This article is based on a Q&A with Nedap Managing Director, Retail Hilbert Dijkstra and Greg Kihlström for The Agile Brand Guide’s Expert Mode series.
Most of us in marketing have spent the last decade working on the top of the funnel, getting better at audience targeting and at proving which campaigns paid off. Whether the product is actually on the shelf when a customer arrives has stayed with operations, tracked in systems marketing rarely looks at.
Hilbert Dijkstra, Managing Director, Retail at Nedap, argues that split no longer holds up. Nedap sells item-level inventory visibility, largely built on RFID, to retailers including River Island and VF Corporation, so he has a direct commercial stake in the argument. We asked him to make it anyway, including the case for waiting. His position is that a wrong stock record is now a brand failure, and that retailers selling through AI shopping agents will pay for it sooner than they expect.
Customers blame the brand for the empty rack
Dijkstra starts with the customer, who has no reason to care which team owns the stock file.
“Inventory accuracy builds brand trust because shoppers judge a brand by whether its products are where the brand says they are. Picture a customer who sees a jacket listed as in stock at her local store, drives over and finds an empty rack. She doesn’t blame the inventory system or the store team. She blames the brand, and she remembers the wasted trip the next time she sees an ‘in stock’ label from that retailer.”
We paid to bring that customer to the product page, and the visit ended in a wasted drive. Dijkstra points out that shopper patience has shrunk since e-commerce put a competitor’s site one tap away from the parking lot. One wrong stock record, as he puts it, now “costs the retailer the sale and wastes the marketing spend that brought her in.”
He cites a survey Nedap conducted with Retail Dive’s Studio: 91% of retail finance leaders said poor inventory visibility hurt customer loyalty and experience, and only 14% were extremely confident in their inventory accuracy. It’s vendor-commissioned and it asked executives rather than shoppers. Still, nine in ten blaming the problem for lost loyalty, next to roughly one in seven fully trusting the numbers, should bother anyone who runs acquisition.
Agents check your stock record, and they can’t check the shelf
An AI shopping agent decides whether to surface a retailer based on the product feed the retailer sends, which lists price and availability for each item. Protocols like OpenAI’s Agentic Commerce Protocol and Google’s Universal Commerce Protocol let the agent read that data and complete checkout, and OpenAI’s feed spec accepts updates as often as every 15 minutes. If the store data underneath is wrong, Dijkstra says, a faster refresh just hands the agent “a fresh copy of the wrong answer.”
“The agent can’t see the shelf, though. It trusts the retailer’s stock record, and most stock records only show whether an item is in the store. They don’t show whether that item is damaged, reserved for another customer or already sold and waiting for pickup. Before a retailer promises anything to an agent, it needs to know what’s available to promise, meaning the items it can sell to the next customer. The agent then learns from what happens after the order. At 70% accuracy, a retailer fails the agent in one of two ways. Say a store has one XXL T-shirt left. If the retailer doesn’t trust its count, it hides the shirt, and the agent sends the shopper to a competitor. If the retailer promises the shirt and it’s gone, the order gets canceled, and the agent is less likely to choose that retailer next time.”
Both failure modes land on marketing’s side of the ledger. When you hide inventory you have, you lose customers you already paid to reach. When you promise inventory you don’t have, you get the empty-rack problem again, with an agent keeping score. Dijkstra expects agents to treat retailers the way Uber treats drivers who cancel too many rides. That part is a prediction. The feed mechanics are already in the published specs.
“Available to promise” is a supply chain term most CMOs have never needed. They need it now, because it’s the figure behind every “in stock” badge and every buy online, pick up in store option at checkout. A retailer that takes 20,000 pickup and ship-from-store orders in a December week and cancels 3% for stock reasons has told 600 customers yes, then no, after paying to acquire every one of them.
Item-level visibility has to pay off with today’s shoppers
We pushed on the obvious objection. Few people shop through agents yet, OpenAI ended Instant Checkout in March, and the prediction comes from someone whose company sells the fix. Dijkstra concedes most of that.
“I also understand the skepticism when someone whose company sells inventory visibility predicts that agents will rank retailers on stock accuracy. If agentic commerce were the only reason to invest in item-level visibility, I’d tell a CMO to wait. Item-level visibility should pay for itself with today’s shoppers, though, through better product availability and fewer canceled orders. Retailers that sell many sizes and colors across stores and online usually see a return on the investment first. A retailer with a small assortment in one channel may not need item-level visibility yet.”
That’s a usable screen, and he applies it against his own sales interest. He also argues against the big-bang rollout, aiming for “80% of the benefits with 20% of the complexity.” Retailers start by tagging items and using handheld readers for counts and back-room-to-floor replenishment in a pilot group of stores. Overhead readers and RFID self-checkout come later, once the basics produce accurate stock.
Timing is where the agent argument returns. Even a fast fleet rollout takes several months after the pilot, and BCG estimates AI-driven traffic could overtake organic search in the US as early as 2028. A retailer that waits for agentic commerce to show up in its sales numbers will spend its first agent-heavy peak season sending agents bad stock data, with the fix still months out.
Look for the lost margin in the back room
Peak season is where the gap opens. Accuracy looks healthy in week one, Dijkstra says, when stores fill with extra stock. By weeks two and three, shoppers have moved items around the floor and associates are rushing through receiving. Theft rises too, and none of those small errors shows up until the next count.
“The lost margin hides in the decisions retailers make with bad stock data. Headquarters skips sending a size to a store because the system says the store already has it. The website offers a jacket for pickup when the last one is gone. After the season, the retailer marks down stock that sat in the back room while the same item was missing from the sales floor.”
Only the first of those decisions stays inside operations. The pickup offer is a customer-facing promise, and the end-of-season markdown is usually something marketing ends up promoting. Once retailers can see every item, Dijkstra says, much of the “missing” stock turns out to have been in the store all along.
His example is River Island. According to Nedap’s case study, weekly counts pushed the UK fashion retailer’s accuracy above 98% and on-floor availability above 99%, letting it carry 30% to 40% less stock. Those figures come from the vendor, though independent coverage of River Island’s earlier pilot reported accuracy rising from 72% to 97%. VF Corporation, rolling Nedap out across more than 1,500 stores starting with The North Face, is extending that visibility to distribution centers and vendor partners.
Dijkstra suggests CMOs take three questions to the COO or head of supply chain. How accurate is stock data at the item level, by size and color in each store? Where does stock go missing between the factory and the customer, and is it theft or errors in transit and stores, since each needs a different fix? Can the systems behind the product page answer far more availability requests once agents start checking 10 or 20 retailers per query?
Then add one number to your own dashboard: the share of online orders canceled because the item wasn’t in stock. Dijkstra calls each one “a broken promise to a customer.” We’ve spent a decade measuring what it costs to bring a customer to the product page. Put the cancellation rate next to your acquisition costs and review both in the same meeting.



