The unit economics of Cash on Delivery for Indian D2C brands

Ask ten Indian D2C founders whether they offer Cash on Delivery and nine will answer with a feeling: it converts better, or it costs too much in returns. Almost none will answer with a number. That's unusual, because COD is one of the few growth-versus-cost tradeoffs in D2C that can be modelled with a formula simple enough to fit on an invoice, using data most stores already collect and mostly ignore.

The reason it doesn't get modelled is that the two effects live in different dashboards. The conversion lift shows up in checkout analytics. The cost shows up two to four weeks later, in a courier's return-to-origin report, in a category most founders check monthly rather than per-order. By the time the RTO report lands, nobody connects it back to the checkout decision that caused it.

Return to Origin, RTO, is a shipment that a courier attempts to deliver and cannot: refused at the door, unreachable customer, wrong address, or a customer who simply changes their mind before the parcel arrives. It is the primary mechanism through which COD orders cost more than prepaid ones, because a prepaid order that fails delivery has already been paid for; a COD order that fails delivery has cost the brand its logistics fee and its packaging for zero revenue.

The published estimates vary by a wide margin, and the honest thing to do is show the range rather than pick the number that makes the best headline. GoKwik's aggregated analysis across more than 180 million Indian shoppers puts the average RTO rate at roughly 23 percent, with COD orders running in the mid-to-high 20s and prepaid orders staying under 2 percent [1]. Other industry estimates put the gap even wider: COD RTO from 25 to 40 percent against a prepaid rate that barely moves [2][3]. A separate operational dataset spanning more than 400 million order items recorded RTO falling from roughly 39 percent in November 2025 to about 21 percent in February 2026, which says as much about how much RTO is seasonal and controllable as it does about any fixed constant [4].

The range matters more than any single figure inside it: a brand that assumes 15 percent RTO because that's what a blog post said, when its own category and pincode mix actually runs closer to 30, is making a COD decision on a number that was never true for them.

Strip away the category-specific noise and the decision reduces to comparing two expected values per order, COD against prepaid, at a given RTO rate.

The reverse logistics cost is the forward shipping fee plus the return shipping fee, since a failed COD delivery usually still incurs both legs. Industry estimates put the all-in cost of a single RTO, courier fees, packaging, and the restocking or write-off of the returned unit, at roughly ₹150 to ₹300 per order depending on category and courier [1][3]. That number belongs in the formula as a real cost, not a rounding error: on a ₹999 supplement order with a 40 percent margin, a single RTO at ₹250 in reverse-logistics cost erases the entire margin on more than half an additional sale.

The number that actually matters for a decision is the break-even RTO rate: the RTO percentage at which COD's expected value drops to equal prepaid's expected value once you account for the conversion lift COD provides. Below that rate, COD is a net positive even accounting for its losses. Above it, every additional COD order offered is destroying more margin than the conversion lift it buys back.

A fit-dependent category, apparel especially, carries a structurally higher RTO rate than a consumables category like supplements or packaged food, for a reason that has nothing to do with payment method: a size or colour that arrives wrong is refused at the door regardless of whether it was already paid for, whereas a consumable that was ordered specifically is usually accepted. Fashion e-commerce RTO analysis published separately puts return-and-RTO combined loss in that category meaningfully above the cross-category average [5], which is consistent with the mechanism rather than a coincidence of the dataset.

This is the actual argument against copying a COD policy from a benchmark blog post written for a different category. A supplement brand and an apparel brand operating at the same national-average RTO figure are not facing the same problem, because the apparel brand's RTO is driven by fit uncertainty that COD availability barely affects, while the supplement brand's RTO is driven mostly by payment-method trust, which is exactly the variable a COD policy controls.

Three interventions follow directly from the model rather than from a generic best-practice list. First, compute your own break-even RTO rate using your real margin and real logistics cost, not an industry average, and compare it against your own courier's RTO report for your specific pincode mix, not a national figure. Second, where COD is still offered, a partial-prepaid model, a small deposit at checkout, moves the expected-value calculation meaningfully because it converts a fully COD order into a hybrid one, and Indian D2C operators using this pattern report RTO reductions in the same range as GoKwik's own published case data [6]. Third, segment by order value: COD's marginal cost per failed order is fixed in absolute rupees, which means it erodes a proportionally larger share of margin on a low-value order than a high-value one, so a COD cutoff above a certain cart value, rather than a blanket toggle, is very often the better-performing policy even though it looks less generous to a customer scrolling checkout.

None of this is a claim that COD should be removed. COD still accounts for roughly 60 to 70 percent of order volume across Indian e-commerce [1][2], which means removing it outright is very often a larger conversion loss than the RTO cost it saves. The point of the model is not to argue for a side, it's to replace a feeling with a number specific to your own store, so the decision can actually be defended rather than just felt.

Every RTO figure cited here is a third-party industry aggregate, not a universal constant, and industry aggregates disagree with each other by a wide enough margin that treating any single number as precise would be dishonest. RTO is also seasonal (the November-to-February swing cited above is real and large) and improves with operational fixes, address verification, NDR follow-up calling, COD order confirmation via WhatsApp, that this article does not model because they change the input rate rather than the decision formula itself. Use the formula with your own numbers; treat every cited industry figure as a starting range to replace with your own data as soon as you have thirty days of it.