How to Reduce COD RTO With AI Calls: A Practical Guide

The short answer

RTO (return to origin) happens when a courier brings a parcel back undelivered. On cash on delivery orders it is expensive, because you pay for the forward leg, the return leg and the packaging, and you earn nothing. The fix with the best return is a short confirmation call made after the order is placed and before it is dispatched.

That call does three jobs: it checks that the buyer still wants the order, it confirms the address, landmark and pincode, and it agrees a delivery window. Orders that fail any of the three are held or cancelled at the cost of one phone call instead of two courier legs. An AI voice agent can make that call on every order, within minutes, in Hindi, Telugu, Tamil, Marathi, Kannada, Malayalam or English.

Why COD orders come back

Most RTO is not fraud. It is ordinary friction that nobody resolved before the courier arrived. The usual causes, roughly in the order store owners report them, are:

• —The buyer ordered on impulse from an ad or a reel and changed their mind within hours.
• —The address is incomplete: no landmark, a wrong pincode, a flat number missing, or a locality name the courier's system does not recognise.
• —Nobody was home or reachable on the day, and the buyer ignored the courier's calls.
• —The buyer expected a different price, size or delivery date from what the ad or checkout page implied.
• —The order was a test, a prank or a competitor's click, which is a smaller share but a real one.
• —A relative who did not know about the order refused to pay at the door.

Notice that almost every cause is solvable with a conversation held before dispatch. A WhatsApp or SMS confirmation helps, but it relies on the customer reading and replying, and many COD buyers do neither. A phone call in their own language gets a spoken answer.

The RTO maths in rupees

The figures below are an illustration with stated assumptions, not a benchmark. Replace each number with your own from your courier statements before drawing any conclusion. The point is the shape of the calculation, which is the same for a store with 300 orders a month or 30,000.

On those assumptions, the calling programme pays for itself if it moves the RTO rate from 25 percent to about 22 percent. Anything better than that is margin. The calculation is deliberately conservative: it ignores the advertising spend that bought the order, the working capital stuck in returned stock, the damaged-in-return share and the courier performance score that falls as your RTO rises.

Two refinements make the maths friendlier still. Unanswered calls are not billed, so the real bill is lower than the all-connected case above. And the customers who do not pick up after two or three attempts at different times are themselves a risk signal that you can act on, for example by holding the order or asking for a prepaid payment link.

The three calls that cut RTO

1. COD confirmation before dispatch

The call goes out shortly after the order is placed, ideally within the first few minutes while the buyer remembers it. The agent states the store name, the product and the amount payable on delivery, and asks one clear question: should the order go ahead. A yes is logged; a no cancels the shipment before it costs anything; an unsure answer triggers a short conversation about what the buyer wants to change.

Timing matters. If the call lands the next morning, a share of orders will already be packed or handed to the courier. Outpero's instant calling dials a new order's phone number within 30 seconds of it reaching the platform, so the confirmation is effectively simultaneous with the order.

2. Address verification

Bad addresses are the most fixable RTO cause. The agent reads back the address it holds, asks for a landmark, and confirms the pincode and a reachable phone number. It captures corrections as structured fields, so the corrected address reaches your system through a webhook or lands in a spreadsheet rather than staying trapped in a recording.

Language is where this call is won or lost. A buyer in Warangal describing a landmark in Telugu, or a buyer in Coimbatore giving directions in Tamil with English building names mixed in, will give a fuller and more accurate address to a voice that understands them than to an English-only script. Outpero is a specialist in regional-language voice AI for exactly this reason, and the agent switches to English the moment the caller does.

3. NDR follow-up

An NDR (non-delivery report) is raised when a delivery attempt fails. Couriers usually give you a short window to act, often a day or two, before the parcel starts its return. The follow-up call asks the buyer why delivery failed, whether they still want the parcel, and when they will be available, then records the answer for your ops team to feed back to the courier.

NDR calls convert better than most cold outreach because the customer already wants something to arrive. The call is cheap compared with a return, and it works best when it goes out the same day the NDR is raised.

