The supply chain world is in the grip of an artificial intelligence fever dream. From generative AI negotiating with suppliers, to machine learning models predicting disruptions weeks in advance, the promise feels limitless. Boardrooms are captivated; pilot programs are multiplying, and the pressure to "do something with AI" is immense.

Yet behind the headlines, a more troubling story is unfolding. Industry research consistently shows that the vast majority of AI projects never make it past the pilot phase. The AI models are powerful. The flaw lies in what's underneath. Companies are trying to build the penthouse of AI on a foundation of sand.

Foundational systems are the digital and operational backbone of any supply chain. They’re the core transactional and record-keeping engines that run your daily operations. Think of them as the central nervous system of your business. In simple terms, they ensure that you know what you have, where it is, where it's going, and what it costs.

Concrete examples include your enterprise resource planning system, which integrates finance, procurement and production; your warehouse management system, which dictates how inventory is received, stored, picked and shipped; and your transportation management system, which plans and executes freight moves.

But foundational systems are more than just software. They encompass clean data infrastructure, where a customer record means the same thing in every database, and standardized processes, where a "perfect order" is defined consistently across the globe. Without these elements working in harmony, you have a collection of fragmented activities. They’re the absolute backbone that determines whether you can deliver on a promise to a customer.

Why AI Needs Strong Foundations

At its core AI is a pattern-recognition engine that learns from historical data to predict future outcomes. This makes it radically dependent on one thing: clean, structured, reliable data. The elegant mathematics of a machine learning model can’t compensate for a messy data reality. If your foundational systems are capturing incorrect inventory levels, duplicate supplier information or inconsistent lead times, the AI will learn those errors and amplify them at scale, only much faster than any human could.

This is the classic principle of "garbage in, garbage out." Real-world examples are plentiful. One North American manufacturer invested heavily in an AI demand forecasting tool, only to discover its forecast accuracy was worse than its old spreadsheet-based method. The post-mortem revealed that its ERP system held years of incorrect customer delivery dates, causing the AI to learn that late shipments were actually on time. In another case, a European retailer built an AI replenishment engine on top of a WMS that had a 20% error rate on actual on-hand stock. The AI, acting on the data it was given, confidently ordered more of products that were already overflowing in the backroom and zero of the items that had just sold out.

In both cases, the AI wasn’t broken; it was simply reflecting the broken reality of the data it was fed. The models failed not because its foundation was a lie.

Common Mistakes Companies Make

The path to AI failure is paved with good intentions, but a recurring set of mistakes almost always guarantees it. The most common and fatal error is jumping to AI before fixing basic operational problems. A company with a chronically late delivery performance needs process improvement to stop it.

Linked to this is a willful ignorance of data quality issues. Leaders assume that because data lives in a modern cloud system, it must be accurate. They don't invest in the hard, unglamorous work of data cleansing and governance. Another classic mistake is expecting AI to magically glue together systems that were never integrated. An AI model can’t create a single view of a shipment if your ERP and TMS have no real-time connection and data sits in silos. Finally, the most dangerous assumption of all is expecting AI to fix broken processes. If a planner manually overrides every system-generated purchase order because they don't trust the master data, an AI recommendation engine will simply be ignored. Automating a bad process just gets you to the wrong answer faster.

What Must Be in Place Before AI

Before you write a single line of Python or issue an RFP for an AI solution, five prerequisites must be solidly in place.

First, you need clean, connected and governed data, where a single source of truth exists for your customers, suppliers, items and financial hierarchies.

Second, you require integrated core systems — your ERP, WMS and TMS must talk to each other seamlessly and in near real-time, creating a digital thread you can trust. 

Third, your core business processes must be standardized and stable. AI can’t work effectively if every site defines "on-time delivery" differently.

Fourth, your workforce needs to be AI-ready. This means having planners, buyers and logistics managers who understand data-driven decision-making and are trained to work alongside algorithmic insights, not against them.

Finally and critically, you need clear, measurable business goals. Are you trying to reduce inventory by 15%? Increase forecast accuracy by 20%? AI without a defined business case is just an expensive science experiment.

Foundation First, AI Second

The companies that win with AI in the supply chain follow a simple, disciplined mantra: Fix the foundation, then layer AI on top. This is a step-by-step journey. Think of it as building a house. You don't install a smart home system before you've poured the concrete and erected the walls.

This requires a profound mindset shift. AI must be seen not as a fixer of broken processes, but as an enhancer of strong ones. Its role is to find the subtle optimizations, the complex multi-variable trade-offs that are invisible to even the best human planner, operating on a rock-solid, standardized and digitized operating model. This approach requires patience and a long-term success mindset. The ROI may not appear in the first quarter. The initial investment goes into data cleanup and system integration work that feels slow and unglamorous. But this is the only path where AI consistently delivers a step-change in value rather than becoming another shelfware statistic.

The lure of AI is powerful, promising a supply chain that thinks, predicts and adapts on its own. But the majority of AI projects fail from a lack of foundational integrity. Clean data and integrated systems are the non-negotiable price of entry for a successful AI journey. Before you ask what AI can do for your supply chain, ask if your supply chain is ready for AI. Invest the time, budget and executive focus to fortify your foundation first. Only then will the intelligence you build on top be real, lasting and transformative.

David Anderson is senior supply chain strategist at SCM Champs.