An online marketplace that offered some customers a partial refund without returning an item saw higher subsequent spending and lower return-processing costs in a randomized experiment involving 853,521 customers.
The option also increased refund payouts and was followed by more return requests, according to the Marketing Science study. The results describe one platform’s policy and should not be treated as a universal formula for handling returns.
Researchers Lu Fang, Chunhua Wu, Zhe Yuan and Kaifu Zhang examined a system that assessed evidence such as customer messages and images. A customer offered a partial refund could keep the item instead of sending it back.
Why a return can cost more than postage
A returned product may need a shipping label, transport, inspection, handling and restocking. Some items cannot be resold at their original price. The customer, meanwhile, may need to pack a parcel and make a separate trip to send it.
A returnless partial refund is one possible alternative: the customer keeps the product and receives part of the purchase price. It can suit a minor defect where an item remains useful. It may make less sense where an item has substantial resale value or the customer cannot use it.
The comparison in the study was an offer of that option against the platform’s traditional return process. Not every person eligible for the offer necessarily received or accepted one. The policy and its automated assessment were tested together, so the results do not isolate the effect of automation from the refund arrangement itself.
What the trial measured
The researchers report a net transaction value gain of $1.44 per customer over 30 days and $4.41 over 90 days. That measure combines subsequent transaction activity with relevant refund and processing costs. It should not be confused with a directly observed profit figure.
Customers exposed to the policy made more purchases later, while fewer parcels were returned. But refund payouts increased. The authors also observed more subsequent return requests with patterns consistent with some opportunistic claims. Such patterns do not establish that any particular customer behaved dishonestly.
Wu’s research summary describes the experiment as a test of a returnless refund policy implemented using multiple AI systems. The paper’s co-author Zhang was affiliated with the partner company, an interest readers should consider when assessing the work.
Questions for a retailer
A useful test would compare the cost of the refund with shipping, inspection, restocking and the value of recovering the product. A retailer also needs to consider whether customers return, how often they request further refunds and whether the policy meets consumer-protection obligations in the markets where it sells.
Businesses should tell customers what alternatives are available and provide a route for resolving an incorrect decision. Local rules and the terms of sale can affect a customer’s rights; the study does not settle those legal questions.
In another retailer experiment covered by MBN, environmental reminders reduced return shipments. Both studies show that returns are affected by policy design, but they test different interventions and cannot be expected to produce the same results elsewhere.