Let me say the quiet part out loud. RFID spent a decade as the technology everyone name-dropped and nobody trusted. It was the pilot that never scaled. The tag tax finance hated. The loss-prevention gadget bolted onto store exits that never touched the P&L in a way anyone could defend.
So retailers filed it under “tried that.” Moved on to computer vision, to demand forecasting, to whatever had a shinier deck.
That filing was a mistake. And the mistake is now visible because the workload changed.
Agents run on state, not reports
Here is what most transformation leaders still miss. Every prior analytics investment tolerated stale data. A weekly forecast could survive a cycle count that was off by thirty percent. A merchandising review could absorb a store that thought it had eight units when it had three. Humans in the loop smoothed over the gaps.
Agentic operations do not smooth over gaps. They act on them.
When an AI agent routes a ship-from-store order, triggers a replenishment, or reprices a slow mover at a single location, it is committing to a decision based on what it believes is physically present. If that belief is wrong, the agent does not fail quietly. It promises a customer an item that is not there. It reorders stock you already have. It burns trust at machine speed.
This is the part the RFID skeptics never had to reckon with, because the old workloads forgave bad data. The new ones punish it. That is exactly why RFID is quietly back. A recent Omni Talk Spotlight conversation with Impinj makes the point cleanly: historically, RFID gave retailers snapshots of where merchandise was last seen, and now improvements in reliability and speed are moving the industry toward continuous streams of item-level data.
Snapshots were enough for dashboards. Continuous streams are the prerequisite for autonomy.
The accuracy math finally clears the bar
Agents amplify whatever you feed them. Feed them a guess and they industrialize the guess. The reason RFID matters now is that it clears the data-quality bar that agentic systems demand, and the proof is no longer theoretical.
Industry research from GS1 UK found that retailers using RFID improved inventory accuracy from typical levels of 65 to 75 percent to 93 to 99 percent, while some retailers also reported sales increases of up to 5.5 percent. Sit with that spread for a second. An agent built on 70 percent accurate state is not a slightly worse agent. It is an agent that is wrong one time in three, which is the same as an agent you will switch off inside a quarter.
The operational payoff compounds beyond accuracy. Retailers with RFID report that ship-from-store fill rates improve from 70 to 80 percent on barcode-based systems to over 95 percent with RFID, because associates can trust the system’s view of available inventory and locate specific items quickly. Trust is the keyword. Fulfillment agents cannot promise what associates cannot find.
And the people closest to the hardware are seeing it too. Per RFID Journal, more item-level data from RFID-tagged products, fed into EPC repositories and enterprise data lakes, enables higher inventory accuracy and richer item-level event data, creating the conditions AI needs to deliver meaningful value. Read that as an infrastructure statement, not a tagging statement. RFID stopped being the feature. It became the substrate.
The Cost of Doing Nothing is now a capability gap
Here is where I get contrarian with my own peers. Most of the RFID coverage still frames the decision as a cost-benefit on tags and readers. Wrong frame. The real question is whether your agentic roadmap has any ground truth to stand on.
Run the CODN. The cost of doing nothing on real-time inventory state is not a line item you can defer. It is every autonomous workflow you cannot ship because your data cannot be trusted to act on. Your competitors are not buying RFID because it got cool again. They are buying the physical-world signal their agents require to run without a human babysitting every decision.
That gap does not show up in this quarter’s numbers. It shows up when your rival’s agents are quietly repricing, replenishing, and fulfilling autonomously while yours are still stuck in supervised pilot mode because nobody trusts the inventory feed.
RFID did not earn its second life by reinventing itself. It earned it because the operating model finally caught up to what the technology always promised: knowing, continuously, what is actually there. The retailers who understand that are not debating tags. They are building the sensory layer their agents will run on for the next decade. Everyone else is about to discover that an agent without ground truth is just an expensive way to be confidently wrong.