I might be able to share some insights on this one : I worked on the uses-cases, tech design and first deployments of RFID at decathlon, back in… 2008-2009 I guess.
We faced major issues back then, but the most important one was the cost of tagging every single product ; both cost of the tag and cost of attaching it (comparatively, antennas and IT where marginal costs at scale). Back in the end of the 2000s, tags were more expensive than today. Decathlon had one major advantage though : it manufactures most of the product. That means we could add the tag in the label at the factory. But still, to make it economically viable, we had to pile-up multiple uses cases. Every single one was important in the balance to justify buying so many tags and tagging so many products. So we went very creative, designing new ways to reduce costs in the whole manufacturing/logistics/storage/retail/after-life chain. Because no single advantage was enough by itself. We also had to imagine these new enablers and uses-cases with the state of this new technology (especially at the time) : airwaves are difficult to predict, and not 100% perfect in terms of detection, collision, …
We invented new ways to do things that were mostly already done, but cheaper (in man-hours for repetitive tasks mostly), better (quality and fiability of results), and/or more frequently. Thing logistics optimization (wrong door / parcel detected earlier), whole lifecycle and whereabouts tracking (for defective product callbacks for example), daily inventory (we had the idea to use cleaning carts with antenas to have a daily store inventory), NOSBOS (« not on shelves but on stock », detecting prodcts that where in inventory but not available at this specific moment in the department/shelf), warranty tracking, …
At the end of the story, self-checkout is mostly a by-product, not a target for the deployment. It’s something that was made possible but not a first target.