← all essays
D2C · India· 7 min read

The 26% problem: how fake COD orders quietly eat D2C margins

Nearly a quarter of cash-on-delivery orders in Indian e-commerce never get paid for. The patterns behind them are detectable at checkout — for less than the cost of the parcel.

Every D2C founder in India knows the number, but few put it on a P&L line: roughly 26% of cash-on-delivery orders are returned to origin. Industry analyses put typical COD RTO between 20% and 35%, and the share of orders that are outright fraudulent at 4–5% of all orders — about 15% of returns. Blended together, that is close to 7% of annual revenue evaporating in reverse logistics.

It is not an ops problem. It is a data problem that ops inherits. The order was fake from the moment it was placed — and the signals were sitting in two fields of the checkout form.

What a fake COD order actually costs

Take a brand doing ₹50 lakh a month with a 25% COD RTO rate. Published industry math puts the annual loss at roughly ₹45.6 lakh — about 7.6% of revenue. Here is where it goes, per order:

  • Outbound and return shipping — ₹120–180, paid twice to the same courier
  • Pick, pack and restock labour that never produced a sale
  • Inventory blocked in transit while real demand goes unfulfilled
  • Manual confirmation calls — one founder reports three people covering calls and still missing 40%
  • Ad spend attributed to a conversion that never existed, quietly poisoning your ROAS numbers
A phone number that doesn’t exist is not a customer. It is a cost centre with a delivery address.

The patterns give it away

Fake COD orders rarely look random at the field level. The same tells repeat order after order:

  • Phone numbers that fail basic validity — impossible prefixes, VOIP-only lines, numbers formatted for a different country than the address
  • IPs from datacenter and proxy ranges — a “customer” ordering from a cloud server
  • Gibberish names and emails — bot-generated strings that pass a form but not a human glance
  • Order sprees from one IP — the same address block placing five COD orders in an hour
  • New-device, high-value, COD-only behaviour — fraud optimises for the path with zero payment friction

The two-check fix at checkout

You do not need a fraud team to catch most of this. Two checks, run against the phone number and the IP at checkout, cover the dominant patterns: is this phone number a real, reachable line, and is this IP a residential connection in the same geography as the delivery address?

The output is a confidence score, and the routing is simple: block scores above ~0.70, step up the 0.30–0.70 band with an OTP confirmation call or WhatsApp verification, and pass everything below 0.30 untouched. Real customers feel nothing. Fake orders never ship.

Why confirmation calls don’t scale

Manual confirmation is the industry default — and it is the wrong default. It treats every order as equally suspicious, so your team burns hours calling genuine customers (who increasingly don’t pick up unknown numbers), while fake orders slip through the fatigue. One operations lead told us their three-person call team still misses 40% of unconfirmed orders.

Humans are excellent at judging the grey zone. They are terrible at being a filter for the whole queue. Score first, call only the uncertain 10–15%, and the same team clears the real risk with time to spare.