This case covers a third-party logistics (3PL) operator running a single 18,000 m² warehouse: 12,000 pallet positions, 8,400 active SKUs across beverages, packaged food and household goods, 60 dock doors, and a WMS that was accurate in theory and 84% accurate in practice. The numbers below are from the six-month review, presented as deployed — including the parts that did not hit the plan.
The baseline problem
The warehouse's pain was not slow counting — it never counted in full. The pain was disagreement: the WMS said 4,120 cases of a SKU; the shelf had 3,600 or 4,700, and nobody knew which without a physical check. Consequences: safety stock padded 18% to absorb the uncertainty, dock-to-stock time averaged 9 hours because receiving verification was manual, and two of its brand customers were threatening penalties for inventory-record errors above 10%.
What was deployed
| Item | Quantity | Notes |
| --- | --- | --- |
| Dock-door UHF portals | 12 | Covers 60 doors, but peak flow used 12 lanes |
| Forklift-mounted readers | 6 | Putaway and replenishment confirmation |
| Handhelds | 8 | Exceptions, cycle counts, pick verification |
| Pallet license-plate tags | 25,000/yr | EPC-96, encoded at label print |
| Case-level tags | ~1.2M/yr | Applied by co-packers at source |
| Middleware + WMS connector | 1 | Event filtering, Odoo/WMS write-back |
Total hardware and integration investment landed at roughly €140,000, with annual tag spend of about €31,000. Deployment ran in three phases over 14 weeks: receiving first (the highest-labor process), then putaway, then dispatch verification.
Results at month six
| Metric | Baseline | Month 6 | Change |
| --- | --- | --- | --- |
| Inventory record accuracy | 84% | 99.2% | +15.2 pts |
| Receiving throughput | 220 pallets/shift | 1,340 pallets/shift | 6.1× |
| Dock-to-stock time | 9.0 h | 1.8 h | −80% |
| Cycle counts completed | 3/month | 44/month | ~15× |
| Safety stock padding | 18% | 9% | −9 pts |
| Customer record-error penalties | Recurring | Zero in Q2 | — |
The payback model, using only the labor and safety-stock lines, crosses break-even at month 11; with penalty avoidance included, month 8.
The two things that underperformed
First, pick-face verification: handheld verification at picking added 11 seconds per pick and pickers routed around it; the fix was moving verification to the forklift mount, where it adds no motion. Second, case-level reads through shrink-wrapped mixed pallets: mixed-SKU pallets under tight film read at 92–94%, below the 99%+ achieved on uniform pallets. The operational answer was accepting the gap for mixed pallets and validating them at the case break, not chasing hardware fixes.
What the project would do differently
In order: start the exception-queue discipline on day one (the first six weeks of "99% accurate" data still sat unactioned because no one owned mismatches); print pallet tags in-house from the start (outsourced pre-printed tags caused a two-week encoding backlog); and negotiate tag prices against a committed annual volume — the per-unit price dropped 22% at contract renewal simply by committing to volume.
Frequently asked questions
**What does a warehouse RFID deployment like this cost?** This one landed near €140,000 in hardware and integration plus roughly €31,000/year in tags for 12,000 pallet positions. Smaller single-process deployments (receiving only, 4 portals) typically start around €25,000–40,000.
**Why did receiving improve more than picking?** Receiving is bulk, uniform and gate-shaped — ideal for RFID. Picking is person-paced and motion-sensitive; verification hardware has to ride on the vehicle or disappear into the process, which took a second iteration.
**Is 99.2% record accuracy normal at month six?** Achievable, not automatic. It required the exception queue, count discipline (44 counts/month) and process ownership. Deployments without an exception process typically plateau near 95–97%.
