How increasing customer retention improved sales to existing clients?

Business Challenge

Company X, a medium-sized e-commerce retailer specializing in premium home and interior products across Poland and internationally, was facing two critical issues: a rising Customer Acquisition Cost (CAC) and declining effectiveness of customer retention campaigns.

The problem was complex due to data fragmentation: transactional data resided in the payment system, behavioral logs were scattered across external analytics tools, while contact information and sales history were held within the CRM. This lack of a unified customer view made it impossible to accurately model Customer Lifetime Value (CLV) or predict how much future revenue could be generated from an individual client. Consequently, the marketing team was operating on incomplete data, often targeting low-potential customer segments.

Company X needed a robust tool for proactive revenue management, moving beyond simply reacting to the last purchase. They required full visibility into which customers were most valuable and when they were at highest risk of churning.

Business Goal

The primary goal was to design and implement a centralized data management system that would integrate all information sources—transactions, behaviors, and contacts—into a single location, establishing a "single source of truth." This system needed to enable the creation of a precise predictive model capable of determining:

• Customer Revenue/Profit Potential - estimating the future purchasing potential of each client.
• Churn Risk - identifying when a customer is most likely to discontinue their relationship with the company.

This model was intended to become the foundation for all marketing and sales activities, shifting the business from reactive marketing to proactive revenue management. The ultimate measure of success was the increase in Customer Lifetime Value (CLV).

Implementation

This was one of my largest and longest-running projects because it required building a data pipeline infrastructure from scratch. The implementation was divided into three distinct phases.

Step 1: Data Audit and Business Process Definition

The initial audit assessed exactly what data Company X was collecting, how they were gathering it, and where the gaps existed. Simultaneously, we defined key business metrics across all stages of the client's operational processes. We also developed a set of internal quality metrics to monitor the integrity of the collected data throughout the project lifecycle.

Step 2: Centralized Data Repository Implementation

Next, we focused on centralizing the existing data. Instead of relying on manual data merging, I designed an automated data processing pipeline using the ELT (Extract, Load, Transform) process. This involved loading raw data from the data warehouse into a cloud-based data lake provider, followed by transformation within the computing cloud environment. After rigorous testing and resolving several edge cases with the internal Company X team, we ensured the system's continuous operation and reliability.

Step 3: Predictive Modeling

Based on the unified dataset, I built an initial predictive model that automatically calculated each customer’s potential revenue value and their associated churn risk. Key variables included country of residence, days since last website visit, and average basket size over the past 90 days.

However, this first iteration was insufficient for Company X. Firstly, the model ignored costs (such as advertising expenses), resulting in only an 81% accuracy rate for revenue potential and a mere 53% accuracy for churn risk. Secondly, gross revenue alone does not indicate true profitability.

In the second iteration, I successfully refined the model to predict customer profit potential, incorporating all statistically significant variable costs. This significantly boosted the model's performance: accuracy rose to 94% for profit prediction and 89% for churn risk.

The predictive results allowed us to develop highly granular customer segments. These insights were then directly applied to marketing strategies, primarily focusing on retention efforts.

Results and Recommendations

The entire system implementation was successfully completed. Subsequent analysis revealed that the primary factor contributing to the decline in CLV was the neglect of a specific segment: high-potential customers who exhibited low initial churn risk. This critical pattern would have been impossible to detect without the developed predictive model.

Key Analytical Findings:

• 18% Increase in CLV: This growth was achieved through personalized campaigns (e.g., emails recommending complementary products based on past purchases), which successfully activated this high-potential group and directly increased the average transaction value.
• Reduced Churn Risk - we identified that the critical window for customer risk occurred 45–60 days after a major initial purchase. Implementing an automated communication sequence providing usage tips during this period significantly reduced the number of customers who abandoned their relationship without cause.

Recommendation for Your Business

If you spend too much time on manual data manipulation, or if your business decisions are based on fragmented views of your customer base, it indicates a fundamental problem with your system architecture. You need one centralized source of truth. Contact me to explore how I can transform your scattered data into measurable sales growth for your business.