> ## Documentation Index
> Fetch the complete documentation index at: https://codking-blogs.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# COD Fraud and Repeat Refusers: Building a Blocklist You Can Defend

> Shopify's fraud analysis never looks at a COD order. What to key a blocklist on, why phone numbers alone fail, and how to graduate treatment instead of banning people.

[← All articles](https://codking-blogs.mintlify.site/blog.md)

COD & RTO · November 2026

# COD Fraud and Repeat Refusers: Building a Blocklist You Can Defend

12 minutes read

COD & RTO

[Apparel Sizing & Returns](https://codking-blogs.mintlify.site/blog/apparel-sizing-cod-returns.md)
[Fraud & Repeat Refusers](https://codking-blogs.mintlify.site/blog/cod-fraud-repeat-offenders.md)
[Festive COD Peak](https://codking-blogs.mintlify.site/blog/festive-season-cod-operations.md)
[Cost of a COD Return](https://codking-blogs.mintlify.site/blog/cod-return-cost-reverse-logistics.md)
[Dispatch Speed](https://codking-blogs.mintlify.site/blog/dispatch-speed-cod-orders.md)
[Pincode Serviceability](https://codking-blogs.mintlify.site/blog/cod-pincode-serviceability.md)
[COD Risk Scoring](https://codking-blogs.mintlify.site/blog/cod-risk-scoring.md)
[Address Quality](https://codking-blogs.mintlify.site/blog/address-quality-cod-orders.md)
[Courier Allocation](https://codking-blogs.mintlify.site/blog/courier-allocation-cod-orders.md)

Payments & Pricing

[Prepaid Incentive Design](https://codking-blogs.mintlify.site/blog/prepaid-incentive-design.md)
[Weight Disputes](https://codking-blogs.mintlify.site/blog/courier-weight-disputes.md)
[COD to Prepaid](https://codking-blogs.mintlify.site/blog/cod-to-prepaid-conversion.md)
[COD Advance Amount](https://codking-blogs.mintlify.site/blog/partial-cod-payment-advance-amount.md)
[Payment Options](https://codking-blogs.mintlify.site/blog/payment-options-indian-ecommerce.md)
[COD Remittance Cycle](https://codking-blogs.mintlify.site/blog/cod-remittance-cycle-cash-flow.md)

Tooling & Checkout

[COD King vs Releasit](https://codking-blogs.mintlify.site/blog/cod-king-vs-releasit.md)
[COD King vs HillTeck](https://codking-blogs.mintlify.site/blog/cod-king-vs-hillteck.md)
[In-House NDR vs 3PL](https://codking-blogs.mintlify.site/blog/in-house-ndr-vs-3pl.md)
[Choosing a COD App](https://codking-blogs.mintlify.site/blog/best-cod-verification-apps-shopify.md)
[Checkout vs COD Form](https://codking-blogs.mintlify.site/blog/native-checkout-vs-cod-order-form.md)
[WhatsApp vs SMS vs IVR](https://codking-blogs.mintlify.site/blog/whatsapp-vs-sms-vs-ivr-cod-verification.md)
[COD Cancellations](https://codking-blogs.mintlify.site/blog/how-to-reduce-cod-cancellations.md)
[Checkout Optimization](https://codking-blogs.mintlify.site/blog/shopify-checkout-optimization.md)

A repeat-refuser pipeline: four match keys of increasing strength feed a ledger, which routes an order to one of four graduated treatments rather than a single blockSummaryShopify's own fraud analysis is built around card orders, so a COD order arrives with no risk recommendation at all — whatever you do here, you build. This covers the four distinct behaviours that get lumped together as "COD fraud", what a blocklist should actually be keyed on, why a phone-number list decays within weeks, and why the most useful entry in the list is "prepaid only" rather than "banned".

Every COD-heavy brand eventually discovers the same thing: a small number of phone numbers are responsible for a disproportionate share of refused parcels. The instinct is to start a list. The instinct is right, and the first version of the list is almost always built in a way that ages badly — keyed on the wrong field, populated from the wrong signal, and enforcing a single punishment that is simultaneously too harsh for honest customers and too weak for anyone deliberately gaming you.

