This audio is auto-generated. Please let us know if you have feedback.
Artificial intelligence initiatives don’t always result in supply chain success.
Starbucks’ recent decision to ditch an AI inventory management tool roughly nine months after implementation illustrates the point. While the system was meant to speed up workers’ ability to take stock using computer vision to automate aspects of inventory counting, Reuters reported in May that the platform occasionally miscounted or mislabeled items.
Some stumbles are to be expected as companies figure out how to most effectively leverage AI within their operations. Fifty-six percent of chief supply chain officers find integrating AI with legacy systems and processes to be a large hurdle, per a Gartner survey released in April. Fifty percent said they had limited internal expertise and talent to manage the technology.
Organizations learning how to manage and implement AI is “compounded with just week-over-week, month-over-month improvements in AI capabilities,” Brian Pacula, a partner in West Monroe’s supply chain practice, told Supply Chain Dive in an interview. “So what you knew last week is already changing going into next week, so you just got to stay on top of it.”
Finding and scaling the right AI tools in this environment can sound daunting. But supply chain leaders can lean on a few guiding principles to more effectively leverage AI within their operations, experts told Supply Chain Dive.
Apply AI in the right areas, with the right data
AI adds the most value in situations where the pace of decision making exceeds human capacity, according to Alan Amling, an assistant professor of practice at the University of Tennessee, Knoxville.
“The question is not ‘where can we use AI?’ It is ‘which decisions in our supply chain are made frequently, under time pressure, with data the organization can actually see?’” Amling, a former VP of corporate strategy for UPS, said in an email to Supply Chain Dive.
Supply chain executives should target areas where decisions are made frequently, be it on an hourly, daily or weekly basis, for potential AI implementation, Dheera Anand, a partner at Bain and Co. focused on supply chain strategy and transformation, said in an interview.
This could include frequent tasks like ordering or inventory deployment, rather than longer-term projects like network modeling, Anand said. For Tractor Supply Co.delivery route building has been a regularly occurring task the retailer is using AI tools to help with, freeing up drivers’ time in the process.
A Tractor Supply Co. storefront. The company has been using AI to assist in its delivery route-building process.
Courtesy of Tractor Supply
Executives should also consider how variable a particular environment is before deploying AI, and how that can impact a company’s ability to produce and distribute products, Anand said. For example, an older manufacturing facility in which there is equipment downtime every week may not be an ideal place to use AI.
Beyond decision frequency, data availability is another important factor to consider before pursuing AI for a particular supply chain use case. Supply chains can be a challenging area to collect data, since organizations are often dependent on their suppliers and logistics partners for critical information, Anand said.
Amling agreed that sufficient data is necessary for whatever decision an organization is automating. If that data isn’t available, the organization needs to address their visibility issues before pursuing AI model usage, he added.
Pilot and deploy AI smartly
Companies often tout instances of successful AI pilots in their supply chains, but early tests can be rife with stumbling blocks, according to Anand.
“I’ve had like at least three conversations in the last week on failed pilots that didn’t deliver value,” she said.
To better position a pilot for success, Anand recommended organizations figure out if the necessary data for the test is available, the return on investment can be demonstrated and employees are comfortable leveraging the AI tool. Employees can also provide useful context for the AI models to learn from.
“Your planners, your inventory managers, they just know things,” Anand said. “There’s so much tribal knowledge that’s not codified anywhere, so you really have to think about codifying that knowledge as you build these models.”
Executives should also determine the pilot’s end goal and know what signals are needed before the company begins to scale the AI tool more widely, Anand added.
“No company would hire a planner, skip onboarding, assign no manager, set no performance expectations, and then fire that person nine months later for underperforming. Yet that is exactly how most organizations deploy AI.”
Alan Amling
Assistant professor of practice at the University of Tennessee, Knoxville
For Toray Industries (America), the company is piloting AI agents to review suppliers’ pallet designs before potentially expanding the tool more widely. If the agent can provide insights on 60% to 70% of bids, the company would save significant time, John Eustis, SVP of U.S. group procurement for the company said during the Institute for Supply Management World 2026 conference in April.
Organizations may be quick to scrutinize an AI tool itself when a deployment struggles, rather than looking at other influencing factors such as workflows surrounding it, Amling said. Instead of treating AI like software, he said companies should approach it as if they were bringing a new employee up to speed.
“No company would hire a planner, skip onboarding, assign no manager, set no performance expectations, and then fire that person nine months later for underperforming,” Amling said. “Yet that is exactly how most organizations deploy AI.”
Know when to move on
If an AI tool continues to fall short of human baseline performance even after making adjustments, or the data foundation it needs doesn’t exist and can’t be built cost effectively, it may be a sign to scrap the initiative, Amling said.
He added that organizations should consider using the “left-on test” to determine if the effort to maintain the AI tool is worth it. The test consists of one question: Would you leave the AI system running unsupervised over a weekend?
“If the answer is no, and it has been no for months with no credible path to yes, you are not running AI,” Amling said. “You are running an expensive suggestion engine, and it is fair to ask whether the juice is worth the squeeze.”
Monitoring key performance indicators can help supply chain leaders determine if an AI application is benefiting the organization, or if the tool needs to be removed, West Monroe’s Pacula said. AI could impact supply chain KPIs such as forecast accuracy, inventory turnover and logistics costs, per a 2025 report from software company NetSuite.
Steering committees can also help determine if the AI initiative’s return matches the effort the organization is putting in, Pacula added. However, organizations should ensure the committee’s members have enough time on their plate to make thoughtful recommendations and properly prioritize the initiative.
“You may need to get through the backlog of everything else you have going on, right?” Pacula said. “If you have an ERP implementation or you’re going through an acquisition, [it’s] probably not the time to overload your plate with AI.”
