Forty-five percent of retailers have scaled artificial intelligence across their operations, while the ones treating it as a core discipline post three times the revenue and EBITDAR growth of their peers. Both figures, as of November 12, 2026, describe the same shelf. The gap is not a software gap. It is the distance between a recommendation arriving in a merchandiser's inbox and a person walking the aisle to reset the physical shelf, and that distance is where most store-level AI budgets quietly stop paying back. Retail News Asia
A recommendation nobody executes is worth nothing
Boston Consulting Group puts the split at 10-20-70: algorithms account for about 10 percent of a retail AI transformation, data and systems 20 percent, and people, processes and institutional structure the remaining 70 percent. Read that way, the 45 percent adoption figure stops looking like a technology problem. It looks like an authority problem, and authority is set by operating models rather than by vendors. Retail News Asia
The failure mode is mundane. A category manager overrides the system because the recommendation conflicts with a supplier commitment. Floor staff do not reset the shelf because resetting is nobody's assigned task before the store opens. The tool logs a win, the like-for-like number does not move, and the next budget cycle treats the whole category as unproven. Retail News Asia
Where the money returns first
Capital is flowing into store and supply chain operations rather than corporate support functions, and the reason is payback speed. Merchandising decisions on assortment, price, promotion schedules and vendor terms release cash quickly enough to fund the next wave of rollouts, while HR, procurement and finance projects sit further from the till. Retail News Asia
Why merchandising beats back-office projects
Merchandising sits close to the till. An assortment change, a price adjustment or a promotion calendar shift shows up in weekly sales, which means the cash released can be reinvested before the next budget review. Back-office projects improve accuracy and compliance, but the link between that work and the sales line is longer and harder to defend. Retail News Asia
The five percent is not evenly spread
An average gain of up to 5 percent in like-for-like sales hides a wide spread. Categories with high substitutability and short shelf life respond fastest, while categories with long replenishment cycles and negotiated listings move slowly. Treating the average as a forecast for every category is how a working pilot gets defunded in the second quarter. Retail News Asia
Shelf assortment is the clearest case in the research. Getting the assortment right yields up to a 5 percent increase in like-for-like sales, which for a grocery format is a meaningful number against thin operating margins. It is also one of the few levers where the output is visible: either the item sells from the shelf or it does not. Retail News Asia
What on-shelf availability really measures
Most retailers still report availability against the warehouse. That number is comfortable and often wrong, because an item can sit in the back room while the shelf is empty and the shopper walks out. On-shelf availability, measured at the fixture, is the metric that connects an algorithm to a customer. Retail News Asia
| Metric | Measured at | What it hides |
|---|---|---|
| Warehouse availability | Distribution centre stock | Shelf gaps behind back-room stock |
| On-shelf availability | Fixture, by SKU and hour | Substitution and lost baskets |
| Distribution coverage | Stores receiving the SKU | Receipt without replenishment |
The middle row is the one that changes behaviour, because it can be assigned to a shift. A store manager who sees a 12 percent gap on a fixture at 6 p.m. has an action; a manager who sees a 96 percent warehouse availability figure has an excuse. Retail News Asia
The vendor toolkit lowers the floor, not the ceiling
Platform vendors are now shipping retail-specific building blocks rather than finished products. OpenAI's retail offering covers customer service, inventory forecasting and personalised recommendations as components a chain assembles against its own systems, which lowers the cost of starting but leaves the operating model untouched. Ster Software
That distinction matters for mid-sized chains. A ready-made assistant can answer the same question a thousand times without fatigue, and it can flag a stockout earlier than a manual walk. Neither capability resets a shelf, renegotiates a promotion calendar or gives a store manager the authority to override a central plan. Ster Software
Agents do not read planograms
Agentic commerce adds a second reader to every catalogue. Software agents assemble baskets against household rules and spending limits, and they do it by reading product attributes, ingredient registries and live inventory across competing domains rather than by admiring visual merchandising. Retail News Asia
The protocol layer is moving quickly. Purchase options were merged into the main agentic commerce branch on November 10, 2026, and Google extended checkout inside AI Mode to eligible Canadian retailers on November 4, 2026. Neither date changes shopper behaviour overnight, but both make machine-readable stock a distribution requirement rather than a nice-to-have. Universal Commerce Protocol
For store operations the implication is unglamorous. If the live inventory figure is wrong, an agent promises an item that is not there, and the failure lands on the store that has to explain it. Availability accuracy becomes a customer service function, not only a merchandising one. Retail News Asia
FAQ
Why do most retail AI pilots fail to scale?
Because the constraint sits in the operating model rather than the algorithm. If buyers can override the system and staff are not assigned the reset task, the recommendation produces no cash.
Which metric should a store manager watch first?
On-shelf availability by fixture and hour. It is assignable to a shift, unlike warehouse availability, which stays high while the shelf is empty.
Does a vendor toolkit replace an operating model?
No. It lowers the cost of starting, but authority, task assignment and incentives still decide whether the output reaches the shelf.
What does agentic commerce change for store operations?
Inventory accuracy becomes customer-facing. An agent that reads a wrong stock figure promises an item the store cannot hand over.
