Every company has a stack of small IT problems nobody ever fixes properly: an old server nobody officially owns, a workaround that quietly became permanent, a support contract that lapsed years ago. New research suggests this stack of ignored problems has grown large enough to be a genuine financial threat. Adding AI automation on top of it makes things worse, not better.
A recent survey by the Uptime Institute found that more than half of companies said their last major computer system outage cost over $100,000, and one in five said it cost more than $1 million. That has held at roughly the same level for two years running. This is not a rough patch, it is a pattern.
The pile of unresolved IT problems sitting behind these outages is large. One widely cited estimate puts the yearly cost of this kind of backlog at over $2 trillion in the US alone. A separate global study, based on scanning billions of lines of company software, calculated it would take 61 billion working days of programmer time to clear everything companies have let slide.
The Southwest Airlines meltdown in December 2022 shows what happens when this catches up with a company all at once. An aging crew scheduling system that Southwest had flagged internally as fragile for years finally broke down under a winter storm. The result was thousands of canceled flights, millions of stranded passengers, and a bill that ran past 800 million dollars.
It was not one broken part failing. It was years of postponed fixes all landing at the same time.
Now companies are adding AI into this same mess. Many are letting AI tools read IT tickets, flag problems, and in some cases take automatic action to fix things without a person checking first.
But an AI tool can only work with the information it is given. If nobody has recorded who owns a system, what depends on it, or why a workaround exists, the AI has no way to know either. It will simply act on incomplete information faster than a person would.
The US National Institute of Standards and Technology has flagged this exact problem in its guidance on AI risk. It warns about people trusting an AI's decision too much even when the AI is working with gaps in its knowledge.
This is not just an IT department problem. A retailer using AI to manage store systems, an insurer using AI to help process claims, a manufacturer running AI on plant equipment, all of them are handing decision power to a tool that inherits every unresolved problem already sitting inside their systems.
Before adding AI on top of existing systems, companies need to know which old systems are undocumented, unowned, or held together by a workaround. AI will not find that gap on its own. It will just move through it faster.