Supply chain teams with AI-ready staff report 71% AI project success, others 28%

Supply chain teams that call their workforce AI-ready report a 71% AI project success rate; everyone else reports 28%. That split comes from The People Paradox in Supply Chain AI Integration, released Oct. 14 by the Association for Supply Chain Management (ASCM) and the Digital Supply Chain Institute (DSCI). Only 17% of respondents put their workforce in the ready camp. The other 83% said their employees are unprepared to use AI effectively, and about 64% of that unprepared group is still in pilot or planning.

The ready group doesn't just succeed more often. According to Material Handling & Logistics, it has put AI to work across an average of six supply chain functions, so the gap shows up in how widely AI is used as well as in hit rate. The publication described the overall pattern as a high-interest, low-execution cycle: companies hold AI on a short leash until talent, trust and measurable business value catch up.

The timing matters because the products have moved faster than the buyers. FreightWaves noted that freight technology vendors are already selling AI agents that can benchmark rates and award freight under rules a customer sets. That suggests the team who will supervise an AI agent could belong on the evaluation list alongside the agent itself. For a supply chain chief weighing whether software should award freight without someone signing off first, that could shape the decision.

ASCM CEO Abe Eshkenazi thinks the industry has the order backwards. Most AI sessions at supply chain conferences, he told FreightWaves, run through the challenges, the implementation and the benefits, then close with a reminder not to forget about the people. "From our perspective that's where you ought to start the conversation," he said.

Planning gets the AI, execution waits

The report scored AI use across 13 core supply chain functions on a five-point scale. A 1 means no AI, and a 5 means autonomous execution with no human review. The average came in at 1.94. Planning scored highest. Execution-heavy work trailed, and returns and reverse logistics landed at the bottom.

Even the leader sits closer to no AI than to autonomy. The report calls this keeping AI on a short leash, with predictive tools used as little more than digital spreadsheets. Eshkenazi made the same point in plainer terms: companies use AI for analysis and prediction, not for doing. He called it "more of a glorified spreadsheet."

He does see room for AI to take routine transactional decisions within set parameters. People would keep the high-risk calls on customer commitments, sourcing and major disruptions. AI's authority needs to grow, he said, but confidence in the technology isn't there yet. A freight-award agent working under customer-set rules sounds a lot like his routine, within-parameters case, which suggests the first execution handoffs could come in transactional work like that, not in sourcing strategy.

Skills and payback tie for first

An unclear business case or return tied for first among barriers with a lack of workforce skills and resources, each cited by 33%. Poor data quality came next at 25%. Executive skepticism didn't make the list of main hurdles. Material Handling & Logistics summarized the report's four business hurdles as an unready workforce, reluctance to trust AI to act on its own, cybersecurity concerns and trouble proving return on investment.

Eshkenazi compared the moment to ERP, which was also supposed to remake supply chains. A system can't repair a broken process or bad data, he argued. That's why he recommends value stream mapping and data analysis to find and fix weak spots before a project starts. The lesson carries over. A team that skips the process work is buying a faster way to repeat its current mistakes.

Measurement is the other half of the payback problem. Eshkenazi called headcount reduction too crude a measure of AI's return. He'd rather track response time: how quickly a company recognizes a disruption, weighs alternatives and acts. A system can't be held accountable for a decision, he added, so a named person has to own it.

DSCI managing director Marko Kovacevic made a similar point in a statement. "Leaders have accepted AI in principle," he said. "What most organizations haven't built yet is the capacity to execute it in practice." With tariff shifts, demand shocks and geopolitical disruption, he added, the companies that sense and re-plan fastest will pull ahead, which makes AI readiness a management problem before it's a technology one.

Respondents' own expectations support that view. Material Handling & Logistics reported that over the next two years, 46% expect the biggest payoff to come from faster decisions through shorter planning cycles, 42% from lower overall supply chain costs and 40% from higher process automation. Leaders who rate their AI deployments as "somewhat successful" make up more than half (55%), a score that points to isolated wins and not broad operational change. A finance team that judges AI only on headcount would miss the gain respondents expect most.

