Companies are pouring billions of dollars into enterprise AI, but inadequate data, weak infrastructure and lack of strategy continue to kill consequences.
Getty
Crossing the boardroom Enterprise AI It has become the biggest item in the innovation budget, but it is also the source of greatest anxiety. Companies deploy large-scale models, generation assistants and forecasting systems at a record-breaking pace, but many results have not been maintained. Behind the PR and pilots there is a growing sense that something is broken across the industry. AI was supposed to unlock new values, but for many, it just adds a layer to the layer of noise.
AI critics such as cognitive scientist Gary Marcus and technology columnist Ed Zitron continue to question where the true moats of AI companies like Openai and humanity are, and Zitron commonly describes generative AI as “general AI.”Financial, ecological, social time bombs“Back in February. For Marcus, it's possible to build a truly amazing AI system, but it's not the mainstream model or approach today. He now has a large-scale language model. Fraud, unpredictable, potentially dangerous.
Andrew Florey, CEO Data axleI think the main problem begins before a single code is written. “The performance gap for Enterprise AI is not surprising. This is what happens when ambition is ready,” he said. “Many companies are investing in AI like products rather than capabilities, and expecting to be able to unlock the switch instantly. But AI doesn't work in vacuum. It's a high-performance engine, and many people are trying to do it with dirty fuel.”
Ai with no direction
Frauley didn't write words when he asked him the main reason for the issue. “The real problem isn't technology itself, it's the foundation,” he told me. “Companies are obsessed with models while they ignore or lack the dependency of models: data.” Fragmented records and siloed systems are the default conditions for most companies. AI exposes these fractures faster and more on a larger scale.
“Some brands, blinded by the possibilities and possibilities of AI, are rushing to deploy immediately. Data InfrastructureHe explained. “The most important steps, such as establishing ownership of data, building governance into workflows, and implementing quality standards, are often pushed aside due to speed.”
But that, according to Frawley, always brings misfires that undermine trust. “Imagine a brand submitting a 'customer loyalty' promotion to someone who just filed a complaint of a major service,” he explained. “AI, which lacks important information about recent negative interactions, makes a self-confident but deaf mistake.”
He is a global view. Udo Foerster, CEO of a German consulting company Advan Teamhe sees similar dysfunction among businesses he advises. “Too many times there is no clear strategy. There is no defined goal. There is no ambiguous roadmap and no system to track progress,” he said.
“And the other big issues are: AI is not a plug-and-play miracle. Companies assume that data is “well” and skips data inventory and cleanup. Does not adapt the process. They don't try to bolt the AI.
Infrastructure Bottleneck
For every lecture on the algorithm, it is an invisible plumbing under the AI. “Bad data is rarely published,” Frawley was excited. “It works quietly and creates a slow erosion of trust and value that businesses don't notice until it's too late.” The damage is not hypothetical, but it is realistic, measurable, and often expensive.
“The flaws are amplified. This can lead to unrelated content, misplaced or false advertising that can waste millions of dollars of spending, or inconsistent narratives that cause brand disruption and alienate important audiences.”
Foerster numbers it. “60-70% of AI's growing pains go back to data infrastructure,” he told me. “It's not flashy innovation, it's overlooked because it's invisible, it's just a tough grind of governance. But without it, corporate AI systems would not be available to date.”
Ken Mahony, CEO Mahoney Asset Managementanother often overlooked bottleneck has been flagged: the physical limits of AI's appetite for energy and infrastructure. “Even if you have the money you want to spend, that may not be possible,” he said. “At this point, transformers, power supplies and cooling equipment are all limited availability. Energy companies have been backlogged for equipment for years and are ready to use, but they cannot create power and equipment.
AI context issues
Many enterprise leaders say AI can think of, and some people believe it completely, but in reality most AI systems are still I can't understand the context Or the human-like nuance is a problem Marcus often discussed, and it may not go away anytime soon, despite how much money he spends on building LLM.
Frawley says that without clear strategies and clean data, the model will confidently push the wrong actions. “Deploying AI on fragmented or inaccurate data is self-vassing,” he said. “It amplifies existing flaws, erodes the quality of your analysis, and gives you false confidence in misinformation decisions.”
Foerster provided a personal example. “In my favourite hotel, I booked a room overlooking the spa garden, No. 138. The AI in the call center didn't realize that my wife and I were regular guests or that this was our favorite room. I was pretty frustrated. The problem was only fixed when humans intervened and turned the AI.” With an AI-driven system, these small mistakes scale quickly and quietly until the customer's trust breaks down.
The same risks extend to more complex developments. “Generative, agent AI in particular, functions as an amplifier,” Frauley said. “With clean and well-built data, you can accelerate progress. With fragmented or inaccurate data, you can amplify errors and biases at speed, perform actions autonomously, and push your business further in the wrong direction before the problem is detected.”
Are you really ready?
The solution is not to abandon AI. It's about approaching it like any other important corporate transformation, with strategy, systems and accountability. Frawley pointed to clients in the communications and health sector who were unable to even unify their customer records before attempting sophisticated personalization or churn forecasting.
“We worked with them to create a single customer view by consolidating, enriching and enriching data from each source, whether digital or offline across all channels,” he said. After the correction, one client saw a 15% improvement in identifying duplicate records. “These improvements have wasteful advertising spending and reduced more meaningful and strategic customer communications at scale.”
Foerster also shared a similar case from the manufacturing industry. “German companies had great ambitions for predictive maintenance, but the deployment was neutral and stuck,” he pointed out. “The breakthrough only happened after standardizing machine data, linking it with master data, and building a central data warehouse. Suddenly, the model was accurate.
Move beyond AI's promises
If there is one thing these leaders agree to, it is this: AI won't fix broken businesses. Rather, it exposes it.
“If I had five minutes in the meeting room, I would ask one question,” Frawley said. “Will you bet on the accuracy of your reputation, your work, and your company's future data? If the answer is anything other than an immediate 'yes', your business is not ready for AI. ”
Mahony added a financial lens to this issue. “I want to know whether you're a new startup, an existing company that claims AI will strengthen your business, I want to know the staged plans and how you quantify how you're generating growth or more efficiency.”
And Foerster's final thoughts were rather dull. “What specific business problems does AI solve and how does it measure success? If you can't answer that, neither the best algorithm nor the biggest budget will help you.”
AI promises will just keep that up until companies have clearer answers – promises.
