Business Introduction
The client is a prominent pharmaceutical and wellness retail chain with a widespread presence across major cities and towns. Along with prescription medicines, they offer a wide variety of FMCG products focused on wellness, beauty, and personal hygiene—available via physical stores, an e-commerce website, and a mobile app. Their stores operate 24/7, ensuring healthcare access around the clock.
With over two decades of excellence in retail and numerous awards—such as Best Omnichannel Retailer and Most Admired Healthcare Retailer—the client continues to expand rapidly. As the business scaled, so did their need for intelligent marketing strategies that could increase revenue, optimize promotions, and identify which products to bundle together in campaigns.
To move beyond gutfeel and guesswork in their bundling approach, the client sought a data-driven solution. That’s when Logesys was brought in—to unlock patterns in massive volumes of transaction data and power smarter bundled promotions.
Business Objectives
The core goal was to perform market basket analysis on recent sales data to identify profitable product combinations that could be bundled together helping the marketing team craft better campaigns, clear excess inventory, and increase average transaction values.
The analysis needed to:
- Identify which products are frequently bought together
- Recommend bundles of 2 to 4 items with strong buying patterns
- Exclude combinations that wouldn’t add business value (e.g., two top-selling items that already perform well individually)
- Handle 20+ million records with high efficiency
Scope of Work
Logesys was entrusted to design and implement an end-to-end machine learning solution that would take six months of transactional data, analyze patterns, and deliver actionable bundling insights via an interactive Power BI dashboard.
Challenges & Solutions
Challenge 1: Massive Transactional Data
With over 20 million records across just two quarters, the volume of historical data was overwhelming. Extracting relevant relationships manually or through traditional tools was simply not scalable.
Solution:
Logesys utilized PySpark within Azure Databricks for efficient parallel data processing. By distributing the load across multiple nodes, preprocessing that once took hours was now completed in a fraction of the time—improving speed tenfold. This allowed us to consistently analyze recent sales and deliver real-time suggestions for campaign planning.
Challenge 2: No Readymade BI or ML System
The client relied on their team’s experience and judgment to decide on bundle promotions. No automated solution or predictive system was in place to drive decisions with data.
Solution:
We built a custom machine learning model leveraging the Apriori algorithm in Python. Apriori is widely known for its precision in market basket analysis. It identified associations between items and suggested bundles based on support and confidence metrics, focusing on pairs, triads, and quartets of products that frequently appear together.
Challenge 3: Performance and Accuracy Concerns
Off-the-shelf analytics platforms either lacked the scale or failed to provide usable, timely results. Accuracy in detecting strong bundle patterns without overloading the system was critical.
Solution:
Rather than using a plug-and-play tool, we customized Python libraries and integrated them with PySpark to handle complex computations on large datasets. We also cleansed the data thoroughly—removing nulls, correcting invalid item codes, and excluding expired or unavailable products—ensuring only valid and actionable records entered the model.
Solution
The final solution delivered by Logesys included:
- Data processed using PySpark on Azure Databricks for high-speed handling
- A machine learning model using the Apriori algorithm to determine frequently co-purchased items
- Recommendations tuned to exclude bundles that don’t add incremental value (e.g., high performers that sell well independently)
- Fresh insights every cycle based on the last 6 months of sales data
Dashboard & Reporting
A dynamic Power BI dashboard was developed to visualize bundling suggestions, confidence levels, and promotional priorities. The dashboard allowed the marketing team to:
- Identify high-confidence product bundles for in-store and online promotions
- Filter suggestions by product category, frequency of sale, and sales region
- View historical bundle performance versus predicted outcomes
It’s fully responsive and accessible across laptops, tablets, and iPads, making it easy to use for marketing managers on the go.
Results
Logesys delivered the project in just two months, with dashboard rollouts completed in three phases. The solution has enabled the client to:
- Launch highly targeted bundle promotions with better returns
- Improve revenue by bundling underperforming products with popular ones
- Reduce manual guesswork and align promotions with actual customer behavior
- Facilitate faster stock clearance through bundled discounts
- Create separate promotional strategies for online and in-store campaigns
An especially unique outcome was the non-intrusive strategy—the model focused on bundle opportunities only where it made business sense. High-selling products were kept unbundled when bundling would offer no additional gain.
Conclusion
Through this project, Logesys helped a leading pharma and wellness retailer move from instinct-led marketing to data-driven promotional campaigns. By combining machine learning with enterprise-scale data processing, we empower their team to create bundles that convert—without wasting budget or inventory.
As a result, their bundling campaigns are now smarter, faster, and significantly more profitable. And with a system that evolves over time, the client is well-positioned to continue driving growth through strategic product insights.