## The platform does not do this for you

It is worth being clear about where the gap is, because merchants often assume some layer of the stack is already watching. Shopify's documentation states that *fraud analysis works with online credit card orders that Shopify can verify*, and that the risk recommendation is produced *for eligible online credit card orders*, indicating *whether an order has a low, medium, or high risk of a chargeback due to fraud*. The page lists order types excluded from a recommendation — test orders, free orders, orders paid in full with a gift card, Point of Sale orders, B2B orders and subscription renewals — and it does not address cash on delivery or manual payment methods at all. The thing being predicted is a chargeback, which is a card phenomenon.

That matters in two ways. First, your COD orders are the ones with real downside exposure and they are the ones arriving without a score. Second, the controls Shopify documents are not merchant-set: it describes Shopify Payments blocking *some high-risk checkout attempts before an order is created*, and temporary IP-level blocks that are *removed automatically, usually within a day*, with *no way to remove it sooner*. There is no documented merchant control for blocking a specific customer. Which means the repeat-refuser problem is entirely yours to build, and the design decisions below are not optional details — they are the whole system.

## Four behaviours wearing one label

"COD fraud" is used to describe at least four different things, and they do not respond to the same treatment. Separating them is the first real step, because a control that is correct for one is actively wrong for another.

**The casual over-orderer.** Someone who placed three orders across two weeks, meant one of them, and let the others come back. No malice, no pattern worth punishing — just a shopping habit that COD makes free. This is the largest group by volume and the one most damaged by a hard block.

**The serial refuser.** A number with a sustained history of ordering and refusing, often across categories, often at similar basket values. The behaviour is consistent enough to be predictive, and this is the group a list is genuinely for.

**The decoy order.** Fabricated or deliberately mistyped contact details, usually placed to disrupt rather than to receive — competitor mischief, or someone testing whether your store will ship anything. These rarely repeat on the same key, which is exactly why a blocklist is the wrong tool and pre-dispatch verification is the right one.

**Delivery-stage abuse.** Opening the parcel at the door and refusing, claiming non-receipt on a delivered shipment, or a swap at the point of handover. This is a courier-process problem — proof of delivery, sealed packaging, open-delivery policy — and no amount of order-time scoring addresses it.

Only the second group belongs in a blocklist. The first needs a nudge, the third needs verification, the fourth needs a different conversation with your courier. A list that mixes all four will be mostly populated by group one, because group one is the most numerous, which is how brands end up blocking their own customers at scale.

## What the list is keyed on decides whether it works

The most common build is a list of blocked phone numbers. It is the obvious key and it decays faster than any other.

A phone number is cheap to change and in India a household often has several. A serial refuser who hits a block simply orders on a different number from the same address, and your list has recorded a fact about a SIM rather than about a person. Meanwhile the collateral damage runs the other way: a number that was recycled by the operator, or belongs to a family member who shares the delivery address, now carries someone else's history.

Address alone fails symmetrically. Shared buildings, PG accommodation, hostels, offices and apartment blocks mean one address legitimately serves many unrelated buyers. Blocking an address in a large residential complex is a blunt instrument pointed at dozens of people.

The pair is what works. A phone number together with a normalised address is specific enough to describe an actual buying pattern and robust enough to survive a SIM change, because the address half persists. Treat a match on either half alone as a signal rather than a verdict: same address, new number is worth a confirmation step; same number, new address is worth noting but is also what happens when someone moves house.

Normalisation is the unglamorous part that determines whether any of this functions. Indian addresses as typed by customers vary enormously for the same physical location — abbreviations, transliteration differences, flat numbers in a different field, pincode in the street line. If you compare raw strings you will match almost nothing. At minimum, strip punctuation and case, collapse whitespace, pull the pincode out as its own field, and normalise the obvious variants in your own data before you compare. Phone numbers need the same discipline: strip the country code, leading zeros and spaces, and store one canonical form.