Strategy stays in the C-suite

Involvement falls off fast below the top floor. Of C-suite respondents, 82% lead, implement or scale AI initiatives. At the director level, only 12% lead AI work directly, and half of directors say they're merely aware of it. Among managers, 51% are only aware of AI work, and among staff that figure climbs to 70%. These are the people who would set an agent's parameters and handle its exceptions.

There's a fair objection to the headline split, too. Readiness is self-reported, and a company that rates its workforce ready may also have cleaner data, bigger budgets or simpler operations. The survey shows the link between readiness and success. It doesn't show which one causes the other. Still, the habits the report ties to progress don't depend on budget. In his statement, Eshkenazi said those organizations treat readiness as part of the technology investment, learn by doing, bring managers in early and build trust one function at a time.

The broader numbers in Material Handling & Logistics show how big the gap is. Some 91% of organizations plan to use AI to augment workers, yet only 14% say they're fully prepared for an AI-augmented future.

The training ground AI is replacing

Few respondents expect AI to take jobs outright. Only 7% predict direct replacement, 51% expect people to move toward higher-value decisions and 40% expect new roles to appear. The second figure comes with a cost Eshkenazi spelled out for FreightWaves. In the past, supply chain professionals developed judgment by doing the work themselves: running analyses, building reports and working through spreadsheets. That hands-on work trained them to catch exceptions and to question whether the numbers looked right.

AI now does much of that groundwork. In Eshkenazi's view, that pushes people into decision-making roles before they've learned the fundamentals. He called powerful tools combined with staff who can't critically assess the output a dangerous mix for any organization. It's an uncomfortable twist. The tasks that get automated first are the ones that used to train the people who now have to supervise the automation.

The MIT Center for Transportation & Logistics described the same squeeze in a May essay by Mackenzie Berry. According to the essay, people working in supply chains must now judge AI tools, make sense of what models produce and know where the data falls short, and many have no computer science background. It cites the World Economic Forum, which expects nearly 40% of core job skills to change by 2030. The Forum also found that 77% of employers intend to make upskilling their workforce a priority as AI transforms their businesses.

The essay also cites an MIT CTL study arguing that generative AI is set to shape decisions in at least 13 major supply chain and operations management domains. Those domains include forecasting, logistics, procurement, inventory management and risk assessment. Compare that reach with the survey's 1.94 average, and the tools could clearly touch far more than companies currently allow them to.

Berry's goal isn't to turn every supply chain professional into a software engineer. It's to build organizations where managers can work fluently alongside AI, ask informed questions about implementation risks and spot where the technology creates measurable value. As autonomous and semi-autonomous systems spread, she wrote, AI literacy will become less a specialty skill and more a baseline business competency.

MIT CTL is offering The AI-Driven Supply Chain, a five-day training focused on practical implementation rather than abstract theory, covering forecasting, optimization, generative AI, computer vision and autonomous agents in real supply chain settings. For operations leaders whose junior analysts once learned the business by building reports, formal training may have to stand in for that lost apprenticeship.

Oversight comes before autonomy

Trust is the other gate. According to Material Handling & Logistics, 91% of leaders want visibility into how AI models produce predictions and recommendations. Another 43% worry about algorithmic bias in supplier selection or inventory allocation, which are exactly the decisions an execution agent would touch. Security controls show a similar gap between concern and practice.

Eshkenazi traced the exposure to the visibility push after the pandemic, when companies started connecting suppliers, carriers and brokers on shared platforms. Much of the weakness, he said, sits with tier-three and tier-four vendors that have few resources of their own. An agent allowed to award freight connects to that same network of partners. That suggests operators could treat its security controls and its authority as one decision, not two.

The figure to watch is the 17%. If more organizations come to call their workforce ready, that could show up in execution scores such as the 1.59 for reverse logistics. If vendors keep releasing execution agents faster than that share grows, more of the industry's AI could stay in planning.