The Four Anti-Patterns That Are Killing Your AI Deployment

In two years of studying enterprise AI deployments across supply chains, I've catalogued a lot of ways these projects go wrong. But when I ran the analysis across 202 case studies, four failure patterns showed up again and again — in supply chains, in healthcare, in financial services, in IT operations. They're not industry-specific. They're structural. And if you can recognize them early, you can stop them before they cost you the project. 

The first is what I call the Trust Deficit. This is the black-box problem, and it's more dangerous than most leaders realize. When an AI agent makes a recommendation that a planner can't explain — can't trace back to a cause, can't defend to their manager, can't rationalize against their own operational experience — the rational response is to ignore it. This isn't technophobia. It's professional self-preservation. The planner's name is on the outcome. The algorithm's name isn't. I've seen this pattern kill S&OP AI deployments repeatedly. The agent is producing accurate outputs. The team simply stops looking at them because they can't explain why the outputs are what they are. Explainability isn't a nice-to-have feature. It's the condition under which adoption is even possible. 

The second is Legacy IT Entropy, which I sometimes call brownfield debt. This is the physical infrastructure problem — 30-year-old PLCs on a plant floor that have no digital interface, ERP systems that predate APIs, equipment that generates data only on a batch basis if at all. You cannot build a real-time AI system on top of infrastructure that doesn't produce real-time data. I've watched organizations spend millions on AI platforms and then discover that the operational environment they're trying to optimize is essentially invisible to the technology. The investment wasn't wrong. The sequencing was. Digital infrastructure has to come before AI, not alongside it. 

The third anti-pattern is the Data Black Hole, and it's most severe in returns and reverse logistics but shows up everywhere. A data black hole is a gap between physical reality and digital record — a situation where the true state of something (inventory location, product condition, supplier capacity) isn't captured in any system until a human physically inspects it. An AI agent making decisions based on system data in a data black hole environment is making decisions based on fiction. The most common version I see in supply chains is ghost inventory — the system says there are 500 units in a location, and there are actually 380, and nobody knows the difference until there's a stockout. No AI can solve a data problem. It can only amplify it. 

The fourth is Algorithmic Collusion, and it's the one most supply chain leaders haven't thought about yet but will. This emerges in multi-agent environments where separate AI systems, each optimizing independently for its own objective, begin to coordinate in ways that weren't intended and that create systemic risk. The clearest example from the research comes from financial services, where algorithmic trading systems were found to be tacitly fixing prices without any explicit coordination — each agent learning that certain behaviors were rewarded and converging on them independently. In supply chains this risk is most acute in dynamic pricing and logistics bidding, where multiple carrier or supplier agents operating in the same market can converge on equilibria that look like collusion even without intent. As agentic AI becomes more prevalent in supply chain operations, this is a regulatory and governance risk that organizations are almost entirely unprepared for. 

What these four anti-patterns have in common is that none of them are technology failures. The Trust Deficit is a design and communication failure. Legacy IT Entropy is an infrastructure sequencing failure. The Data Black Hole is a data governance failure. Algorithmic Collusion is a governance architecture failure. In every case, the AI is doing what it was designed to do. The failure is in the organizational and environmental conditions around it. 

This is why I've become increasingly skeptical of AI readiness assessments that focus primarily on the technology stack. The technology is almost never the constraint. The constraint is in the three layers that surround it — the organization, the data infrastructure, and the operating environment. Fix those first, and the technology will work. Skip them, and even the best AI in the world will quietly underperform until the project gets cancelled and everyone wonders what went wrong. 

If you recognize any of these patterns in your own organization, that recognition is actually valuable. It tells you exactly where to focus before the next deployment. The organizations that get AI right aren't the ones with the best technology. They're the ones that do the unglamorous diagnostic work first. 

Brad Rogers is a Director at PepsiCo Beverages North America and a DBA candidate at Fairfield University researching agentic AI adoption in enterprise supply chains. He is the founder of ChainLytix, an AI readiness advisory practice.