A customer walks into one of your stores to return an item they bought from a third-party seller on your site. Who ends up holding the stock, and whose money pays the refund?

One counter, three parties, and a customer who sees one brand

The customer believes they bought from you. They browsed on your site, paid through your checkout, and the parcel arrived in packaging carrying your name on the marketplace listing. From their point of view walking into your store is the obvious thing to do. Everything difficult about this scenario comes from the gap between that belief and the commercial reality, which is that another company sold the item, owns the goods, holds the margin and wrote the returns policy.

Once you frame it that way the design question is clear. The customer must experience a single retailer, and the platform must therefore absorb every discontinuity — decision, money, physical goods, and reconciliation — and settle it with the seller out of sight. Candidates who try to expose the seam to the customer, by asking them to contact the seller from the shop floor, are describing something that works on paper and is abandoned within a week of launch.

Seller of record decides almost everything downstream

The first thing to establish is who sold the item, because it determines what is even permitted. On a marketplace the seller of record is usually the third party, and their returns policy, their return window, their condition requirements and their acceptance rules govern the transaction. Your own store policy governs your own stock. The customer holds one order that may contain lines under both regimes, which is why the answer has to be per line and never per order.

That has an uncomfortable consequence: two lines in the same parcel can be returnable on different terms, and a colleague at a till cannot be expected to reason about it. The resolution is not to harmonise the policies by fiat, because you cannot impose your window on a seller unless your seller agreement says you can. The resolution is to make eligibility a platform decision, computed from the seller's terms, and rendered to the store as a plain yes or no with a reason. Whether the seller agreement grants the platform authority to accept returns on the seller's behalf, and on what terms, is the single most important thing to have negotiated before any of this is built — and saying that in an interview shows you understand where the real constraint lives.

The colleague needs one answer, fast

At the counter, the scan of a receipt or an order barcode has to resolve to an authoritative decision within a few seconds: is this line returnable here, in this condition, today, and what is the refund amount. That decision needs the order line, the seller's terms, the line's own history of prior returns, and any fraud signals against the customer, and it needs to be recorded as an authorisation with an identifier so the rest of the process hangs off one object.

Two failure modes are worth naming. The first is a store estate whose till system cannot reach the platform, either because of a network fault or because it was never integrated, in which case colleagues are guessing and the exception rate becomes the process. Deciding in advance what a store does when the decision service is unreachable is part of the design; refusing politely is usually better than accepting something you cannot account for. The second is that the return happens at the counter and the customer leaves with their money before anything has been agreed with the seller, so every subsequent step is a compensating action rather than a negotiation. That is a deliberate trade and worth stating as one.

You are now storing another company's goods

The physical side is where this gets under-designed. A returned marketplace item arriving in your store is stock you do not own, in a location that is not built to hold it, and it must not be booked into sellable inventory under any circumstances. If it is, you will sell another company's unit as your own and discover it during a stock count that cannot be reconciled.

So the item needs a distinct holding state — received on behalf of a seller, in a non-sellable location, with a link to the return authorisation. Then somebody has to decide its disposition, and there are only a few sensible options: consolidate it and ship it back to the seller in batches, hold it for the seller to collect, dispose of it locally under an agreed instruction, or buy it from the seller and take it into your own stock at an agreed price. Which one applies is a per-seller commercial term, and the cost of the reverse leg has to sit with somebody explicitly. Store floor space is expensive and finite, so the design also needs an ageing rule and an escalation, otherwise stockrooms silently fill with items no process is pulling.

The related trap is your own availability figures. Marketplace units held in a store must be excluded from every projection your site and your other channels read, or you will promise stock you do not own from a location that cannot fulfil.

Fund it now, recover it later

The money follows the same principle. The customer is refunded promptly by the platform, on the tender they paid with, because making them wait for a seller to agree is the behaviour the whole feature exists to avoid. That refund is the platform's cash going out, and it creates a receivable against the seller.

Recovery normally happens by offsetting the refund, plus any agreed handling or carriage charge, against the seller's next settlement. That works while the seller has a payout balance. What separates a senior answer here is naming what happens when they do not. A seller with a shrinking order flow, a seasonal seller between peaks, or a seller who simply stops trading has no future payout to net against, and your funded refunds become an unsecured loss. The controls are commercial rather than clever: reserves or rolling holdbacks proportionate to return risk, eligibility gates so in-store returns are offered only for sellers above some standing, exposure limits per seller, and a settlement position that is monitored as a credit risk rather than as an accounting artefact.

Platform commission on the original sale is a further wrinkle. Whether your fee is reversed with the refund is a term in the agreement, and whichever way it goes it must be applied consistently, because a partial refund with a partially reversed fee is where seller disputes are born.

The seam is the reconciliation, and it is where this quietly rots

Every one of these returns writes to at least three ledgers: your payments and refunds, your inventory, and the seller's settlement account. They are updated by different systems at different times, and none of them individually notices when they disagree. A refund issued with no corresponding settlement adjustment, an item received with no return authorisation, an authorisation with no item ever received, a settlement deduction for an item that came back to a different store — each is invisible until an auditor or an angry seller finds it.

So the deliverable is not the happy path. It is a daily three-way reconciliation with an owner and a worklist: authorisations without receipts, receipts without authorisations, refunds without deductions, and physical items whose age exceeds the disposition rule. That report is also the thing that makes the credit exposure visible while it is still small.

The customer must see one retailer, which means the platform pays first and argues later. Design for the argument: a per-line eligibility decision the store can trust, goods held in a state that can never be sold as your own, and a settlement position you watch like a credit line.