How do RFID counts work? Batch reading and discrepancy processing
Good RFID stocktaking depends less on scan speed than on resolving every discrepancy. A complete count has four steps: on-site preparation, bulk reading, data reconciliation, and discrepancy resolution. The fourth often accounts for over 60% of the work. Across several warehouse and retail projects, we have broken the process into standardized, reusable tasks.
Point 1: Pre-intake preparation to take stock of
The label health check is earlier than the device starts. There are three types of common preparation blind areas: the tags are off, the labels are wrong and the reader antennas are blocked. 48 hours before the inventory, a sample of 10-15% labels is taken, confirming that the EPC (electronic product code, "ID number" for each label) area can be activated normally, and the signal strength (RSSI value, unit dBm, values are weaker). The RSSI labels below -65dBm are "AH", and the support labels are re-painted or added.
Environmental interference is also checked in advance. Metal shelves, liquid stacking areas, and over-crowded clothing will significantly reduce the recognition of UHF RFID (UHF, working at 860-960 MHz bands). Clearing the operating lanes before counting, leaving a net area of 1.5-2 metres ahead of the antenna.
Reader-end confirms that solid version, antenna feed, reading and writing power corresponds to the last count. The scene of multiple readers’ network (e.g. portals) is coordinated in small ranges to ensure that time synchronisation and trigger logic are consistent.
Point 2: Do not mix the three ways of mass reading
Handheld Terminal
The industrial RFID handheld terminals can identify 100-200 labels/seconds within 2-3 metres of the sky, provided that the antenna direction is stacking up against the label. One job recommends a single-stop scanned shelf to avoid a faulty relationship between the shelves.
Channel Gate
The portal is usually equipped with 4-8 antennas covering a width of 1-3 m, and the recognition rate can be more than 99% when you cross 200-500 boxes. Note the effect of stacking on inner labels, and then tops the antenna if necessary.
Fixed Dragon Gate
Suitable for full-pallet counts of large or irregular items, such as palletized finished goods or individual pieces of large equipment. A multi-antenna portal with a constant-speed lane can read a whole pallet in two minutes, but the goods must travel in a fixed orientation on standardized pallets.
The three methods are not mixed. The mix makes the data list confusing, and then it is difficult to handle the links. In the projects we've done, the single-way count of data discrepancies can be kept within 0.3%, with a mix of them often at 2-3%.
Element 3: Data matching to divide differences into three categories
Once the inventory data enters the backstage, the differences are divided into three categories for processing:
- Type A: Systems are in place and not read on siteThe most common is 70% - 80% of the difference. The reason is that labels are masked, misdirected, read too far away. The method is to do a field sweep: change angle, close to reading, remove the shield and then sweep it.
- Type B: On-site read, system not available (Unexpected Tag). It is possible that the new arrival is unrecorded, the labels are mislabelled and the last count of excess tags are not cleared. The manual check must be made to physical verification, not directly to the system.
- Type C: Same number but misattribution (Miscplaced Tag)The system does not see a problem at the data level, for example, when the label A is on the B shelf. Only if you compare the tag ID-position ID map to the map. It is recommended that each count should be run and verify the script.
The priority for processing is C > B > A. Category C means that system data are no longer credible and must be processed first; category B requires on-site evidence before entry into the accounts; and category A allows batch cleaning and one-time updates.
Element 4: Discrepancies are processed in four steps
First step, export three different statements.(Missing / Unexpected / Misplaced), print by the place number. Each line contains at least: tag EPC, belonging to SKU (inventory unit, i.e. cargo number), system position, field measured location, different type of item.
Step two, field-by-site verification.The list of differences in each line requires physical confirmation and cannot be based on the "brain" report.
Step three, system data update.Confirms the batch import, records the person, time, original/new value and allows for retroactive. If WMS (storage management system) or ERP (enterprise resource management software) are used, the permissions and operating logs are confirmed before updating.
Step four, remaining discrepancies reported.Where physical damage to the label is not possible to activate, the owner or purchaser is reported on a separate list with physical photographs and inventory records and does not sell directly.
Point 5: Rewind is more valuable than a single count
The end of a single count is not the end. The next count will be faster and less variable depending on this reset.
The refill answers three questions: what is the difference rate, which is the highest, and which of the differences is concentrated in which slots or SKU. The category A difference exceeds 1% twice in a row, to be sorted through the reader layout; category B continues to appear, to check the registry entry; category C multiple distribution indicates that the standard of the place sticker was not implemented.
It is recommended that an inventory discrepancy analysis table be produced after each count, with improvements. After two or three rounds of double-discretion files, the difference can normally drop from 1 to 2 per cent, which is common in industry, to less than 0.5 per cent.
The core of the RFID inventory is process standardization and discrepancy loops. Prepare-read-reciprocal-disposal-reverse five steps to fix, each with a checkable output, and the difference rate will naturally be crushed. It is recommended that the whole process be run through a particular class of cargo first, then spread over to the entire warehouse, not just go up there and spread it out in full.

Ms. Lei
Mr. Pang