How automated recommendations increased sales in online store?

Business Challenge

Company X, a high-quality product supplier, was struggling with a low Average Order Value (AOV). Despite significant website traffic and customer interest, the value of individual transactions remained unstable and stagnant.

An initial audit revealed several key areas for optimization:

• Incomplete Purchase Process - customers were often purchasing only one item, failing to realize that their needs frequently required a complete set of complementary products.
• Untapped Product Relationships - there was no systematic knowledge of natural product pairings (e.g., buying coffee almost always requires filters and sugar). These patterns were invisible to the store owners.
• Suboptimal Site Layout - products were displayed randomly, forcing customers to expend excessive effort in finding complete solutions.

For Company X, this represented a direct financial risk: every unnoticed purchasing pattern was lost revenue that could have been recovered through intelligently designed product suggestions.

Business Goal

The primary goal was to build a robust basket analysis model. The objective was not merely to identify best-selling items, but crucially, to statistically prove and leverage the patterns by which products are purchased together. Success was measured directly by increasing the Average Order Value (AOV).

Implementation

While this project involved elements similar to academic coursework—using data exploration techniques for basket analysis—the real-world implementation required a more complex, three-stage approach:

Step 1: Data Audit and Customer Segmentation

I integrated the full set of transactional data (what was purchased) with behavioral data (how long users spent on specific categories). Since the data was already well-processed, I could focus quickly on analysis. We conducted an initial customer segmentation to understand if purchasing patterns differed among new, occasional, or loyal customers.

Step 2: Association Rule Mining Using Basket Analysis

Following customer segmentation, I extracted product relationships using association rule mining (basket analysis). Simply stating that products are bought together was insufficient. The key was applying and filtering three critical statistical metrics:

• Support - defines the frequency with which a specific set of products appears in a basket.
• Confidence - measures the probability of purchasing Product B, given that the customer has already purchased Product A.
• Lift Score - this was the most crucial metric. The Lift Score indicates whether the relationship is stronger than expected by chance. Only rules with a Lift Score above 1.2 were considered statistically significant and actionable.

Step 3: A/B Testing of Product Relationships

To prove the causality behind the AOV increase, I did not implement the recommendation system globally. Instead, I first conducted a controlled comparative test on key purchasing paths.

• Control Group (A) - standard site layout without any product suggestions.
• Test Group (B) - layout featuring implemented, statistically verified recommendation blocks ("Customers Also Bought," "Complete Your Set").

Results and Recommendations

The system implementation had an immediate, measurable impact on Company X's e-commerce operations. The 22% increase in AOV was achieved through the introduction of automated recommendations ("Customers Also Bought...") and overall product layout optimization. Furthermore, Company X identified a secondary area for improvement: inventory management in light of increased sales of complementary products.

Recommendation for Your Business

Do not allow your sales potential to be limited by random purchases. Implementing professional basket analysis is an investment that guarantees the maximization of every customer's value and delivers immediate revenue growth.