Ask a retail board about returns and you will usually get a conversation about policy: how many days, who pays for postage, whether the label goes in the box. Ask the team on the goods-in dock and you get a different conversation entirely — about vehicles arriving unannounced, about bench space, and about how long a coat sits in a cage before anyone decides what it is worth.
Both conversations are about the same garment. Only one of them is about the cost.
A return is not a reversed sale
When a customer buys, the business has weeks of notice. Demand is forecast, stock is allocated, pick faces are replenished, labour is rostered against a plan. Everything about the outbound flow assumes you knew it was coming.
When a customer returns, none of that happens. The item arrives with no purchase order, no expected date, no known condition and no confirmed destination. Operationally it is an inbound delivery that nobody planned — and it arrives in ones, not pallets.
A return is an inbound delivery with no purchase order, no expected date and no known condition.
Seven touches before it can be sold again
7
handling steps between a return and a resale
Each one is a place where cost accumulates and information gets lost.
2
of those steps are usually costed
The first and the last. The expensive middle is rarely measured per unit.
- Initiation. The customer tells you something is coming back — or doesn't, and it simply turns up.
- Collection. A carrier network or a store handover, each with its own tracking, timing and failure modes.
- Goods-in. The parcel is received against an order that closed weeks ago.
- Inspection and grading. Saleable, refurbishable or written off? A human judgement, rarely recorded consistently.
- Refurbishment. Steam, re-tag, re-bag. Real labour, and almost never costed per unit.
- Disposition. Which channel does it go back to — full price, outlet, marketplace, clearance?
- Re-listing. It becomes available again, at a price somebody has to decide.
Most retailers can tell you what step one and step seven cost. Very few can tell you what steps three to six cost, per unit, by category — which is precisely the range where the money goes.
Outbound
Inbound
The outbound path is planned weeks ahead. The inbound path is planned when the vehicle arrives.
The part that is genuinely predictable
Here is what makes returns unusual: they are one of the most forecastable flows in retail, and one of the least forecast.
Return rate is stable by product attribute. Fit-sensitive categories behave differently from accessories. A particular fabric, a particular fit block, a particular size curve — these repeat, season after season. In fashion the signal is often strong enough to see a return problem in the first fortnight of a launch, long before the returns themselves arrive.
What we have seen work
- Forecast returns per line per week, rather than applying a blended rate across the range.
- Grade at the point of receipt, not in a batch on a Friday.
- Decide disposition automatically for the clear cases; escalate only the genuinely ambiguous ones.
- Feed grading outcomes back into buying and design, not only into finance.
Where AI actually earns its place
Not in the chat window. The useful applications sit further back in the process.
Predicting return likelihood at the point of sale
Not to block the sale — to plan capacity, and occasionally to intervene. A size guidance prompt on a line with a known fit problem is worth considerably more than the same insight delivered as a report three months later.
Consistent grading
Grading is where value is created or destroyed, and it is usually the least consistent step in the chain. Image-based assessment gives a repeatable baseline; the human handles the edge cases, and the system learns from what they decide.
Disposition routing
Which channel, at what price, given current stock position and weeks remaining in the season. This is an optimisation problem with a clear objective — exactly the kind worth automating, and exactly the kind that people are slow and inconsistent at.
Automate the decisions that are repetitive and well-defined. Keep people on the ones that are ambiguous and expensive to get wrong.
The question worth asking first
Before any of that: can you say, today, what a returned unit costs you by category?
In our experience most retailers cannot — not because the data is missing, but because it sits in three systems that were never joined. That is a smaller problem than it sounds, and solving it changes the conversation from policy to economics.
In short
- Treat returns as unplanned inbound supply, not as reversed demand.
- Cost the middle of the process, not just the two ends you can already see.
- Forecast return rate at line level — the signal is there and it is stable.
- Automate the clear grading and routing decisions; keep people on the ambiguous ones.
- Join the cost data before buying any technology to act on it.