Every Way to Cut RTO, Ranked by the Evidence Behind It
Search for how to reduce RTO and you will find a great deal of advice and almost no arithmetic. Verify addresses. Improve product photos. Offer prepaid discounts. Send tracking updates. Every list contains roughly the same ten items, in roughly the same order, and almost none of them tell you the one thing that decides where to start: how much each intervention is actually worth.
This piece ranks them by the evidence. Where a number exists, it is here with the source attached. Where no number exists, that is said plainly rather than dressed up, because a tactic with no published effect is a hypothesis, and knowing which of your ten tactics are hypotheses is worth more than the tenth tactic.
First, the cost you are actually carrying
Return to origin is usually priced as a courier line item, which is why it gets underestimated. The parcel travels twice, so you pay forward and reverse freight. Pathao Courier, which moves over 250,000 parcels a day, puts the round trip at roughly Tk 195 for a typical merchant. But the freight is the cheapest part of the loss.
You also carry the working capital that sat in that unit while it travelled for nine days, the handling on both ends, the quality risk on a product that has been in a van twice, and the opportunity cost of stock that was unavailable to a customer who would have kept it. Merchants who model RTO as freight alone will systematically underinvest in fixing it, which is the single most common mistake in this whole area.
Scale it and the number stops being a line item. Pathao's own worked example: a store shipping 100 orders a day at a 20 percent return rate burns roughly Tk 117,000 a month on parcels that come back.
The finding that reorders every priority list
GoKwik scores cash-on-delivery risk across 180 million Indian shoppers. Their attribution is the most useful number published in this field: 60 to 70 percent of returns come from low buying intent, and only 20 to 25 percent from genuine delivery failure.
Read that carefully, because it inverts the standard playbook. Most RTO advice optimises delivery: better addresses, better couriers, better tracking. That work addresses roughly a quarter of the problem. The majority of your returns were decided before the parcel ever moved, by a customer whose intent was soft at checkout and who was never asked to confirm it.
Meesho's audited IPO filings make the same point from a different direction: 75.5 percent delivery success on cash-on-delivery orders against 97.8 percent on prepaid. Same products, same couriers, same customers, same addresses. A 22-point gap that exists purely because of how the promise was made. No amount of address verification closes a gap that is made of intent.
Tier one: interventions with a measured effect
These have published numbers behind them. Do these first, in this order.
Converting the order to prepaid, or taking a partial advance through mobile money, cuts returns by up to 55 percent. It is the strongest single lever on record, and the mechanism is obvious once stated: a customer with money already committed answers the rider's call. Even a small advance changes the psychology of the doorstep entirely.
Automated pre-dispatch confirmation cuts returns by 20 to 30 percent on its own. This is the highest-leverage intervention that does not require the customer to part with money before delivery, which makes it the realistic starting point for most merchants in COD markets.
Between them, those two account for more measured effect than every other tactic on the standard list combined. If you do nothing else this quarter, do these.
Tier two: sound logic, thinner published evidence
Address verification and normalisation. Clearly useful, and it addresses the 20 to 25 percent of returns that are genuine delivery failures. What is missing is a credible published figure for how much it moves the total, so treat the size of the win as unknown rather than large.
Risk scoring at checkout, then routing risky orders into confirmation and clean orders straight to dispatch. The logic follows directly from the GoKwik attribution, and it is what makes confirmation economically viable at volume, since calling every order is not affordable. The measured effect belongs mostly to the confirmation step it triggers rather than to the scoring itself.
Courier selection by route performance. Real, but bounded by the same ceiling: it can only affect the quarter of returns that are delivery failures.
Tier three: repeated everywhere, measured nowhere
These appear on nearly every list. We could not find a published, sourced figure for the effect of any of them on RTO specifically. That does not make them wrong. It makes them unranked, and it means nobody should be spending a quarter on them before finishing tier one.
Better product photography and descriptions. Plausible, since expectation mismatch causes refusals at the door, but the effect on RTO is asserted rather than measured.
Prepaid discounts. The mechanism is the same as tier one, and this is really a pricing lever aimed at the same outcome. What is unpublished is how much discount buys how much conversion, which is the only number that tells you whether it pays.
Tracking notifications. Genuinely reduce support load, which is a real benefit, and are frequently credited with reducing RTO without a number attached. Worth doing for the WISMO saving alone.
Blocking repeat offenders. Sensible, and increasingly built into risk scores, but published effect sizes are scarce and the false-positive cost is rarely discussed.
COD limits above a cart value. Reduces exposure per failure rather than the failure rate itself. Useful, different thing.
Why the ritual collapses at exactly the wrong moment
Sellers in COD markets already know confirmation works. In Bangladesh, where around 75 percent of e-commerce is paid in cash at the door, the confirm-everything phone call is close to universal among serious f-commerce sellers. The knowledge is not the gap.
The gap is that the ritual does not scale. At fifty orders a day it needs one person doing nothing else. At two hundred it needs a shift plan, a script, and someone to manage the people following it. So it gets rationed, then skipped, and the return rate drifts back toward the market band while everyone blames the courier.
This is the actual shape of the problem. Not that merchants do not know what works, but that the thing that works is made of labour, and labour does not scale with order volume at a constant cost.
What changes when the confirmation is not made of people
The interventions in tier one are both actions rather than answers. Something has to dial a phone, hold a conversation in the customer's own language, understand what was said, and then write the result back onto the order so dispatch can act on it. That is why a messaging widget cannot deliver either of them.
This is the loop Omnistra runs. Incoming orders are scored against delivery history and order shape. Risky ones go into a confirmation queue and an Omnistra voice agent calls within minutes of checkout, in Bangla, English or any of 60 other languages: it confirms intent, reads the address back, and requests a partial advance where the merchant's policy calls for one. Clean orders ship untouched. Failed deliveries go into a recovery workflow rather than straight to return.
Underneath sits the part that makes it auditable rather than merely automated. A workflow engine defines what the agent may do, so behaviour is something a human wrote down. Order OS holds the order and courier status together. Every conversation is classified against a taxonomy of more than a hundred tags, tracked for emotional trajectory while it happens, and analysed after it ends, so the effect on your return rate is measurable in your own numbers rather than borrowed from someone else's case study.
Gartner projects that agentic AI will autonomously resolve 80 percent of common service issues by 2029. Confirmation calling is the least glamorous version of that forecast and, in a cash-on-delivery market, the most valuable.
What we are not claiming
The numbers above are other people's, and they are honest about their origin. GoKwik's attribution is from the Indian market. Meesho's is an audited filing from one large marketplace. Pathao's cost example is Bangladeshi. None of them is a guarantee for your store, your category or your city, and any vendor quoting them as a promise is overselling.
It is also worth saying plainly that no national RTO statistic is published for Bangladesh. The 20 to 30 percent figure widely quoted for the market is a practitioner estimate triangulated from Pathao's published success band and India's audited numbers. The direction is beyond dispute. The precise number for your store is something only you can measure.
Which is the real recommendation underneath all of this: instrument the thing before you optimise it. Measure your own return rate by channel, by category, by city and by payment method for one month. Then start at tier one, because that is where the evidence is.
Every figure here was checked on August 15, 2026 and carries a named source. Corrections are welcome: contact@omnistra.io.
