005693950  0.7000000 2.739554
2 {pip fruit,                                                        
   root vegetables,                                                  
   other vegetables}   => {whole milk} 0.005490595  0.6750000 2.641713
3 {butter,                                                           
   whipped/sour cream} => {whole milk} 0.006710727  0.6600000 2.583008

• Specific Product Promotion: In some instances, you are interested in growing sales of a specific product. In this case, you limit your analysis to all the rules containing your product on the right hand side. Some of the insights you may get are trivial and some will be interesting. For example, putting  bottled beer next to liquor and red wine is not exactly a breakthrough insight, but it at least ensures that right product promotions are done. For domestic eggs, you see the need to cross promote with whole milk, butter, whipped/sour cream, root vegetables, and other vegetables. You can see similar insights around soda. You can prioritize the products critical to your profitability and find ways to increase their sales through cross-promotion and packaging.

  lhs                 rhs                support confidence     lift
1 {liquor}         => {bottled beer} 0.004677173  0.4220183 5.240594
2 {red/blush wine} => {bottled beer} 0.004880529  0.2539683 3.153760

lhs                     rhs                 support confidence     lift
1  {other vegetables,                                                    
    whole milk,                                                          
    butter}             => {domestic eggs} 0.003050330  0.2654867 4.184394
2  {root vegetables,                                                     
    butter}             => {domestic eggs} 0.003253686  0.2519685 3.971331
3  {other vegetables,                                                    
    whole milk,                                                          
    whipped/sour cream} => {domestic eggs} 0.003558719  0.2430556 3.830852

lhs                        rhs        support confidence     lift
1 {rolls/buns,                                                    
   candy}                 => {soda} 0.003050330  0.4285714 2.457726
2 {yogurt,                                                        
   rolls/buns,                                                    
   bottled water}         => {soda} 0.003050330  0.4285714 2.457726
3 {bottled water,                                                 
   shopping bags}         => {soda} 0.004067107  0.3703704 2.123961
4 {bottled water,                                                 
   fruit/vegetable juice} => {soda} 0.005185562  0.3642857 2.089067

Applications

There are several applications of association rules in identifying cross-sell opportunities and benefiting from them. Here are some examples:

• Retailer: Promote commonly purchased products in the marketing material. Place them in physical proximity and for an online store, place product recommendations next to individual product pages.
• B2B Company: Bundle services and promote them. Provide price discounts for bundled services if necessary.
• Online Publisher: Package advertising solutions. Provide related keywords or behavioral targeting segments based on advertiser’s current purchases.

Segment Transaction Data

Clustering and segmentation of transaction data can help you create actionable insights that are more accurate and flexible.

• Time period: You might see different patterns based on day of the week, hour of the day, or day of the month. Make sure that you are interpreting the insights based on time period of customer transactions.
• Region: Compare insights from one region to another as they can significantly vary, such as comparing a warm place to cold place or urban to rural.
• Demographics: Customer purchase patterns often vary by demographics of the customer. By comparing cross-sell opportunities by demographics, one can apply the insights appropriately based on customer profile, especially in an online environment. You can use other customer identifiers to segment customers and learn more about cross-sell opportunities for specific targeted groups.

There is plenty of data you have on your customer transaction and hope you take advantage of it. Please share your thoughts and feedback.