jlk – You’ve probably shopped at a supermarket or online store and noticed certain products being paired or recommended together. For example, when buying toothpaste, you might also be offered a toothbrush.
Or when purchasing clothes, you may be suggested to buy matching pants. Is this a coincidence? Certainly not. It’s the result of a data mining technique known as market basket analysis.
Market basket analysis is a technique used to analyze patterns of product purchases frequently made by consumers in a single shopping basket.
Its goal is to understand consumer behavior by identifying relationships between purchased products. Consequently, retail businesses can boost sales by:
- Placing frequently bought products together in close proximity, both physically (on store shelves) and virtually (in e-commerce site catalogs).
- Offering promotions, discounts, or incentives for frequently purchased products, such as bundled packages or cross-selling.
- Enhancing customer retention by providing product recommendations based on their needs and preferences.
- Expanding market share by targeting potential consumers with similar purchasing patterns as loyal customers.
Market basket analysis employs an algorithm called association rules mining, which generates association rules between purchased products. These association rules take the form of:
{Product A} => {Product B}
This implies that if a consumer buys Product A, they are also likely to purchase Product B. To evaluate the strength of these association rules, several metrics are used:
- Support: The relative frequency of Products A and B being purchased together in all transactions. Higher support indicates more frequent occurrences of the association rule.
- Confidence: The proportion of transactions buying Product A that also buy Product B. Higher confidence signifies a stronger relationship between Products A and B.
- Lift: The ratio of support to the product of individual supports for Products A and B. Higher lift indicates a more significant relationship between Products A and B.
For example, considering 10 shopping transactions:
One association rule formed from this data could be:
{Pet Food} => {Syrup}
This means that if a consumer buys pet food, they are also likely to buy syrup. The metrics for this association rule are:
- Support: 4/10 = 0.4, meaning 40% of all transactions involve the simultaneous purchase of pet food and syrup.
- Confidence: 4/5 = 0.8, indicating that 80% of transactions buying pet food also buy syrup.
- Lift: 0.4/(0.5 x 0.6) = 1.33, suggesting that the likelihood of buying syrup after purchasing pet food is 1.33 times greater than random chance.
Based on these metrics, it can be concluded that the association rule {Pet Food} => {Syrup} is strong and frequently observed.
Retail businesses can leverage this rule to enhance sales by placing pet food and syrup together on shelves or the same page on e-commerce sites, offering discounts for their combined purchase, or recommending syrup to customers buying pet food.
However, market basket analysis has limitations, such as its inability to capture cause-and-effect relationships between purchased products, the exclusion of other factors influencing consumer behavior (e.g., demographics, psychographics, seasons, trends, or competitor promotions), and the need for large, high-quality transaction data to generate valid and reliable association rules.
Therefore, market basket analysis should not be used as the sole tool for business decision-making but rather as one source of information that can be combined with other methods like cluster analysis, sentiment analysis, or predictive analysis. This way, retail businesses can gain a more comprehensive and accurate understanding of their customers and market.