The CLV Paradox: Why High-Value Customers Are Not Always the Best Retention Targets
June 2026 | Marketplace Economics | Customer Lifetime Value | Empirical Research Note | ← Back to Blog
This analysis draws from a working paper on profit-aware pricing in two-sided marketplaces, examining customer lifetime value, repeat purchase behavior, and retention investment allocation across 110,840 transactions from the Olist Brazilian e-commerce marketplace.
The Finding
High-value customers are not always the best retention targets.
In the marketplace I analyzed, I segmented customers by their predicted repeat purchase probability using out-of-sample propensity scores. The pattern that emerged was counterintuitive.
The bottom 50% of customers by repeat propensity spent BRL 178 on their first purchase, compared with BRL 106 for the top 20%. Yet the low-propensity segment returned at only 2.45%, versus 4.40% for the high-propensity segment.
"Retention budgets should follow future value, not past revenue. The highest-spending customer is not always the best customer to retain."
The CLV Paradox
The low-propensity segment contributed 64% of total estimated customer lifetime revenue (BRL 8.43M of BRL 13.13M), but most of that value came from the first transaction rather than future purchases.
This creates a paradox for standard CLV-based retention frameworks. If you rank customers by lifetime value and allocate retention budget accordingly, you would invest most heavily in the segment least likely to return. The CLV metric is not wrong; it accurately reflects the revenue generated. But it conflates one-time purchase value with repeat purchase potential, which are different problems requiring different strategies.
This distinction matters because many retention programs are still allocated using historical revenue rather than expected future behavior.
Why the High-Spending Customers Did Not Return
These customers were not disengaged. They were buying products tied to one-time occasions: wedding gifts, single furniture pieces, and electronics upgrades. The barrier to retention was structural, not behavioral.
The behavioral profile confirms this interpretation. Low-propensity customers had the highest average order values at BRL 178, but also the highest freight costs at BRL 23.86, the lowest satisfaction scores at 3.92, and the lowest items per order at 1.03. They were making single large purchases from remote geographies, often in categories with no natural repeat purchase occasion.
Retention marketing is unlikely to overcome a structural retention barrier. If the purchase occasion does not recur, as with a wedding gift, a bedroom set, or a laptop upgrade, no loyalty program or follow-up email changes that. The customer was never going to return at a rate that justifies retention investment.
Why the Lower-Spending Customers Were the Right Retention Targets
The high-propensity segment spent less per order at BRL 106, but showed a very different behavioral profile. Items per order averaged 1.39 versus 1.03 for the low-propensity segment. Freight costs were lower at BRL 16.84. Satisfaction scores were higher at 4.28. And actual repeat rates matched predicted rates almost exactly (4.40% actual versus 4.49% predicted), confirming strong model calibration.
These customers were buying recurring household products in multi-item orders with positive platform experiences. Their repeat behavior was structurally sustainable. Retention investment in this segment, including loyalty programs, free shipping incentives, and sequential recommendations timed to the 29-day median return window, had positive expected ROI. In the low-propensity segment it did not.
The Segmentation That Actually Matters
The analysis suggests that the segmentation variable that matters for retention investment is repeat purchase propensity, not CLV rank or first purchase value. These two dimensions are not the same and in this marketplace they pointed in opposite directions.
A customer with BRL 178 average first purchase and 2.45% repeat rate is a profitable acquisition. They should be served well on the first transaction. But they are not a retention investment target. A customer with BRL 106 average first purchase and 4.40% repeat rate is both an acquisition and a retention target. The lifetime value program belongs here, not in the high-spend low-propensity segment.
This distinction has direct implications for how platforms design loyalty programs, allocate promotional budgets, and sequence post-purchase communications. Blanket retention programs applied across all customer segments waste budget on customers who were never going to return regardless of the incentive offered.
A Note on Measurement
The propensity scores were estimated out-of-sample to avoid overfitting. Actual and predicted repeat rates aligned closely across segments, suggesting the model captured meaningful behavioral differences rather than statistical noise.
Conclusion
The CLV paradox in this marketplace is not a data artifact. It reflects a genuine structural feature of e-commerce markets with heterogeneous purchase occasions. High first-purchase value and high repeat propensity are different customer attributes that happen to be negatively correlated in this setting.
The practical implication is straightforward. Retention budgets should follow future value, not past revenue. The BRL 178 customer is a profitable one-time transaction and should be treated as such. The BRL 106 customer is the foundation for lifetime value programs.
The highest-spending customer is not always the best customer to retain. Sometimes the most valuable customer is the one who spends less today but has a reason to return tomorrow.
This analysis is part of a broader working paper on profit-aware pricing in two-sided marketplaces, examining demand elasticity estimation, profit optimization under cost uncertainty, customer lifetime value modeling, and implementation frameworks across 110,840 transactions from the Olist Brazilian marketplace. The full paper is available on my research page.
