Retailers are getting serious about AI in 2026, shifting focus from isolated tools to embedded, agent-driven architectures that connect real-time data, decision intelligence, and execution inside core retail workflows.
Retailers have spent the past several years experimenting with artificial intelligence, deploying chatbots, copilots, and analytics tools designed to boost efficiency and improve decision-making. But as AI adoption matures, the conversation is shifting away from experimentation and toward operational impact.
“What’s a little different this year than last year is that retailers are getting, I would say, more serious about AI,” Kristin Howell, part of SAP’s retail industry product management team, told Supply Chain Management Review.
That seriousness reflects a growing realization: AI value does not come from isolated tools, but from embedding intelligence directly into core business processes supported by clean, real-time data and clear guardrails.
From AI features to AI architecture
Howell describes SAP’s retail strategy as one that intentionally combines applications, data, and AI into a single operating environment.
“My job is really to look across [rather than focusing on individual products], which is really important … because part of the SAP strategy’s really what we call combining applications, data and AI,” she said.
At the center of that approach is what SAP refers to as retail data products: integrated, live data entities that bring together information such as product catalogs, pricing, vendor data, inventory positions, and replenishment plans.
“These retail data products just help you combine and integrate your product catalog with your price master, with your vendors, your sourcing strategy,” Howell explained. “Think of that as kind of a common foundation of data products.”
Those data products then support AI agents that are embedded directly into workflows functioning inside the systems where decisions and execution actually happen.
“We really believe that from an AI perspective the real unlock for retailers is going to be connecting those agents, embedding those agents right at the heart of the decision, right at the heart of the execution,” Howell said.
Agents that act, not just analyze
One example is SAP’s order reliability agent, designed to proactively monitor customer orders and resolve fulfillment issues without requiring human intervention.
“I’ve accepted this order and customer says they want to pick it up in store,” Howell explained. “Do I have the stock in store? I had the stock available to promise when I took the order but … is it still available?”
If inventory discrepancies arise, the agent can identify alternatives such as shipping from a distribution center or transferring stock from another store and then execute that decision directly in the system.
“The agent’s got access to the right data,” Howell said. “It’s got the artificial intelligence to make smart decisions … but then also connect it to SAP from an execution perspective so that … the agent can actually take that next step.”
Why data accuracy matters more than autonomy
Agents, though, are only as reliable as the systems they depend on. If warehouse or store systems show inventory that is not physically available, AI-driven decisions can quickly unravel.
“The reason we’re so focused on embedding AI and embedding agents in the business process is to prevent or architect around some of these potential challenges,” Howell said. “Inventory is not just a fixed view from a week ago or last night. It’s inventory in the moment, real time.”
That level of granularity—knowing whether goods are on a truck, at the dock door, or received into store inventory—is essential to preventing AI from making decisions based on outdated assumptions.
“The key to successful agents is clean, robust, accurate data,” Howell said.
Build, buy, or blend?
Retailers increasingly ask whether they should build their own AI agents or rely on those provided by vendors. Howell’s answer: most will do both.
Understanding the data model, in either model, is the key. In the case of an SAP agent, it is able to handle what Howell described as the “blocking and tackling”—common workflows that drive efficiency across retail operations.
At the same time, retailers may want to customize agent behavior to reflect brand-specific priorities, allowing them to tailor the customer experience to things like loyalty and previous spending amounts.
In those cases, retailers can extend or adapt standard agents using their own customer data and business rules.
“You want to … spend your time investing in and developing the agents that drive differentiation or deliver your secret sauce,” Howell said.
Guardrails before autonomy
Despite growing interest in agentic AI, most retailers remain cautious, particularly when AI decisions involve pricing, procurement, or customer promises.
“They’re not quite yet taking the guardrails off to say, ‘oh yeah, go spend my money,’” Howell said.
Retailers are increasingly evaluating AI not just as a tool, but as an architectural decision that requires governance, compliance, and continuous data discipline.
“That agent is going to operate on the data you provided,” Howell said. “If your data isn’t clean … the agent’s going to think it’s doing the right thing.”
From curiosity to intent
Perhaps the most telling shift Howell sees this year is how retailers frame their AI conversations.
“Last year, retailers were asking us, what use cases do you see?” she said. “I see it turning a bit this year where retailers come to us and say, ‘Here’s what I’m trying to do.’”
That evolution from curiosity to intent signals that AI in retail is entering a more consequential phase. The next wave of value will not come from experimentation alone, but from disciplined execution, strong data foundations, and carefully designed guardrails that allow AI to act without introducing unacceptable risk.