When a voice broadcast is enough, and when it is not

A voice broadcast plays one fixed message to a list and ends when the message finishes. It is not a conversation: the listener cannot answer, correct or ask anything. It is billed at a lower rate than AI conversation minutes, which makes it attractive for high-volume notices. The decision depends on whether the call needs an answer.

A sensible set-up uses both. Confirmation, address and NDR calls are conversations. Shipping updates and out-for-delivery reminders can be broadcasts. Where a conversation reaches a request the agent should not decide, such as a refund dispute, the call can be transferred to your team.

What a confirmation call sounds like

Here is a short opening in Hindi with the English words a real buyer would hear and say.

From there the conversation branches. If the buyer says yes, the agent reads back the address and asks for a landmark. If the buyer hesitates, it asks what has changed and offers to cancel or hold. If the buyer asks for a person, the call can be transferred to your team. Every branch ends with a recorded outcome.

How Outpero handles COD calls

Outpero is an AI employee: a voice agent that calls, holds the conversation and writes up the result. For an online store the pieces fit together like this.

• —Orders reach Outpero by CSV upload for a batch, or one by one through a webhook, Zapier or Make for instant calling.
• —The agent calls in the buyer's language and handles the English mixed in, with the same per-minute price in every language: Standard ₹5 per connected minute, Premium ₹6.8, no setup fee, prices excluding 18 percent GST.
• —A dedicated +91 business number costs ₹649 a month including 1 channel, and each extra channel, which is one more simultaneous call, is ₹649 a month.
• —Outcomes, a recording, a transcript, a summary and your custom fields (confirmed, cancelled, new address, preferred time) go to the dashboard and to your webhook.
• —Bulk campaigns honour calling windows, retry unanswered numbers and screen against DND. Calls come from a DLT-registered line.
• —The agent recognises returning callers, so a buyer who calls back about an order is not treated as a stranger.

Outpero does not run your store or your courier. It reports what the buyer said, and your own system or team decides to ship, hold or cancel. That boundary is deliberate: the decision rules belong to your business.

Setting it up in a week

1. 01Pull your last 60 to 90 days of COD orders and mark which became RTO. Note the RTO rate by pincode, order value and product category.
2. 02Choose where the call adds the most value. High-value orders, first-time buyers and risky pincodes are the usual starting group.
3. 03Write the three outcomes you need from every call: confirmed, cancelled or on hold, and a corrected address if any.
4. 04Test the agent on your own phone, in the language your buyers use, with two or three awkward replies.
5. 05Run a small slice of live orders for a few days and compare RTO against the untouched slice.
6. 06Widen to all COD orders once the numbers hold, and add NDR follow-up as the second workflow.

Keep a holdout group at the start. Without one, a lucky week looks like a result. Splitting by order number (odd and even) is enough to compare like with like.

How to measure whether it worked

• —RTO rate on called orders versus the holdout, measured over the same dates and the same pincodes.
• —Confirmation rate: the share of called orders confirmed, cancelled and left unanswered.
• —Cancelled-before-dispatch count, which is the direct saving. Multiply by your ₹ direct cost per RTO.
• —Corrected-address count and the delivery success of those parcels.
• —NDR recovery: the share of NDR cases that delivered on the next attempt after a call.
• —Cost per order called, taken from the credits shown on each call log, against your direct cost per RTO.

When this will not help

AI calling will not fix a product that disappoints, a courier that fails a pincode, or a listing that promises what the product cannot deliver. If most of your RTO comes from a handful of pincodes where your courier is weak, switching couriers or restricting COD there is cheaper than calling. If your RTO is already a few percent on COD, the break-even above gets harder to clear.

There is also a customer experience cost if calls are clumsy. Keep them short, in the buyer's language, and give a clean way to cancel or reach a person. Calling an existing customer about their own order is generally treated differently from cold outreach, but the exact position depends on your business; confirm it with your compliance advisor.

Frequently asked questions