Faisal Saeed is the Founder & CEO at Promptev Inc. At the forefront of AI’s next leap.
getty
The problem with enterprise AI isn't the technology. It's everything around it.
In the spring of 2013, a major U.S. airline experienced a widespread outage in its reservation system that grounded flights and disrupted operations for several hours. Public reporting attributed the incident to a technical failure during a system update. The disruption highlighted weaknesses in testing, rollback preparedness and operational resilience, and resulted in strained customer confidence during a critical period for the airline.
That became a case study in why even powerful, well-proven technology can fail without the right operational backbone. Today, as AI migrates from experiments to mission-critical operations, a similar fault line is opening quietly and is going largely unacknowledged by the organizations most exposed to it.
Most companies have poured their energy into asking which AI model is right for them. This is a reasonable question. But it is the wrong one to obsess over.
Based on my analysis of findings from McKinsey’s State of AI research and related industry discussions, fewer than one-third of organizations are seeing substantial business impact from their AI initiatives. The culprit was not the model itself. It was everything surrounding the model: how it received information, how its outputs were governed, how its behavior was monitored and how its decisions could be explained to a regulator or an auditing committee.
Right now, most enterprises have no good answer to any of those questions. And that is where things go wrong.
The failures tend to cluster around three basic problems.
The first is context: The AI does not know enough. Models are trained on enormous amounts of data but not your data. Not your policies, your products, your customers' history or your internal processes. So you feed it some documents, set up a few integrations and hope for the best. The result is an AI that sounds confident but is often working with incomplete or outdated information. It fills in the gaps the only way it knows how: by guessing.
The second is reliability. You cannot trust it to be consistent. You may receive two different answers if you ask the identical question twice under slightly different circumstances. That is fine for a chatbot helping you plan a vacation. It is a serious problem when the AI is recommending loan approvals, flagging compliance risks or drafting customer communications. Without a way to monitor and test its behavior, you are essentially flying blind. In regulated industries, that drift can carry legal exposure. In consumer-facing products, it erodes trust.
The third is governance. Nobody can explain what it did. Boards want accountability. Regulators want audit trails. Customers want to know how decisions about them are being made. But most AI deployments today are built to produce outputs, not explanations. When something goes wrong, and it will, there is no clean record of what the AI was told, what it decided or why. Layering it on afterward is expensive, partial and often unconvincing.
When these issues surface, the typical response is to bolt something on. Add a logging tool. Write a new policy. Assign someone to manually review AI outputs. These fixes buy time, but they do not solve the underlying problem.
The issue is structural. Most AI systems were designed to be powerful, not to be managed. The team that built the model is different from the team that handles compliance. The data pipeline was set up quickly and has not been properly maintained. Governance documentation exists somewhere in a shared drive but is disconnected from what the AI is actually doing day to day.
A shift is happening among the enterprises getting this right. They are not asking which AI to use as their first question. They are asking how to build an environment where AI can actually be trusted.
That means treating context, reliability and governance not as problems to fix later, but as things you design for from the start. It means having one place where the AI's knowledge is managed and kept current. It means being able to see exactly what the AI was given, what it produced and when. If something goes wrong, you can trace it, fix it and prove it will not happen again.
Think of it the way companies thought about cloud infrastructure a decade ago. At first, everyone just spun up servers and hoped. Then the smarter ones realized they needed a proper operating layer (monitoring, security, access controls and deployment pipelines) before they could scale safely. AI is at exactly that point now.
If you are a business leader making decisions about AI right now, here is the honest truth: The companies that will win are not necessarily the ones with the most sophisticated models. They are the ones who can deploy AI reliably, catch it when it drifts and explain it when they have to.
That advantage is still up for grabs. But the window will not stay open indefinitely. As regulations tighten and as early adopters prove what is possible, the gap between organizations with a real AI operating foundation and those without one will become very hard to close quickly.
The companies building that foundation now are making a bet. Not on a particular AI model, but on their own ability to use AI responsibly at scale. That bet is looking smarter every quarter.
AI does not fail because it is dumb. It fails because it is running inside systems that were not built to support it: bad information in, no one watching, no paper trail when things go sideways.
Fixing that is not glamorous work. It does not make for exciting product announcements. But it is the work that separates the companies who will look back on this period as a turning point from the ones who will look back on it as a very expensive lesson.
The question is not whether your AI is smart enough. The question is whether your organization is ready to actually use it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。