Every risk team running a COD-heavy e-commerce platform has a version of the same rule. Place too many orders that get rejected at the doorstep and the platform restricts your ability to place future COD orders. The logic is clean. The execution is fast. And for a meaningful segment of the buyers it catches, it is exactly the wrong intervention.
This is not an argument against COD block rules. They work. The data is clear that a subset of buyers systematically reject packages with no intention of ever accepting a delivery, and blocking them reduces failed delivery costs measurably. The problem is not the rule itself. The problem is that the rule cannot distinguish between a fraudulent actor and a confused first-time buyer, and it treats both identically.
The new buyer problem
Across four South Asian e-commerce markets, a consistent pattern emerges when you look at the buyer composition behind COD block triggers. A significant share of the buyers being flagged are new to the platform. Many are placing their first or second order. Many have never successfully completed a delivery.
For a portion of these buyers the failure is not intentional. South Asian e-commerce markets have meaningful frictions that Western risk frameworks do not account for. Many buyers genuinely do not know how to cancel an order they no longer want. The cancellation flow requires navigating an app interface that is not always intuitive, particularly for first-time users in markets where smartphone e-commerce adoption is still maturing. The path of least resistance is not cancelling. It is simply saying no when the rider arrives.
Others placed an order impulsively and changed their mind but assumed the order would be automatically cancelled if they ignored it. Others did not realise the order had gone through. Others had a family member place an order on their phone without their knowledge.
None of these are fraudulent behaviors. All of them produce a failed delivery. And under a standard COD block rule, all of them result in the same outcome: the buyer is restricted after a small number of failures, frequently before they have had a single successful transaction.
The platform pays twice
When a new buyer is blocked after two or three failed deliveries the platform has already absorbed the logistics cost of those failures. That cost is real and it is right to want to prevent it from recurring. But the block creates a second cost that does not appear on the same dashboard.
A new buyer who is blocked before their first successful delivery is a customer the platform spent money to acquire, potentially through paid marketing, affiliate commissions, or campaign subsidies, who never generated a single unit of retained revenue. The acquisition cost is sunk. The lifetime value is zero. And in many cases the buyer never returns to the platform at all because the experience of having their payment method restricted feels punitive rather than helpful.
The risk dashboard shows the logistics cost saving from the block. It does not show the customer lifetime value that was permanently suppressed. Both are real. Only one is being measured.
What a smarter approach looks like
The question is not whether to block buyers who systematically fail deliveries. The question is whether a block is the right first intervention for every buyer who triggers the threshold.
A more precise framework segments the flagged population before applying a hard restriction. Three cohorts emerge naturally from the data:
The first cohort is buyers whose failure pattern is clearly behavioral and deliberate. High failure rates, multiple accounts, device-level evasion, no successful deliveries across an extended history. For this cohort a COD block is the correct and immediate response.
The second cohort is buyers with a short history, low failure count, and no signal of deliberate evasion. For this cohort the right intervention is not a block. It is a proactive contact before the next delivery attempt. A short call or message confirming the order, explaining the cancellation option, and setting delivery expectations. Many of these buyers will accept the delivery when someone from the platform reaches out. Those who do not can be escalated to a block at that point with much stronger confidence that the failure is intentional.
The third cohort sits in between. Some history, some failures, but also some successful deliveries. For this cohort a graduated friction approach works better than a binary block. Reducing the COD order limit, requiring a smaller initial COD value, or adding a confirmation step before COD orders are processed. These interventions reduce exposure without permanently suppressing a buyer who may still become a loyal customer.
The role this creates
This kind of segmentation cannot be owned by the risk team alone. It requires someone asking a different question than the one risk teams are trained to ask. Risk asks: how do I stop this buyer from costing the platform money today? Buyer lifecycle asks: what is this buyer worth to the platform over the next two years and what intervention maximises that value?
Those two questions need to be answered together before a block decision is made. Right now in most platforms they are answered in isolation by different teams using different data with different objectives. The result is a risk function that is technically correct and commercially suboptimal at the same time.
The platforms that close that gap, that build the connection between risk data and lifetime value data before applying interventions, will have a structural advantage in COD markets over the next decade. Not because they are more lenient with bad actors. Because they are more precise about who a bad actor actually is.