WW/SUPPLYCHAI
AI supply-chain projects face five barriers to operational value
Five barriers can derail AI in freight and supply chains: poor data, broken workflows, legacy integration, employee resistance and missing strategic ownership.
- By
- Elena Vasquez
- Filed
- Length
- 967 words
- Read
- 5 min

Key points05
- Five recurring barriers undermine AI adoption: unreliable data, poorly defined processes, legacy integration, organizational resistance, and weak capability or strategic alignment.
- AI applications include demand forecasting, inventory optimization, transportation, procurement and risk management.
- Incomplete sales records can cause stockouts and lost sales, while inaccurate inventory can increase carrying costs and excess or obsolete stock.
- Many companies still use legacy software described as decades old, while integration can require custom interfaces, specialist expertise and significant modifications.
- Ana Perdomo has about 20 years of logistics experience across Canada, Mexico, the United States, China and Argentina.
By Elena Vasquez
Five implementation barriers can stop an AI supply-chain project before it produces measurable operational value. Consider a manufacturer that buys predictive-maintenance AI for production machinery; without compatible sensors, cloud connections and real-time data, the system cannot identify equipment failures before they occur.
What actually blocks AI supply-chain projects?
The proposed applications span demand forecasting, inventory optimization, transportation, procurement and risk management. Amid growing complexity and volatility, AI promises better decision-making, efficiency and responsiveness, yet easier access has not made implementation easier.
Five recurring barriers emerge:
- Unreliable sales, inventory and operational records
- Poorly defined workflows and informal communication
- Legacy equipment and software that cannot exchange data
- Employee resistance and unclear decision ownership
- Limited internal expertise and weak strategic alignment
Most of these obstacles concern foundations rather than algorithmic sophistication. Ana Perdomo, an assistant teaching professor of Operations and Supply Chain Management at the University of Lethbridge, has about 20 years of professional experience across Canada, Mexico, the United States, China and Argentina. Her conclusion is direct: AI cannot substitute for strong process, data and system foundations.
What does bad data cost?
Machine-learning models for demand forecasting, inventory routing and risk management need massive volumes of clean, accurate, real-time and historical data. Legacy systems, fragmented spreadsheets and siloed departments often prevent procurement data from communicating with warehouse data.
Perdomo captures the risk in the principle "garbage in, garbage out." When companies feed messy or incomplete records into an AI system, the technology amplifies bad decisions at scale instead of correcting underlying inefficiency.
An incomplete sales history can make a forecasting model underestimate demand, causing stockouts, lost sales and dissatisfied customers. Conversely, inaccurate inventory records can trigger unnecessary replenishment, raising carrying costs and the risk of excess or obsolete stock.
When companies connect AI recommendations to automated decisions, one flawed assumption can influence multiple operational activities simultaneously. That makes data governance and cleansing a prerequisite, rather than a later technical cleanup.
Can a model fix a broken workflow?
AI-powered transportation planning can optimize delivery routes and reduce fuel costs, but dispatch conditions determine whether it works. If dispatchers follow inconsistent procedures, drivers receive last-minute informal messages and priorities remain unclear, the model can produce inefficient routes that fail actual needs.
Those routes can increase transportation costs and customer dissatisfaction despite the investment. Technology accelerates a broken process; it cannot cure one.
Firms that adopt trendy tools to reverse declining performance without mapping and optimizing baseline workflows merely digitize flawed operations, including manual guesswork and ad hoc communication. Before automating, companies should standardize dispatch procedures, clarify responsibilities and establish reliable communication protocols.
Can legacy equipment connect to AI?
Modern AI requires cloud computing, integrated ERP systems and APIs, yet many companies still depend on rigid, decades-old legacy software. Connecting old and new can prove difficult, expensive and risky, which stalls or ends digital-transformation projects.
If production machinery runs software that cannot communicate with modern sensors or cloud platforms, predictive-maintenance AI cannot access real-time equipment data. Bridging the gap may require custom interfaces, specialized technical expertise and significant modifications, which can delay implementation, raise costs and disrupt critical production systems.
Companies therefore need not replace every legacy system. They should identify the data a specific use case needs, test whether existing systems can supply it reliably and estimate integration costs before committing to modernization.
Why do employees bypass AI tools?
Technology adoption remains a human challenge. Employees and middle managers may resist because of low digital literacy, fear of job displacement or a preference for experience-based judgment over algorithmic recommendations.
Without a culture of data-driven decisions and continuous learning, workers may bypass AI tools, rendering the investment useless. In a logistics example, experienced managers may ignore demand forecasts and resource-planning recommendations in favor of manual methods, preventing the company from realizing the technology's benefits.
Organizations should train employees, explain AI's purpose and limits, involve users in implementation and create clear lines for challenging recommendations and making decisions.
Who must own the model?
Successful AI adoption also requires data scientists, supply chain analysts who understand technology and cross-functional leadership. Many firms buy off-the-shelf software expecting it to run itself, but without internal talent to interpret outputs, customize models and govern them, AI becomes a "black box."
Perdomo warns that opaque conclusions erode trust and can lead teams to abandon the technology. In a supplier-risk example, a company may struggle to select relevant risk indicators, validate recommendations or determine how results should shape sourcing decisions without specialists who understand both AI and supply-chain risk.
If the tool does not fit procurement strategy and risk tolerance, it can produce technically sound insights that are commercially irrelevant. Technical expertise must therefore combine with industry knowledge and explicit strategic objectives.
What should supply-chain teams fix first?
The implementation sequence begins with auditing and cleansing data sources, then establishing rigorous governance protocols. Companies should map and optimize baseline workflows, fix fragmented operations before automation, assess the move toward cloud-native infrastructure and fund continuous training.
They also need internal talent or cross-functional leadership so people can interpret models rather than treat them as uninterpretable systems. For shippers and logistics providers, this order limits the risk that inaccurate demand, inventory, route or supplier signals will amplify stockouts, excess stock, higher transport costs or poor sourcing choices.
It also gives dispatchers and managers a practical way to validate outputs, assign decision responsibility and use AI without surrendering operational control. Perdomo's forward view is that intelligent supply chains will depend as much on organizational readiness as on technological advancement. As AI becomes easier to access, clean data, standardized work, reliable integrations and accountable decisions will determine whether the technology produces operational value.
Source: scmr.com
More from Elena Vasquez
Show full bio
News editor covering people, launches and day-to-day practice in Ocean and air freight, container shipping at Waybill Wire.
956 articles
Related05
AI to Orchestrate Procurement in 30% of Firms by 2030, Gartner Says
83% of supply chain leaders say their workforce can't use AI
57% expect autonomous AI for over a quarter of warehouse decisions
AI Spending Still Leaves Tariff Management Exposed
AI to handle more than a quarter of warehouse decisions by 2028