Apparel is the largest item-level RFID market, and the reason is not fashion. It is arithmetic: a tag costing three to ten cents goes on an item that is individually valuable, individually returned and individually stolen, and it survives industrial washing for years.

The same EPC then serves five different systems — the factory, the distribution centre, the shop floor, the point of sale and the returns process — which is why the economics work here and nowhere else.

Where the chain actually starts: source tagging

Value is created at the factory, not in the store. By the time the carton reaches the DC, the tag is either a reliable identity or a liability.

Specify tag placement in millimetres, including the fold and the seam, so the inlay is not bent or shadowed by the care label.
Require the EPC to be written at source with an SGTIN that matches the purchase order; validate the mapping in the ASN.
Audit a sample per delivery: 20 items per carton lot, read on a fixed portal, with a target above 99 %.
Agree who pays for rework. A supplier that knows rework is charged at your labour rate tags correctly.

Distribution centre: receiving, put-away and the 98 % rule

The DC is where RFID first pays, because it replaces a scan-per-carton process with a read-per-pallet process.

Receiving: a portal reads a 60-carton pallet in under two seconds. The expected count comes from the ASN; any difference becomes an exception, not a manual recount.
Put-away: overhead readers confirm the carton reached the right location. This is where most deployments discover their first antenna blind spot.
The 98 % rule: if a portal reads below 98 % of expected tags on a controlled test pallet, stop and fix the portal. Every downstream number inherits that error rate.
Cross-docking: read at the inbound door and again at the outbound door, and reconcile. A carton that appears at neither is a real problem worth investigating.

Store backroom and shop floor: the accuracy that sells

Store accuracy is the benefit that customers feel, and it is the one most often left on the table.

Backroom: a handheld sweep of the stockroom takes 10–15 minutes instead of two hours, and it happens weekly instead of quarterly.
Shop floor: a rail of 200 garments is read in seconds, which makes daily replenishment checks realistic rather than aspirational.
Sizes and colours: the value is not total quantity but the size curve. Knowing you have 40 units of a style and no medium is a different fact from knowing you have 40 units.
Fit and display: a weekly read of the floor also finds items in the wrong place, which is a merchandising problem no planogram audit catches.

Fitting room and POS: the two places that decide ROI

Two read points carry most of the return, and both are frequently under-designed.

Fitting room: an antenna at the entrance counts items going in and coming out. Items in minus items sold equals conversion pressure, and it gives the merchandiser a size-level demand signal.
POS: the same EPC is deactivated or detached at sale, so the tag cannot be read after the customer leaves. This is a privacy obligation as much as a stock control step.
The linkage: what the fitting room reads and what the POS sells must reconcile in the ERP, otherwise the two numbers become rival versions of the truth.

Returns and reverse logistics

Returns are where most apparel programmes quietly lose the accuracy they gained.

Read the returned item at the service desk; a tag that no longer exists in the sales record is a fraud or a system error, and both need a rule.
Re-tag only if the original tag is damaged, and record the new EPC against the same item identity.
Items that are resellable go back to the shop floor with a location event; items that are not go to a disposition queue with a reason code.
Never return a tag to stock without a physical read. A manual keyed return is exactly the transaction RFID was supposed to eliminate.

Omnichannel: ship-from-store and the single inventory pool

The commercial payoff of item-level accuracy is that the store becomes a warehouse.

Ship-from-store requires a trustworthy on-hand figure per store per SKU per size; without it, orders are routed to stores that cannot fulfil and cancelled at the last mile.
Reserve logic must respect in-transit and fitting-room stock; a garment in a cubicle is not available.
Click-and-collect needs a read at handover, not just a status change.
Publish one accuracy metric across channels, or finance will keep two sets of numbers and trust neither.

A staged rollout that does not stall

Most programmes stall because they try to do all seven stages in one release.

Stage one: DC receiving with source-tagged inbound only. Narrow, measurable, and it proves the portal.
Stage two: one store cluster with cycle counting and fitting-room reads.
Stage three: POS deactivation and returns handling.
Stage four: omnichannel fulfilment, only after store accuracy is measured above target for two consecutive months.

Each stage has an exit criterion, a baseline and a named owner. Without the criterion, stage one becomes the permanent pilot.

Bottom line

Apparel RFID works because one identifier serves five systems and the tag outlives the garment's first season. The value, however, is not evenly distributed.

Source tagging protects everything downstream; skimp here and every later stage inherits the error.
The DC and the store floor are where the measurable labour and accuracy gains appear first.
Fitting room and POS decide whether the programme pays for itself.
Returns and omnichannel are where accuracy is either preserved or quietly lost.

Do the stages in order, measure before and after each one, and the programme stops being a technology project and becomes a retail operating model.