Email is a weak key for COD — frequently absent, frequently throwaway. Device and session signals are stronger but belong to your checkout or app layer, and they are also the ones most likely to collide in shared-device households.

## Build the ledger before you build the list

A blocklist is a conclusion. The thing worth building first is the record the conclusion is drawn from — a per-key ledger of outcomes, not opinions.

For each normalised key, keep the count of orders placed, the count delivered and paid, the count refused at the door, the count returned undelivered for other reasons, the dates, and the basket values. That is it. The discipline is in what you leave out: no free-text notes about customers, no judgements recorded by whoever handled the call, nothing that cannot be traced back to a delivery outcome in your courier data.

Two reasons for that strictness. The practical one is that outcome counts are the only thing you can set a threshold on. The other is that this is a record about identifiable people, and a ledger of facts you can point at is defensible in a way that a collection of staff impressions is not.

The ledger also makes the distinction the four-behaviours section asks for. A key with two refusals and four paid deliveries is a customer with a habit. A key with six refusals and no deliveries is something else. You cannot tell those apart from a count of refusals alone, which is why the paid-delivery count belongs in the record and is the field most often missing.

One more field earns its place: whether the refusal reason came from the customer or the courier. A parcel marked undelivered because the rider could not find the address is not a refusal, and pooling those two together will fill your list with people who live somewhere hard to find.

## Four treatments, not one

The single biggest improvement available here is replacing the binary with a ladder. A block is the most expensive response you have — it forgoes all future revenue from that person to avoid one return — and it is almost never the right first move.

**Tier one: require confirmation.** One prior refusal routes the order into verification rather than straight to pack. Cost is a message; downside is a short delay. This absorbs most of the casual over-orderers, because a person who meant the order will confirm it.

**Tier two: require a partial advance.** Two or more refusals, or one refusal at high basket value, and the order needs some money upfront before it ships. This is the first tier with real teeth and it still lets the customer buy. A buyer who pays part of the value has demonstrated commitment in a way no confirmation message can.

**Tier three: prepaid only.** COD is withdrawn for that key; the store remains fully open. This is where most of the value in the whole system sits, and it is the tier brands skip. You stop funding the forward leg for someone whose history says they will not pay for it, you keep the customer relationship, and you keep the revenue if they still want the product. It is also the easiest tier to defend if the customer asks, because nothing has been taken away except a payment method.

**Tier four: block.** Reserved for keys where the pattern is unambiguous and the volume is material. Rare, reviewed by a person, and reversible.

Framed this way, the question stops being "who do we ban" and becomes "at what history does COD stop being something we extend to this key". That is a credit decision, and treating it as one is what makes it tractable.

## Setting the thresholds on your own numbers

There is no correct refusal count at which to escalate, because the right threshold depends on what a return costs you and what a customer is worth — both of which are yours specifically.

The arithmetic to run is simple in shape. Work out your actual all-in cost of one refused COD parcel: forward freight, return freight, the COD handling fee, packaging, the labour in handling the NDR and the return, and the inventory cost of whatever comes back unsellable. Then work out the contribution margin on one delivered order. The ratio between those two numbers tells you how many good orders you can afford to lose from a key before the control costs more than the returns it prevents.

That ratio is what should set the tier boundaries, and it moves by category. A brand with thin margins and expensive reverse logistics should escalate on a single refusal. A brand with healthy margins and cheap parcels can absorb two or three before restricting anything, and will make more money being patient. Copying another brand's thresholds is copying their cost structure.

Two adjustments worth making. Weight recent behaviour more heavily than old behaviour — a refusal from fourteen months ago says much less than one from last month, and a ledger with no decay slowly blocks your entire customer base. And scale the threshold with basket value, because the expected loss on a high-value order justifies friction that would be wasteful on a small one.

## The cost of being wrong is asymmetric and invisible

A blocklist produces two kinds of error, and only one of them shows up in your reporting.

