Technology often arrives with a promise larger than the power of the tool itself. A new tech stack simplifies the work, and a new platform connects the enterprise. A new dashboard creates visibility, and a new algorithm sharpens decision-making. AI has entered the same room with the loudest promise yet: faster analysis, cleaner forecasting, automated service, content on demand, and fewer constraints on human capacity. AI—in its various forms—promises power, but does the organization understand how to use it? Is it ready to be changed by it, or just eager to say that it launched a shiny new tool?
What’s an unanticipated risk of dependence on AI? And the answer is: AI can amplify retail dysfunctions and accelerate problems before you’re aware of it.
The Diagnosis Problem
The allure of the AI promise is real. Individuals are writing, analyzing, summarizing, and connecting faster than they could a few years ago. Consumers have shown increased levels of adoption. But at the enterprise level, the use cases are more complicated. Many retailers believe that once AI is adopted, integration and transformation will automatically follow.
Technology does not instantly make organizations better; it amplifies what they already are. Yes, they become more efficient, measurable, scalable, and often more profitable. But the key to paying off AI’s promise is whether the organization is ready for it. If the organization is aligned, technology amplifies the alignment. If it is confused, technology amplifies the confusion. If incentives reward collaboration, the tool strengthens collaboration; if they reward turf protection, the tool gives every function a faster way to defend its turf. If the data is clean and governed, AI helps people see patterns; if it is fragmented and poorly owned, AI exposes the mess at a mind-boggling speed.
As a result, many AI transformations disappoint not as technology failures but as misdiagnoses and human failures. New systems cannot replace bad processes to compete for success. When owning a process matters more than solving a problem, and the stated mission is not the rewarded behavior, AI cannot correct these dysfunctions, it only amplifies and accelerates them.
The solution often arrives before the problem is properly diagnosed. The evidence now bears this out. In BCG’s 2025 study of more than 1,250 firms, only about 5 percent, the ones it calls “future-built,” were generating AI value at scale. Roughly 60 percent reported little to no material benefit despite real investment, and the remaining 35 percent were scaling but admitted they were not moving far or fast enough. BCG’s own diagnosis reads like a restatement of the problem: Too many companies approached AI incrementally, as a way to do more of the same a little faster, when the work required was reinvention. Nearly nine in ten of the value-generating firms expected most of their AI value to come from reshaping business processes, not from the tools themselves. The winners are not adopting AI; they are redesigning around it.
Rewarding the Launch
Somewhere a decision is made that the company needs AI, a new platform, or a particular system before anyone has diagnosed what is actually broken. The temptation is understandable: A tangible tool is easier to approve, budget, and announce than an ambiguous problem is to solve.
AI often gets layered onto an existing process, on the assumption that the new tool will run that process faster. But is that process productive in the first place? Maybe instead of a new AI tool, the process should be reimagined, but we all know that reimagining is slow, disruptive, and politically expensive, while bolting a new tool onto the current workflow gets rewarded. So, the technology accelerates and amplifies a process no one stopped to question. The AI launch becomes the win and a headline. The organization rewards the launch, not the outcome. The enthusiasm for AI doesn’t address the more fundamental question of what the company should become to sustain success.
Retail Misfires
Target Canada is a well-documented cautionary tale of a multi-causal operating-model failure that was impossible for software to hide. The brand’s expansion to Canada moved faster than the supply chain, pricing logic, and, above all, the data were ready to support. What happened? Inaccurate inventory and pricing data flowed into the system, then breaking replenishment, and customers were met with unforgiving, empty shelves. The tech system did not cause that breakdown; it executed faithfully on a foundation of human failure.
In September 2025, Starbucks rolled out an AI inventory tool, built with its vendor NomadGo, across the 11,000+ company-owned North American stores as part of an effort to improve product availability and fix chronic shortages. Nine months later, the tool was retired; it miscounted and mislabeled items, sometimes failing to recognize a product sitting directly in front of the camera, and baristas ended up recounting by hand what the system had already counted. The enthusiastic launch promised speed and accuracy the tool could not deliver in real store conditions. The AI did not fail in isolation—it was deployed at scale before anyone confirmed it worked, and instead of fixing the shortage problem, it revealed the disorder faster than the organization could absorb it.
Amazon is a counter-case. Its robotics story is not “Amazon bought robots.” It presciently acquired Kiva Systems in 2012 for $775 million and then spent more than a decade progressively redesigning fulfillment around automation of storage logic, picking flow, routing, safety, facility layout, and the human-machine interface. By mid-2025, it had one million robots and was running a generative AI model, DeepFleet, to help coordinate the fleet. AI leverage didn’t come from the number of robots; it came from redesigning the work around the capability, instead of bolting the capability onto the old work.
Deploying AI for Results
Before deploying AI, retailers should ask the hard question: Are we trying to solve a real problem or are we keeping up with the trends? These are distinctions leaders need to make before their next AI initiative. The question is not “where can we use AI?” That one is too easy, and it produces pilots, demos, and scattered productivity gains. The better question is “what work should change because this capability exists?”
Retailers need to resist the urge to solve an operating-model problem with an AI strategy label. They do not need more demos and dashboards; they need to ask the right questions and be sure they are measuring what matters. Technology can absolutely make an organization better, but only in service of deliberate redesign. Without that, AI simply does more of what the organization already is—just faster.
AI is the new language of retail. But a tool cannot answer a question no one has asked. Before the next initiative, the discipline is not deciding where AI fits—it is diagnosing what is actually broken and being honest about whether a new system will fix it or simply scale it. The worst decision a leader can make is not buying the wrong tool. It is buying any tool before asking what was broken in the first place.
