This was the week retail AI crossed a line. Not better chatbots. Not smarter recommendations. Actual operation — AI running inventory, pricing, transactions and markdowns, with humans supervising the system rather than approving every decision inside it.
Four signals. One story.
1. H&M built the sensor layer autonomy needs
H&M Group rolled out GreyOrange’s gStore platform to hundreds of locations across three continents. This isn’t a pilot.
Worth being precise about the numbers, because they get quoted wrong. GreyOrange says its technology now runs in more than 3,800 retail locations globally and has managed roughly 200 million inventory items across all live deployments — that is the vendor’s entire customer base, not H&M’s footprint. What is specific to H&M is the depth: gStore in hundreds of stores on three continents, RFID readers locating any item to within three to five feet, and a real-time inventory heat map that tells the retailer which SKUs sell fastest, which get abandoned in fitting rooms, and where stock is running low.
Be equally precise about what it does. H&M describes gStore as driving faster replenishment and better picking efficiency — associates acting on machine-generated instructions. This is not yet an autonomous store. It is the thing autonomy is impossible without: item-level ground truth. H&M put GreyMatter into its warehouses in 2021. Five years later the same orchestration logic is on the shop floor. The store became a sensor grid with an operating system on top, and that is the precondition for everything in the three signals below.
2. Agentic checkout went mainstream — Walmart, Gemini, Visa
Salesforce projects that AI agents will handle close to half of all online shopping. That’s not a forecast for 2030. That’s the 2026 trajectory.
Walmart integrated Google Gemini into its product catalogue for conversational shopping. Visa pushed AI agent payment rails into high-frequency daily purchases — the boring transactions where autonomy actually matters. Walmart’s Sparky agent already plans full trips: checks inventory, weather, baskets sale items, checks out.
The gap this exposes is the one every board should be asking about. McKinsey’s State of AI work found that 88% of companies now use AI in at least one function, but only 39% see any impact at EBIT level — and where impact exists it is usually under five percent. That is a cross-industry figure, not a retail-specific one, and you will see it misquoted as “88% of retailers” all over the vendor internet. The underlying point survives the correction: near-universal adoption, rare economic value. The difference is passive AI versus agentic AI. Dashboards don’t move margin. End-to-end agents that forecast demand, schedule labour and transact do — vendor case studies claim order-error reductions around 30% and excess-inventory reductions of 20–30%, which are supplier-reported and worth pressure-testing before you put them in a business case.
3. Luxury went AI-first — TheCode’s invite-only relaunch
Love the Sales rebranded as TheCode and relaunched as an invite-only marketplace with more than 100,000 curated styles from around 500 brands. Its own framing is the interesting part: deliberately luxury-first and quietly off-price. Proprietary AI handles product relevance, intent recognition and a personalised edit for every shopper.
The strategic bet isn’t that AI kills the end-of-season sale. It’s that AI makes the sale invisible. Off-price is where luxury brand equity goes to die — discount depth becomes the product. TheCode is wagering that curation and intent modelling let brands clear stock through a channel that leads with desirability rather than markdown percentage. If it works, it changes who controls the clearance relationship.
Meanwhile Kering is running a different play. Luca de Meo’s ReconKering, unveiled at April’s Capital Markets Day, is a group turnaround strategy rather than an AI programme: restore financial discipline, rebuild desirability, refurbish or relocate two-thirds of the Gucci network, cut selling space by 20%, take a billion euros out of inventory, double jewellery revenue by 2030. Note what sits inside it — inventory reduction and sales-density targets at that scale are not achievable on spreadsheets and instinct. The AI requirement is implicit in the operating targets, which is arguably a more honest signal than a press release about agents. LVMH, for its part, has unified data across its Maisons on Google Cloud for agentic commerce. Luxury stopped treating AI as a website feature. It’s becoming the merchandising model.
4. Thailand is operating, not piloting
Thailand consistently ranks near the top of regional AI adoption-growth rankings, and the deployments behind that are real. CP AXTRA — operator of Makro and Lotus’s across 2,600+ stores — has put AI at the core of its retail operations through its Tencent Cloud partnership. Lotus’s Pick & Go, cashier-less with TrueMoney scan-and-walk-out, is live and running on Tencent Cloud infrastructure. The Mall Group’s Power Mall is positioning itself as an AI retail and innovation destination.
The shift in language matters: Thai retail AI is moving from assistant to operator — pricing, restocking and promotions running with review rather than line-by-line sign-off. Computer-vision smart stores, GenAI catalogues, virtual try-on, AI-assisted store layout.
And the sharpest work isn’t coming from the biggest balance sheets. Declaring my interest plainly: I’m CEO of Xponential and Chief Digital Officer of Siam Piwat Group, so treat the following as a practitioner’s note rather than neutral analysis. Xponential is a roughly 44-person team inside Siam Piwat (Siam Paragon, ICONSIAM and more) shipping AI daily: computer-vision footfall that feeds retailers, generative enrichment of what shoppers actually browse, and agents surfacing insight to leadership through the day.
The position I’ll defend in public is this: the “360-degree customer” is a myth. You cannot see what a shopper buys at the bank, the insurer, or the shop next door — and under PDPA you shouldn’t try. So personalise on what you can actually observe, in near-real-time, instead of building for a fantasy profile. That constraint is not a weakness. It is what makes the system shippable.
Asia transacts while the West announces. Same pattern as last month, accelerating.
The connection
Physical AI (H&M’s sensor grid) + transactional AI (Walmart/Visa agents) + merchandising AI (TheCode, and the targets buried in Kering’s turnaround) + operational AI (Thailand’s pricing and restock). Four different layers, same direction: AI isn’t helping humans run retail anymore. It’s running retail, with humans supervising.
The adoption-versus-impact gap tells you who’s stuck. If your AI still needs a human to click “approve” on every decision, you’re in the 88%. The 39% let it act.
What this means / What to do
- Audit for autonomy, not adoption. “Do we use AI?” is the wrong question. “What runs without human touch?” is the right one. If the answer is nothing, you’re behind.
- Fix the catalogue for machines, not humans. Shopping agents buy on materials, durability and sizing data — not brand story. AI agent optimisation is becoming what SEO was. Structured product feeds win.
- Start with inventory truth. H&M’s play works because RFID gives the system ground truth. No sensor layer, no autonomy. If you don’t know where every item is to within a few feet, agentic anything is theatre.
- Watch Thailand, not Silicon Valley. Some of the most aggressive autonomous retail deployment this quarter is happening in Bangkok, not San Francisco. If you’re operating in SEA, the benchmark just moved.
Sources: GreyOrange / GlobeNewswire (1 Sept 2026), Chain Store Age and Sourcing Journal on H&M gStore; Drapers and Retail Technology Innovation Hub on TheCode (Sept 2026); Kering Capital Markets Day, ReconKering (April 2026); McKinsey State of AI; Salesforce Connected Shoppers; CP AXTRA × Tencent Cloud; NRF.
Disclosure: I am CEO of Xponential and Chief Digital Officer of Siam Piwat Group, both referenced above.
You must be logged in to post a comment.