When you ship to someone you should have restricted, you get a return, a cost line and a clear record of the mistake. When you restrict someone you should have shipped to, you get silence. There is no row in any report for the order that was never placed, and the customer does not write in to explain that they have gone elsewhere. So the feedback loop runs in one direction only, and a list left to drift will become steadily more aggressive — every error it can see argues for tightening, and the errors arguing for loosening are structurally unobservable.

Three habits counter it. Hold a small holdout: a random share of keys that would have been restricted get shipped anyway, so you can see what the restriction was actually preventing. Review the tier-three and tier-four population on a schedule rather than only on complaint, and check what share of it is single-incident. And keep the escape hatch obvious — a restricted customer who pays prepaid and receives the order should earn their way back down the ladder automatically, because a list that only ratchets upward is a list that eventually contains everyone who ever had a bad week.

## Keeping the list defensible

This is a file of phone numbers, addresses and behavioural judgements about named individuals, which makes it one of the more sensitive things your operation holds. A few practices keep it both useful and defensible, and they are worth adopting before anyone asks you about them.

Record outcomes rather than characterisations — "three refused deliveries" rather than any description of the person. Keep it to the fields the decision actually needs. Give entries an expiry so the record does not outlive its usefulness. Keep it inside your own systems rather than a spreadsheet on a laptop, and limit who can read it. Log who added or removed an entry, which matters more for your own debugging than for compliance. And be careful about shared or industry-wide blocklists: a key added by someone else, on evidence you cannot inspect, is a decision about your customer made by a party with no relationship to them.

India's Digital Personal Data Protection Act, 2023 governs the handling of digital personal data and is worth reading properly with your own advisers rather than taking a blog's summary of it. The operational point stands regardless of how the rules are enforced: minimal fields, stated purpose, bounded retention and controlled access are the version of this that will not embarrass you.

## What to track

Four numbers tell you whether the system is working. The share of your COD volume that matches a ledger key with prior refusals, which tells you how concentrated the problem actually is — in most catalogues it is smaller than the team believes and more concentrated than the reporting shows. The RTO rate of restricted keys compared with your holdout, which is the only honest measure of whether the restriction prevented anything. The conversion rate of tier-three keys to prepaid orders, which tells you whether "prepaid only" is retaining revenue or quietly functioning as a block. And the number of keys that have earned their way back down a tier, which is your false-positive correction rate — if it is zero, your ladder has no way down and is not a ladder.

The second of those is the one that justifies the whole exercise, and it needs the holdout to exist. Without it, you are comparing restricted keys against unrestricted ones and attributing the difference to the restriction, when the difference is mostly why you restricted them in the first place.

## The short version

Shopify's fraud analysis is a card-chargeback tool and does not produce a recommendation for a COD order, so this layer is entirely yours to build. Separate the four behaviours before building anything, because only the serial refuser belongs in a list and the casual over-orderer will dominate it if you let them. Key on a normalised phone-and-address pair rather than a phone number, and spend the effort on normalisation, because without it nothing matches. Build the outcome ledger before the list, and keep paid deliveries in it alongside refusals. Replace the ban with a four-tier ladder and put most of your weight on prepaid-only, which protects you without closing the store. Set thresholds from your own return cost and contribution margin, decay old history, and scale with basket value. Then hold a holdout, because the errors that argue for loosening the list are the only ones you will never see on their own.

Where COD King fits: the ledger lives in your order data, and the enforcement happens at order time — COD rules by pincode and order value, confirmation before dispatch, and partial advance collection are the mechanisms tiers one to three are actually built from. The thresholds, the review cadence and the decision about who comes off the list stay yours, and they should.

Quoted statements about Shopify fraud analysis are from Shopify's [Reviewing orders with fraud analysis](https://help.shopify.com/manual/orders/fraud-analysis) help page, checked 19 November 2026. Platform behaviour changes; verify against the current page. India's Digital Personal Data Protection Act, 2023 is referenced only as context, not as legal advice; check the current text and rules with your own advisers.

