What to know before deploying RFID
RFID is not a faster barcode. The read model, the data model, and the event model are all different, and most failed RFID projects fail by carrying barcode assumptions across. Read this before writing the first line of integration code.
How RFID differs from barcodes
| Barcode | UHF RFID | |
|---|---|---|
| Who decides when a read happens | A person deliberately scans one code | The reader continuously scans every tag in the field |
| Reads per item | One scan is one read | One tag produces many reads per second |
| Range | Tens of centimetres, needs line of sight | Metres, through cartons, no line of sight |
| Items at once | One code at a time | Tens to hundreds of tags at once |
| Identity | A GTIN names a product type | An EPC names an individual item |
| Outcome | Clearly succeeds or fails | Probabilistic: reads can be missed or stray |
| Event boundary | The beep is the event | No event exists, you define it |
The last two rows matter most: a barcode hands you an event, while RFID pours out a continuous stream of raw reads, and turning that into events is your job.
One EPC is read many times
With a barcode, one beep is one product. RFID works differently: the reader sweeps the field in rounds, each round only tens of milliseconds long, and on every round it reports all the tags it can currently see. A tag that has not left the field gets reported again on the next round. In SDK documentation this scan round is called inventory, which shares a name with, but is not, the stock count operation in Easy Inventory, see the glossary.
For example, a shirt lying on a counter for two seconds:
Three hundred reports are still one shirt. Two rules follow:
- To count goods, count distinct EPCs, never the number of reports.
- Your software must fold many reports into one business event. That step is deduplication, every RFID application needs it, and reading tags and deduplication shows exactly how.
Nothing tells you a tag has left
The reader only tells you what it currently sees, never what disappeared. "The tag has left the field" is a timeout you implement: how long without seeing an EPC before you treat it as gone.
So do not design a business flow around a "goods left the warehouse" event coming from the reader. That event does not exist.
Reading is probabilistic
Two kinds of error are always present and must be handled in business logic:
- Missed reads: the tag is in the field but is not reported. Usually caused by metal, liquid, tag orientation, other stock blocking it, or tag density.
- Stray reads: a tag outside the intended zone is read anyway. Usually caused by too much power or reflections off metal racking.
The direct consequence: not seeing a tag does not mean the item is absent. An RFID count needs an acceptance threshold, a rescan pass, and a defined way to handle the difference. Never let a single sweep automatically write stock down because a tag was not seen.
An EPC identifies one individual item
The barcode on a carton of milk says "this is milk type X". The EPC says "this is carton 4172 of milk type X". That is what makes RFID powerful, and it is also extra work:
- You need a rule for allocating EPCs and writing tags. See EPC encoding.
- You need a mapping from EPC to SKU and to documents.
- You need somewhere to put unknown or unallocated EPCs. Never silently ignore them, and never accept them into stock automatically.
A read is not a stock movement
A read carries exactly one fact: this EPC appeared at this antenna at this moment. Business meaning comes from the read point and the process around it:
- The same EPC read at a dock door means goods leaving, at a till means goods selling, on a shelf means goods in stock.
- The same EPC read twice at the same read point in one shift is usually still one event.
Decide what the business event is first, then choose the dedupe window and antenna placement to match.
The environment defines the read zone
The real read zone is set by physics, not by the datasheet:
- Metal reflects the waves and detunes tags mounted on it, so they never gather enough energy to answer.
- Liquid absorbs the signal, so a tag on the far side of a water carton is very hard to read.
- Raising power grows the read zone in every direction, including directions you do not want.
Always test with the real tag, the real product, the real packaging, and the real way stock is stacked. Site survey and antenna placement are covered in the RFID deployment guide.
Settle these questions before deploying
- What business event must be produced, and at which read point?
- How many events should one pass of the goods create? How long is the dedupe window?
- What read accuracy is acceptable, and how is the difference handled?
- Where is the tag applied on the product, and what is the product made of?
- Who allocates EPCs, under which scheme, and at which step are tags written?
- What happens on an unknown EPC, a stray read, or a reader that drops mid-session?
- Which band, power, and channel-hopping rules apply in the country of installation?
Answer these seven and the integration code becomes straightforward. Leave them unanswered and correct code still produces wrong numbers.
Read next
- Integrating RFID into your software if you are wiring this into an ERP or WMS
- Quickstart to connect your first reader
- Reading tags and deduplication for the read sequence and the three dedupe layers
- RFID deployment guide for site survey and antenna placement