Delivery failure prediction is software that predicts which parcels are likely to fail before they are sent, and then acts to save them. It confirms the order with the customer, chooses the courier most likely to deliver it, and books the delivery for a time when the customer will be at home. It is usually a part of shipping software.
A delivery fails when the rider cannot hand over the parcel. The customer may be out, the address may be incomplete, or the customer may refuse a cash-on-delivery parcel at the door. The courier then sends the seller a non-delivery report, known in India as an NDR. After two or three failed attempts, the parcel goes back to the seller. This is an RTO, a return to origin, and the seller pays for shipping both ways. For a small seller, a high RTO rate can turn a month's orders into a loss.
The model learns from the seller's past orders. It looks at the pin code, whether the address has a house number and a landmark, whether the customer has refused parcels before, the value of the order, the hour at which it was placed and how long the customer spent on the website. From these, it gives each new order a failure risk, as a percentage.
Take, for example, a footwear seller in Agra that receives an order for a ₹2,400 pair of leather shoes, to be paid by cash on delivery. The order arrives at 1 a.m., and the address has no house number. The model gives it a 41% chance of failing, against 8% for the seller's average order. So the software acts before the parcel leaves. It sends the customer a WhatsApp message asking her to confirm the order and complete the address. The customer confirms, adds the house number and asks for delivery after 6 p.m. The software then offers her ₹50 off if she pays online now, and she pays by UPI. It chooses the courier with the best record of evening deliveries in her pin code. The rider delivers the shoes on the first attempt.
If an attempt does fail, the software reads the courier's NDR at once. It contacts the customer to arrange a second attempt, before the courier sends the parcel back.
This part is different from COD risk scoring at the checkout. COD risk scoring decides, before the order is placed, whether to offer cash on delivery at all. Delivery failure prediction works after the order is placed, on the parcel itself. The two often use the same data. The courier that finally carries the parcel usually runs an express courier network, and the rider's stops come from route optimisation software.