Data mining is a set of techniques for the automated discovery of statistical dependencies, patterns, similarities or trends in very large databases. Of course, for those involved in sales techniques and correcting sales results, data mining is an extremely valuable tool.
Market basket analysis help increase profits and improve competitiveness.
Due to ease of obtaining large amounts of online data, data mining (here web mining) has become an interesting issue for e-commerce, especially for exploring purchasing trends by analyzing customer market baskets. This involves finding a relationship between purchased products by discovering links using simple rules, known as the rules of association. Market basket analysis can effectively present product offers, create more effective promotions and develop more effective marketing campaigns. Simply put, they help increase profits and improve competitiveness.
Construction of association rules
While analyzing a market basket we are looking for associations. This is done by combining elements together and finding the relationship between them, simultaneously discovering the shopping habits of customers. This is done with special algorithms that will be presented in the next post. The result of the analysis are rules of association in the form of : A → B, ie. IF [body] then [head]. The body and head can be freely extended and consist of a conjunction of many conditions and create a more complicated rule.
But let’s start with a simple example: [frozen pizza ] → [ketchup]. This rule assumes that if a customer bought a frozen pizza there is a substantial likelihood that it will buy ketchup. Of course, it is not necessarily true one hundred percent of the time. The fact that these two products are strongly linked determine specific indicators which define quality rules.
Measures of quality rules
The quality of the rules, that is how much they are true and valuable from the business point of view, outlines several measures: support, confidence and lift.
- Support for the rule is the percentage of transactions containing the rule in all transactions in a retail outlet. In our example support determines the probability of buying both frozen pizza and ketchup by a randomly selected customer in a particular store.
- Confidence is the probability that the B product will appear when the A product also will be present. In our case it is likely that the customer will buy ketchup when he buys frozen pizza.
- Lift informs the impact of the sale of the product A on sale of the product B. If the lift of value is 1, then we say that the products do not affect each other, if it is less than 1, we consider the products to be antagonistic, and if is greater than 1 the products are complementary
Let’s take another example in which we highlight support and confidence. Suppose that the sales report in a supermarket shows that last Thursday, of 1,000 customers, 200 bought loaves of bread, and from those who bought loaves of bread, 50 bought butter as well. Here we see a rule that if someone buys bread, also buys butter, which support is 5% (50/1000), and the confidence is 25% (50/200).
Market basket analysis in e-shops
Initially, the analysis of market baskets included transactions in supermarkets. Now, with the increasingly growing importance of e-commerce, analysis of market baskets in e-stores seems even more relevant, because the internet is the perfect place to get to know user habits. Transactional data collected in an automated way from online stores provides reliable information about customers and the products they buy. Additionally, the utility developed on the basis of the rules and the effectiveness of their actions can be easily estimated by, for example, calculating the ROI.
The most common use of association analysis is finding products which are complementary to each other and are therefore commonly consumed together. It positively affects sales – knowing that the customer buys product A, you can show him products B, C and D, which are usually purchased together with A. The online bookstore Amazon can be used as an example here, where while we search a particular title, we also get a list of books with similar themes. These include items that other customers bought together with the title (or viewed before making a purchase) and those that are recommended by other customers. But for the purpose of market basket analysis in e-commerce you can use also other types of association rules:
- Negative association rules discover links between the products present in the basket and those which are not. On this basis, you can find significant customer buying behavior like that when someone buys Coca-Cola but not Pepsi, or if someone buys juice he will not buy bottled water. Negative rules are equally useful in the preparation of an online store offer as positive rules (presenting products sold together), because it lets you dynamically modify the appearance of the page so that the customers do not see the content they do not want to buy.
- Cyclic association rules take into account the repeatability/cyclical nature of purchases over time and thereby customize the appearance of the website to the customer.
- Inter-transactional association rules do not look for dependencies in a single transaction but examine the relationship between purchases made in a certain period of time. These rules take into account the context of shopping (eg. time, place, customers), which is ignored in the intra-transactional rules. They aim to find sequences of events.
- Ratio association rules – in addition to information about purchased products they also contain information about the amount of money spent on various goods and services. As a result, it is possible to predict the value of the transaction and the purchase of specific items from the e-store’s offer.
Data mining is often characterized as “mining for dollars”
The most important is the relevant data. Data mining is often characterized not as “the search for rules in the database”, but as “mining for dollars” because it helps to increase profits. It allows you to optimize marketing campaigns by identifying the probability of the purchase of individual products, develop effective methods for cross-selling (selling a product or service to the customer associated with the purchase of another) and up-selling (selling more expensive product versions). It is necessary to know customers’ buying habits if you want to construct an offer consistent with their expectations.
After finding high-quality rules you can easily reach each customer with a suitable offer with dynamic content available on the site (in the form of banners and pop-ups) and email. You can also profile the message based on customer segments. And if, despite a personalized offer, your customers add products to the market basket but don’t buy them, you can automatically send them an email regarding the items in the abandoned basket and a discount to encourage them to complete the transaction.