GreyOrange said on 1 September 2026 that more than 3,800 retail locations worldwide are now live on its technology, and named H&M Group as one of the largest of those deployments. The company badges this as “physical AI”, but strip the marketing back and what sits underneath at item level is unglamorous and well understood: passive UHF RFID tags on garments, overhead and handheld readers picking them up, and a software platform turning those reads into work for shop floor staff.
GreyOrange is a US headquartered warehouse and retail automation company, known first for its GreyMatter fulfilment orchestration software and the mobile robots that run under it. Its retail product, gStore, is a store inventory and fulfilment platform that ingests RFID reads alongside sales data to hold a live picture of what is on the shop floor and what is in the stockroom. H&M Group is the Swedish fashion retailer behind H&M, COS, Arket and Weekday, and operates in the region of 4,000 stores worldwide.
Read the 3,800 figure carefully
That last number matters, because the headline figure is easy to misread and several syndications of the announcement have misread it. The 3,800 locations are GreyOrange’s total installed base across every customer it has, not H&M’s store estate. H&M Group’s own footprint on the platform is described only as “hundreds of retail locations spanning three continents”, plus GreyMatter orchestration across multiple distribution centres. Because H&M runs roughly 4,000 stores in total, 3,800 sits close enough to look like a full chain rollout. It is not one.
GreyOrange’s other figures give a better sense of scale: roughly 200 million retail inventory items under management through gStore, more than one billion US dollars of inventory moving through its systems each month, and 130,000 hardware and software agents across stores and warehouses globally, 100,000 of them added in the two years to September 2026. All are vendor supplied and unaudited, and “agents” is GreyOrange’s own unit covering everything from a robot to a software process, so it is not comparable to a reader count.
Overhead readers change what inventory accuracy means
The technical substance here is the shift from periodic to continuous reading, and the two approaches produce numbers that look alike and behave nothing alike. Handheld UHF RFID cycle counting is the established method in fashion retail: an associate walks the floor with a handheld reader, sweeps the fixtures and produces a count. It is fast, cheap in capital terms, and accurate on the day it is done. But it is a snapshot. From the moment the count finishes, every sale, every return and every garment carried into a fitting room and abandoned there degrades it, and in a busy store that decay is measured in hours. Counting weekly means working from data that is usually several days old.
Fixed overhead readers mounted in the ceiling read the same tags continuously, so the output is a rolling state rather than a count. That supports tasks such as “this fixture is empty now” instead of “this SKU was short on Tuesday”. GreyOrange has claimed since January 2024 that gStore with overhead reading holds item level inventory accuracy at around 99%, against roughly 90% for traditional point in time cycle counting, and locates items to within three to five feet. That precision is enough to say which fixture or zone an item is in. It is not enough to say which peg, and any deployment plan assuming otherwise will disappoint.
gStore is sensor agnostic, using hardware from suppliers such as Zebra and Keonn rather than a reader line of its own. The economics split cleanly too: overhead arrays suit high volume stores where continuous data pays for the ceiling infrastructure, while handhelds remain the sensible choice for lower volume sites. That is why the largest chains are going overhead first, and it is the same pattern visible in other overhead deployments reported during 2026.
The Zebra tie-up did the groundwork
The September 2026 milestone follows an announcement on 8 January 2026 in which GreyOrange and Zebra combined Zebra’s SmartLens overhead RFID hardware with the gStore execution platform. Overhead readers gather item level data continuously, gStore processes the reads in real time, identifies stock gaps and pushes tasks to associates on mobile devices, removing manual cycle counts and replacing reactive replenishment with scheduled work.
Two details there matter more than the headline. The combined system integrates with a retailer’s existing handheld RFID infrastructure rather than replacing it, which counts for a lot when most chains considering overhead reading have already spent years building handheld estates and tagging programmes. And the first deployment, at an unnamed national fashion retailer, reached 98% on shelf availability. That is a strong figure, but on shelf availability and inventory accuracy are different measurements, and one unnamed site is not a trend.
Troy Siwek, General Manager of gStore by GreyOrange, framed the problem as latency rather than data volume, saying retailers “don’t struggle because they lack data”, but because “they can’t act on it fast enough.” That is a fair description of where item level RFID has stalled in a lot of stores: the tags are on the garments, the reads are happening, and the resulting exception list never reaches anyone with time to walk the floor.
A warehouse relationship that grew into a store one
This is not a new partnership, which is how some coverage has framed it. H&M Group first deployed GreyMatter at a distribution centre on the US West Coast in 2021, and the store side came later. That sequence is the more interesting part of the story: a fulfilment automation supplier working from the back of the network towards the shop floor, which is the direction of travel for several vendors in this space.
Akash Gupta, Co-founder and CEO of GreyOrange, credited H&M with shaping the product rather than simply buying it, citing its “scale, operational complexity, and willingness to co-develop the future of omnichannel technology rather than simply adopt it”. David Pino, Head of Emerging Technology at H&M Group, said scaling successfully requires “specialized understanding of how stores, fulfillment operations and inventory flows work together”. Those are the quotes a press release produces, but the underlying claim, that this took five years and started in a warehouse, is checkable.
Continuous reading creates its own problem
Reading tags all day rather than once a week surfaces a difficulty handheld counting largely hides: readers do not respect the boundaries retailers care about. Stockroom stock reads as shop floor stock, a fitting room reads as a fixture, and an adjacent department’s tags bleed into a zone they do not belong to. In an open plan store that is the normal condition, not an edge case, and it is what stands between a continuous read stream and a count anyone will act on.
GreyOrange’s answer is SoftFence, announced in October 2024, a virtual shielding layer tuned per site that uses received signal strength, signal movement and angle of arrival to judge whether a tag genuinely sits inside the zone being read. Its initial focus was stray tags confusing associates during omnichannel shipping and receiving with handhelds, but the same physics applies to ceiling mounted arrays. That a vendor with this installed base needed a dedicated product for it says plainly that zone discipline, not raw read rate, is the hard engineering problem in overhead retail RFID.
Where this leaves the category
GreyOrange’s other named retail customer is Fabletics, which deployed gStore in August 2025. Keonn is named as an RFID hardware and smart mirror partner, Cognizant as a retail concept partner. Reporting by Chain Store Age on 1 September 2026 adds that gStore combines RFID reads with sales data and smart mirror interactions to build real time inventory heat maps, the same data used to flag fast moving SKUs, garments abandoned in fitting rooms and low stock alerts. The customer outcomes GreyOrange has published, a 45% productivity improvement, a 5% sales increase and 60% more daily customer assistance, all come from unnamed sites, and the accuracy numbers quoted across retail RFID have a long history of travelling without their baselines.
What is solid is the installed base. A category that was still pilot shaped five years ago now has a supplier claiming 3,800 live stores and a global fashion group running it across three continents. The open question for 2027 is whether continuous overhead reading becomes the default in high throughput fashion stores or settles as a top tier option while handhelds carry the long tail. That will turn less on reader performance than on whether retailers can restructure store labour around a task list that never stops updating, a change to how a rollout is run rather than to the technology in the ceiling.
More: retail RFID, fashion RFID, and our guide to implementing RFID for stock control and retail inventory.
