COD Fraud and Fake Orders: Detection That Does Not Kill Real Sales

Fake COD orders are real, and so is the damage from over-reacting to them. A seller who blocks every order that looks slightly odd loses more real sales than fraud ever cost. This is a risk-scoring approach that catches the orders that were never going to be accepted, while letting a first-time buyer in a Tier 3 town with a landmark address through.

What “COD fraud” actually is

Mostly it is not fraud in the criminal sense. It is orders placed with no intention of accepting them: pranks, competitors, buyers testing whether the site is real, duplicates placed twice to see which arrives first, and impulse orders the buyer forgot about. The effect is the same as fraud, an RTO with all its cost, and the detection is the same: signals that the buyer has no stake in the order.

Genuine fraud (organised fake orders to drain a seller’s freight budget, or address rings) exists but is a small fraction. Design for the large fraction.

The signals, and their weight

Each signal on its own is weak. Together they are predictive. A simple score:

Signal	Points	Why
Phone number has refused a COD parcel before	40	The strongest single predictor
Same phone or address ordered in the last 24 hours	25	Duplicate or test
Address under 25 characters or no house number	15	Undeliverable
Pincode with RTO over 35 percent in your own history	15	Lane risk
Order value in your top 10 percent	10	Higher-value COD refuses more
Email is disposable or a random string	10	No stake
Phone does not match the pincode’s circle	5	Weak, but adds up
Placed between 1 am and 5 am	5	Impulse hours
First-time buyer	5	Not a red flag alone

Score under 30: ship. 30 to 60: confirm before shipping (a WhatsApp with one-tap confirm). Over 60: prepaid only, with a polite message. Tune the thresholds on your own data after a month.

The false-positive problem

Every rule blocks some real orders. A rule that blocks 100 orders of which 60 would have been RTOs and 40 would have delivered is a bad rule if your margin on a delivered order exceeds 1.5 times the cost of an RTO. Most sellers’ margin does exceed that. So the bar for auto-blocking is high, and the bar for “confirm first” is low.

That is why the middle band exists. Confirmation costs a rupee and loses almost no genuine buyers; blocking costs the sale. Route most of the risk to confirmation, and reserve blocking for the top of the score.

Where the signals come from

• Your own order history for phone, address and pincode behaviour. This is the most valuable data you have and it improves monthly.
• The RTO report by pincode from your shipping platform, which shows RTO rates by lane across your volume.
• Checkout validation for address quality, phone format and email domain.
• Shared lists from some checkout providers, which flag phone numbers with refusal history across many stores. Useful for first-time buyers you have no history on.

What to do with a flagged order

Not silence. A message: “Thanks for your order. To ship COD we need a quick confirmation; tap to confirm, or pay online now for a 5 percent discount.” The genuine buyer taps. The test order does not. The prank order does not. The real fraud does not. Twenty-four hours later, the unconfirmed order is cancelled with a note.

Measuring the rule set

Every month:

• Orders by band (ship, confirm, block) and the RTO rate in each.
• Confirmation rate in the middle band and RTO among confirmed orders.
• Estimated real orders lost in the block band (sample: message them; some will convert to prepaid and prove they were real).

If the “ship” band’s RTO is over 12 percent, the thresholds are too loose. If the “block” band converts more than 20 percent to prepaid when messaged, it is too tight.

Frequently asked questions

Is there a list of known fake-order phone numbers?

Some checkout and fraud tools share refusal history across stores. Your own history is more reliable for repeat buyers; the shared list helps with first-timers.

Should I block orders from high-RTO pincodes entirely?

Rarely. Confirm them, or offer prepaid only above a value. A pincode with 40 percent RTO still has 60 percent of buyers who accept.

Do fake orders cost anything if I catch them before shipping?

Almost nothing: a message and a cancelled order. Every rupee of fraud cost comes from shipping the parcel, which is why the score runs before dispatch.

Can the shipping platform score orders for me?

Courier Uncle gives you the lane-level RTO history and the per-buyer shipment history to build the score from; the scoring itself lives at your checkout or order management, where the order can be held.

What about address rings and organised fraud?

They show up as many orders to slightly varied addresses in one area, often with sequential phone numbers. The duplicate and address-quality signals catch most of it; a weekly look at orders clustered by pincode catches the rest.

Does confirmation reduce conversion?

A single WhatsApp with a one-tap confirm loses very few real buyers. Repeated calls do. Keep it to one message and one reminder.

Ship smarter with Courier Uncle

RTO by pincode and by buyer history on the dashboard, so your risk score runs on your own numbers. Start free.