On this page[The platform blind spot](#blind-spot)[Four behaviours](#four-behaviours)[What to key on](#keys)[Build the ledger first](#ledger)[Four treatments](#ladder)[Setting thresholds](#thresholds)[The cost of being wrong](#false-positives)[Keeping it defensible](#data)[What to track](#measure)[Start for free](https://codking.tech)[Contact sales](https://codking.tech)

COD KingPractical guides, strategies, and product insights for Shopify merchants managing Cash on Delivery.

Blog
[All articles](https://codking-blogs.mintlify.site/blog.md)
[Understanding COD Verification](https://codking-blogs.mintlify.site/blog/understanding-cod-verification.md)
[How to Reduce RTO for COD Orders](https://codking-blogs.mintlify.site/blog/reduce-rto-for-cod-orders.md)
[Apparel Sizing and COD Returns](https://codking-blogs.mintlify.site/blog/apparel-sizing-cod-returns.md)
[COD Fraud and Repeat Refusers](https://codking-blogs.mintlify.site/blog/cod-fraud-repeat-offenders.md)
[NDR Management](https://codking-blogs.mintlify.site/blog/ndr-management-cod-orders.md)
[Courier Allocation](https://codking-blogs.mintlify.site/blog/courier-allocation-cod-orders.md)
[Address Quality for COD Orders](https://codking-blogs.mintlify.site/blog/address-quality-cod-orders.md)
[COD Risk Scoring](https://codking-blogs.mintlify.site/blog/cod-risk-scoring.md)
[COD Pincode Serviceability](https://codking-blogs.mintlify.site/blog/cod-pincode-serviceability.md)
[Dispatch Speed for COD Orders](https://codking-blogs.mintlify.site/blog/dispatch-speed-cod-orders.md)
[What a COD Return Costs](https://codking-blogs.mintlify.site/blog/cod-return-cost-reverse-logistics.md)
[Running COD Through a Festive Peak](https://codking-blogs.mintlify.site/blog/festive-season-cod-operations.md)
[RTO Benchmarks for Indian D2C](https://codking-blogs.mintlify.site/blog/rto-benchmarks-india.md)
[COD to Prepaid Conversion](https://codking-blogs.mintlify.site/blog/cod-to-prepaid-conversion.md)
[Designing a Prepaid Incentive](https://codking-blogs.mintlify.site/blog/prepaid-incentive-design.md)
[How Much COD Advance to Collect](https://codking-blogs.mintlify.site/blog/partial-cod-payment-advance-amount.md)
[Courier Weight Disputes](https://codking-blogs.mintlify.site/blog/courier-weight-disputes.md)
[Best COD Verification Apps](https://codking-blogs.mintlify.site/blog/best-cod-verification-apps-shopify.md)
[In-House NDR vs 3PL](https://codking-blogs.mintlify.site/blog/in-house-ndr-vs-3pl.md)
[Native Checkout vs COD Form](https://codking-blogs.mintlify.site/blog/native-checkout-vs-cod-order-form.md)
[COD King vs Releasit](https://codking-blogs.mintlify.site/blog/cod-king-vs-releasit.md)
[COD King vs HillTeck](https://codking-blogs.mintlify.site/blog/cod-king-vs-hillteck.md)
[WhatsApp vs SMS vs IVR](https://codking-blogs.mintlify.site/blog/whatsapp-vs-sms-vs-ivr-cod-verification.md)
[How to Reduce COD Cancellations](https://codking-blogs.mintlify.site/blog/how-to-reduce-cod-cancellations.md)
[Checkout Optimization for Shopify Stores](https://codking-blogs.mintlify.site/blog/shopify-checkout-optimization.md)
[Payment Options for Indian E-commerce](https://codking-blogs.mintlify.site/blog/payment-options-indian-ecommerce.md)
[COD Remittance Cycles and Cash Flow](https://codking-blogs.mintlify.site/blog/cod-remittance-cycle-cash-flow.md)

COD King
[Shopify App Store](https://apps.shopify.com/cash-on-delivery-cod-order-confirmation)

© 2026 COD King


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.