Up until now, enterprise AI has been largely dominated by a familiar conversation: which models should we use, where do we deploy copilots and agents, and how quickly can we move from experimentation to measurable business value?
These have all been pertinent questions, but as teams reach a new stage on their AI journey, they're no longer the pressing ones.
Chief Delivery Officer and Partner, Intellias.
The first wave of AI was one of discovery, with organizations asking a simple question: ‘Can AI help us at all?’ That was followed by a period of rapid experimentation, as enterprises launched proof-of-concepts and pilots across every conceivable business function to understand where AI could deliver value.
Now, the conversation has changed again. Enterprise leaders aren't struggling to identify AI use cases - most organizations already have dozens - the real challenge is moving from experimentation to scaling. It's no longer about asking ‘Where can we use AI?’ but ‘How do we make it work consistently across the business?’
And this shift in focus has exposed a problem many organizations originally underestimated: data.
AI exposes weakness in the foundations
In the earlier stages of developing AI, many enterprises focused on accessing proprietary datasets for one primary purpose: model customization. Although this challenge still exists, there’s now a bigger issue at play that involves data quality, accessibility and governance across the organization.
AI is only as effective as the information and processes it operates on. The tricky part here is that AI is a master of exposing weaknesses that have existed inside organizations for years. And that should serve as a wake up call for enterprise leaders investing in AI right now. Because when data is fragmented, processes are inconsistent and operational maturity is lacking. In this scenario, AI won’t fix the problem – it’ll simply amplify it.
Viewed this way, enterprises should best understand AI as a force multiplier, not a correction mechanism for the legacy inefficiencies. However, the opposite is equally true. When organizations treat data as a first-class product, AI has the potential to multiply the quality coming out of it.
And this is the key difference between implementing AI and being genuinely prepared for it. You might have a strong model, but if you’re working with fragmented and unstructured data, your efforts will only continue to produce inconsistent results. I’ve seen it first-hand.
This isn’t just an anecdotal point. Gartner predicts that through 2026, 60% of AI projects will be abandoned because they aren't supported by AI-ready data, while 63% of data management leaders say they either lack - or aren't sure they have - the data management practices AI requires.
Those findings reinforce what we’re seeing many organizations discover first-hand: AI success is increasingly determined by the quality of the underlying data rather than the sophistication of the model.
The data pilot trap
Many organizations saw encouraging results during their early AI pilots, and that isn’t surprising. Typically, pilots are built on curated samples of data that are absolutely real, but to some extent pre-selected and considered synthetic.
On that kind of data, it's natural to see success because the information has been carefully selected to demonstrate the technology's potential.
The real test begins when organizations start scaling AI and release it from the boundaries of those pilots into the real enterprise environment.
Suddenly, models have access to many different types of data that all have to work together. They're exposed to years of duplicated records, conflicting business definitions, incomplete customer data, inconsistent documentation and disconnected systems. Information that appeared reliable in a controlled pilot becomes far less dependable in production.
That's the point where leadership teams often get their biggest surprise. They realize the true state of their data is much worse than they expected. What worked well during the proof-of-concept stage simply doesn't work in the real production environment – not because the AI has become less capable, but because it has finally encountered the reality of enterprise wide data.
Unfortunately, that experience is becoming increasingly common. What’s abundantly clear is that the defining challenge now is not proving that AI works, but ensuring the data behind it is ready for enterprise deployment.
Signs your data isn’t ready for AI
How do you know if your data’s ready for AI? In many cases, the answer only becomes obvious after projects fail to deliver the expected results. But long before that happens, there are usually warning signs.
Your data governance has gaps: Can you identify where your data lives, who owns it and whether it can be trusted? If every analysis depends on manual reconciliation, AI will simply scale those inconsistencies. Years of organic growth often leave enterprises with siloed data, conflicting definitions and inconsistent governance. Which means before AI can deliver reliable results, organizations need clear ownership, consistent standards and dependable data pipelines.
AI initiatives are happening in isolation: When different teams are experimenting with AI independently, it's often a sign that the underlying data isn't connected. And it’s happening more often than you might think – McKinsey research found that fewer than 30% of organizations have their AI agenda directly sponsored by the CEO. Valuable information remains trapped in departmental silos or legacy systems, making it difficult to build a complete picture of the business.
Treated like this, it’ll always remain more function-level experimentation rather than coordinated enterprise transformation. AI performs best when it can draw on integrated, trusted data rather than fragmented datasets created for individual functions.
Your data isn't connected to business outcomes: AI creates value by improving business decisions and processes, not by analysing data for its own sake. If it's unclear how your data supports the outcomes you're trying to achieve, AI initiatives are unlikely to produce meaningful results. Incomplete, outdated or poorly maintained data will also undermine confidence in AI outputs, making it harder to move beyond isolated pilots.
You're spending more time choosing models than improving data: Selecting a foundation model is important, but it's rarely what determines success. The bigger challenges are preparing enterprise data, identifying high-value use cases, embedding AI into existing workflows and driving adoption across the business. Continually chasing the latest release will not deliver tangible results, investing in the data foundations that will make a model effective, will.
Potential will only be realized with strong foundations
Arguably, the biggest shift facing enterprise leaders is one of mindset: moving the focus from advanced models to strong foundations. McKinsey's research supports this, finding that organizational readiness accounts for 48% of the difference between companies that successfully capture value from AI and those that don't—making it a stronger predictor of success.
For enterprises, the ultimate goal with AI should not simply be to automate existing tasks, but how to rethink how the business operates. That means redesigning processes around AI's capabilities, rather than layering AI onto inefficient ways of working.
For that to happen, data quality, governance and clear ownership can no longer be treated as minor, back-office concerns, they need to be recognized as strategic priorities.
The opportunity for enterprise AI is enormous, but without trusted data to build on, its potential will remain just that, potential.
We've reviewed, rated, and ranked the best data removal service.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit