Should a person call every cash-on-delivery order, or an agent?

The comparison is about the window, not the cost

Most write-ups of this decision start with cost per call. That is the wrong place to start, because a person calling orders by hand is very good and quite cheap right up until a specific thing goes wrong, and the thing that goes wrong is not price.

Confirmation calls only help before dispatch. That means they all have to happen inside the same few hours: after the orders have accumulated and before the courier collects. One caller can work through that window at a fairly steady rate. Add orders and you do not get a slower queue — you get a queue that does not finish, and the orders at the bottom go out unconfirmed, which is the same as not having a confirmation step for those orders.

So the question is not "is a person more expensive than an agent". It is "does the queue still fit in the window", and the honest answer for a small operation is yes.

What a person is genuinely better at

This is worth stating plainly rather than as a hedge, because it decides where the boundary sits.

Hearing tone. A customer who is annoyed, embarrassed, distracted or being pressured by somebody else in the room sounds different, and a person adjusts inside a sentence. An agent cannot do this reliably, which is why the correct behaviour when it detects something unexpected is to escalate rather than improvise — the reasoning is in what an agent must never do.

Judgement about exceptions. A person can decide, on the spot, to waive a delivery charge, split an order, or promise a callback from a specific colleague. Those are commercial decisions and an automated agent should not be making them.

Complaints and history. Anything referring to a previous order needs somebody who can look it up and talk about it. An agent that tries will produce a confident answer, and a confident wrong answer on the phone is worse than a transfer.

Selling. If your calls are really upsell conversations wearing a confirmation's clothes, that is a human activity and automating it will cost you the upsell.

What automation is genuinely better at

Doing the same thing every time. The first pass of a confirmation call is identical: is this the right person, are these the items, is this the amount payable at the door, is this the address, is there anything to change. A person on their fortieth call of the afternoon does that pass less carefully than on their first. That is not a criticism of the person.

Not caring how long the queue is. Capacity is a configuration number, not a hiring decision.

Producing a structured outcome. A caller writes a note. An agent writes one of six values plus a recording, a transcript and any change as a field-level edit. The difference matters downstream: a note has to be read by a human before anything happens, and a value can release stock, re-print a label or fire a webhook without anybody reading anything.

Being auditable. If a confirmation looks wrong, you can listen to the moment it was decided. With hand-dialled calls you are usually reconstructing from memory.

The hybrid is the actual answer

Framing this as either-or is the mistake. The arrangement that works is: automate the first pass, and route the exceptions to your people.

The agent takes every order, confirms what it can, records changes, and escalates anything it is not certain about — an upset customer, a question about a refund, an unclear answer. Your callers then spend their day on a much shorter list of calls that all genuinely need a person, with the recording and transcript already in front of them so the customer does not have to start again.

The effect merchants report first is usually not the cost. It is that the pending queue stops being undifferentiated. Instead of a hundred rows all waiting for someone to decide, you get orders ready to pack, orders to release, orders that need a change applied first, and a short list for a human.

What we will not tell you

We publish no figure for how much a confirmation step improves delivery success, and we do not repeat the ones circulating online. Bangladesh's largest courier does publish the band most merchants operate in, which is a real, dated, attributable number and is in the table above — but the improvement attributable to confirmation is not something we found published by anyone whose name we could put next to it.

Your own number is also the only one that matters, because it depends on your category, your price point, your ad channel and your courier. The calculator on this site works only from figures you enter yourself, which is why it asks what you currently lose before it tells you anything.

If you want to see what the automated half sounds like before deciding, the call demo plays a scripted confirmation turn by turn, and Features covers what happens to an outcome after the call ends.