AI success depends on data foundations

Organizations achieving the greatest value from AI are those with solid data foundations—discoverable, high-quality, and safely accessible—rather than those with access to the latest models, according to an analysis of AI readiness. The biggest challenges often lie in data silos, governance, and integration, especially as agentic AI systems demand trusted data for autonomous actions. Successful AI programs spend as much time on data architecture and governance as on developing agents themselves.

The conversation around AI is often dominated by models, tooling and the latest advances in generative and agentic AI. Organisations are understandably keen to understand how these technologies could improve operations, unlock efficiency or create new opportunities. However, in many cases the biggest challenge isn’t choosing the right AI solution. It’s ensuring the foundations are in place to support it. The organisations seeing the greatest value from AI are not necessarily those with access to the latest models. More often, they are the organisations with solid data foundations: discoverable, high-quality, and safely accessible. This is particularly important as businesses begin exploring agentic AI, where systems move beyond generating content to reasoning, making decisions and initiating actions. Start with the business problem One of the most common mistakes organisations make is approaching AI as a technology initiative rather than a business initiative. Before considering models or platforms, it’s important to understand the problem being solved and what information is required to solve it. Only then can organisations determine whether they have the necessary data of sufficient quality. In our experience, AI readiness often comes down to three fundamental data questions: - Do we have the right data? - Is that data sufficiently trustworthy and complete? - Can it be accessed in a way that enables AI systems to use it effectively? Answering these questions may not be as exciting as experimenting with the latest AI tooling, but it is often the difference between success and disappointment. Agentic AI raises the stakes Many organisations are now looking beyond traditional machine learning and generative AI towards agentic systems. These systems have the potential to be highly valuable because they can be given a broad goal and work out the best way to achieve it within security bounds across disparate systems. However, they also place greater demands on the underlying data. A chatbot that provides an inaccurate answer is a problem. An autonomous system acting on inaccurate information can create much larger challenges. As organisations increase the level of autonomy, they are willing to give AI systems, the importance of trusted, accessible and well-governed data increases accordingly. This is why successful AI programmes often spend as much, if not more, time focusing on data architecture, governance and integration than they do on developing and evaluating agents themselves. Data silos remain one of the biggest barriers For many large organisations, data exists across a complex estate of operational systems. These systems have often evolved over many years, serving individual business functions well but making it difficult to create a holistic view of information across the organisation. Data may exist in multiple formats, be duplicated across platforms or be subject to different ownership and governance controls. This creates challenges not only for analytics, but also for AI. We’ve seen examples where valuable information exists within an organisation but cannot easily be used because it is fragmented across multiple systems. Bringing that information together in a governed and accessible way can be a significant step towards enabling future AI capabilities. The prototype is only the beginning Many organisations have already built promising AI prototypes. Demonstrations can often be created quickly and can generate considerable enthusiasm among stakeholders. The challenge typically emerges when organisations attempt to move those solutions into production. Production AI systems require security controls, operational safeguards, monitoring, testing and governance. They must perform reliably under real-world conditions, not simply in a controlled demonstration environment. The same is true of the data underpinning these systems. Inconsistent, inaccessible or poorly understood data may not prevent a prototype from functioning, but it can create significant obstacles when trying to scale. Legacy data platforms often become an AI challenge Many organisations pursuing AI initiatives eventually discover that the greatest barriers are not related to AI at all. Legacy data platforms, data quality issues, and fragmented architecture can all limit the ability to adopt new capabilities effectively. This is not surprising. If information is difficult to find and access, poorly quality or held within systems that are expensive to change, introducing AI on top of those foundations is unlikely to solve the underlying problem. In many cases, successful AI adoption and modernisation go hand in hand. Focus on fundamentals The AI landscape will continue to evolve rapidly. Models will improve, tooling will mature and new approaches will emerge. What is unlikely to change is the importance of data. Organisations that understand their data, invest in its quality and create architectures that make it accessible will be better positioned to take advantage of whatever comes next. Those foundations support analytics, machine learning, generative AI and agentic systems alike. The organisations that succeed with AI won’t necessarily be those that adopt the newest technology first. They will be the organisations that have done the work to make their data usable, trustworthy and accessible. Because ultimately, AI success depends on data foundations.