A parcel sorter is judged by one number: how many parcels per hour arrive in the correct chute. Everything else — the reader model, the antenna count, the inlay — is instrumental to that number. Most RFID sortation projects that disappoint do not fail because the readers were wrong. They fail because the read decision was placed somewhere it could not be won, or because nobody defined what should happen when a parcel is not read at all.

Sortation is an RFID problem, not a barcode problem

Manual scanning at induction caps throughput at the speed of the slowest operator and injects a human error rate that grows with fatigue.

A single induction position scanning by hand sustains perhaps 900 to 1,200 parcels per hour on good labels, and materially less on wrinkled polybags.
Barcode read rate on soft mailers, cylindrical tubes and shrink-wrapped multi-packs is the weakest link in the whole line, and it fails exactly where volume is highest.
Every misread or missed scan becomes a manual exception at the end of the belt, which is the most expensive place in the building to fix anything.
RFID moves identification upstream: the parcel is read in a tunnel before it reaches the sorter, so the divert decision is already made when it arrives.

The trade is real. RFID costs more per parcel than a printed label, and it only pays back if it removes manual induction labour and reduces mis-sorts.

The read tunnel: geometry before power

The most common design error is turning up the power to compensate for a tunnel that was built wrong.

Tunnel length should give at least three to five read opportunities per parcel at line speed. At 2.5 metres per second, a two-metre tunnel is under a second of exposure — marginal.
Antennas should face the parcel from at least two directions, because a flat tag lying against a carton wall is invisible from the side it is stuck to.
Do not aim antennas at the belt surface. The reflection from the belt and the metal frame dominates the field and desensitises the readers.
Use shielded tunnels with absorber panels on the frame. In a sorting hall there are dozens of readers and hundreds of tags in motion, and an unshielded tunnel reads its neighbours' parcels.
Set the reader to report every tag on every pass during commissioning, then tighten the filters. You cannot tune what you cannot see.
Keep the tag orientation tolerance in mind: a parcel tumbling down the chute may present the tag edge-on, and no amount of reader power fixes a zero-projection angle.

The target at the tunnel exit is a read rate above 99.5 % on the first pass, measured over at least 100,000 parcels, not over a demo of 200.

Tag placement is a packaging decision

In a parcel network the tag is applied by the shipper or by the hub, and placement is rarely controlled.

Specify a placement zone on the largest flat surface, at least 50 millimetres from any metal edge, battery or liquid container.
For polybags, use a label with a flexible inlay and specify that it must be applied without creasing across the antenna.
For reusable plastic totes, mount a hard tag in a recessed pocket. A surface-mounted tag on a tote is destroyed within weeks by impact and washing.
Test the worst case, not the average case: a carton full of liquid, a box of metal parts, a parcel wrapped in foil-backed insulation.
Provide shippers with a small validation kit and a documented placement picture. Tag placement guidance is cheaper than a rework station.
If the network cannot control placement, design the tunnel to tolerate it, and accept a lower first-pass rate with a fast recovery path.

The no-read path is the design

Every sorter has a no-read path, whether or not anyone designed it. If it was not designed, the path is the sorter's default, which is usually the reject chute and a manual keying station.

Define the no-read action explicitly: reject to an exception lane, re-circulate once, or hold for a second read.
Re-circulation is attractive but must be bounded. A parcel that loops three times consumes sort capacity that the throughput model never accounted for.
The exception lane needs a person, a screen and a scanner, sized for the expected no-read rate — typically 0.3 to 1 % of volume in a well-run network.
Every no-read should be logged with the parcel image, tunnel, timestamp and reader diagnostics. Without that, the cause is unknowable and the rate never falls.
Do not let the sorter controller silently guess. A guessed divert is a mis-sort with a delay, and mis-sorts cost far more than exceptions.

Induction, singulation and the time budget

RFID does not remove the need for singulation; it moves the constraint.

Parcels must still be separated enough that the tunnel reads them individually, or the sorter receives a batch of identities for one position.
If two parcels are read together, the controller has to decide which one is which. That decision is guesswork unless the tunnel can resolve position, which most cannot.
The end-to-end budget from tunnel exit to divert is typically 300 to 800 milliseconds. Reader filtering, middleware and WMS lookup all eat into it.
Do the WMS lookup locally with a cached route table. A round trip to a central server per parcel will not fit the budget at peak.
Time-stamp at the reader and at the controller. Clock skew between the two is the usual reason a correct read produces a late divert.

Cross-belt, popup roller and sliding shoe

The sorter type changes where the read must happen, not whether it must.

Cross-belt sorters carry parcels in trays, and a tag on the underside of the tray is often a better read target than a tag on the parcel. Tray identity works if the tray-to-parcel binding is enforced at induction.
Popup roller and wheel diverters handle irregular shapes and are gentler on soft mailers, but they present the parcel at changing angles, which argues for a longer tunnel.
Sliding shoe sorters run at high speed and short pitch, leaving little physical space for a tunnel. In these lines the read usually happens at induction only.
Whatever the type, the divert command must be idempotent. A duplicated read must not produce a duplicated divert.
Tray and shoe sorters should support a reconciliation report that compares inducted identities to diverted identities at end of shift, by chute.

Integration with the WMS and the sorter controller

Keep the reader layer dumb and the routing logic in the WMS or the middleware. Firmware-side sorting rules become unmaintainable within a year.
Send a compact event to the controller: tag identity, tunnel, timestamp, read count and a confidence or signal indicator if available.
In Odoo, inbound and outbound moves can be driven by scan events through a middleware queue, but do not write stock moves directly from the PLC. Use a validated queue with replay and deduplication.
The WMS should own the destination mapping, because chute assignment changes with carrier cut-offs and wave plans several times a day.
Record the divert outcome against the read event. Without the outcome, you measure reads, not sortation.

Commissioning and acceptance

Run the tunnel for a week with logging only, before any divert is enabled. Establish the baseline read rate and the top ten no-read causes.
Test with the real mix: the smallest polybag, the heaviest carton, the liquid pack, the metal-containing box.
Measure at three speeds, including the peak design speed, not the comfortable one.
Verify the no-read path end to end with deliberately unlabelled parcels, and check that the exception appears on the operator screen within the expected time.
Audit mis-sorts separately from no-reads, and reconcile the two counts against the chute report every shift.

The difference between a good and a great sortation line is rarely the reader. It is the discipline of knowing what happens to the parcel that was not read, and having a person, a screen and a process ready for it before the first peak season.