Artificial intelligence (AI) is often sold as the shortcut to smarter decision-making. There’s an assumption that if the right models are deployed and the correct tools connected, every organisation will operate faster and more effectively. But according to Jonathan Oaker, CloudZA’s CEO, the companies seeing meaningful results are not prioritising models, they’re focused on their data foundation.
“Your data is the fuel to your organisation scaling,” says Oaker. “If the data foundation is correctly put together, that fuel flows. You will start to see your organisation take off.” AI cannot produce reliable insight when it is built on inconsistent, scattered or poorly structured data. The real differentiator is not the model; it’s the data feeding it. “You want to be able to get the quickest performance out of AI while also reducing the cost,” he says. “You will never get that when your data is fragmented.”
Preparing usable data
AI readiness is not a question of model sophistication. It’s determined by whether the organisation’s data can support intelligent computation at scale. This is why AI fails earliest at the data layer. When information sits across disconnected systems, mismatched schema and legacy formats, the model spends its time reconciling context rather than reasoning with it. “It is all about context. Fragmented data forces the model into a larger context window,” says Oaker. These failures are often mistaken for AI limitations. In reality, the model is only reflecting the constraints of the data beneath it. When that foundation is unstable, everything above it becomes unpredictable.
The first step in avoiding these failures is preparing data in a way that models can interpret reliably. AI does not need more information; it needs information that is consistent across sources and structured in a form it can compute efficiently. A modern data lake, for example, helps organisations achieve this by bringing different datasets into a single environment where governance, anonymisation and schema alignment can be applied. “The data foundation is where it all is,” continues Oaker. “Once the data is clean and sufficient, the model can actually interact with it effectively.”
The data that matters
The next step, says Oaker, is separating valuable data from noise. Organisations often assume that everything they collect should run through AI systems, yet irrelevant inputs only degrade performance. “It’s the same with human beings. If I give you a whole bunch of confusing information, you will take longer to get to the core of it all,” says Oaker. “A model works in the same way. Not everything needs to go into AI; you want the data that represents business value.”
This misconception often leads businesses to push every dataset into the pipeline, applying AI to data that has no value. One materials business that CloudZA worked with tried recalculating stock values for every webpage visitor. It added needless computation, a change that only increased compute cost without improving accuracy or insight. “But is there actually a return on this investment?” questions Oaker.
Oaker is often asked how to judge whether a data foundation is ready for AI, and his view is that the system shows this before anything else does. If responses slow down, if results shift or if the model needs more compute to stay consistent, the foundation needs work. “We try to push for sub 5% word error rate for production,” explains Oaker. “When both structured and unstructured data sit within that range, you can trust the output.” But the impact should also be practical. The user experience should improve before the organisation sees the metrics move. “The whole point is to make myself 10 times faster. Or can I make my staff 10 times faster?” he asks. “It all comes back to the data foundation.”
What readiness looks like
For Oaker, the organisations advancing fastest with AI are those modernising their data platforms with open formats, unified storage and automated governance. They are rebuilding the part that determines speed and consistency. “Success is moving beyond silos and legacy barriers,” adds Oaker. “Yes, your data might be clean and yes, your data might be efficient, but is there business value? If we can find an efficient way to talk to AI, so it’s not taking too long to give responses, then our foundations are in a good space to start scaling up.”
The message, says Oaker, is to modernise the platform in open formats and open standards on object storage so data can be queried efficiently at scale. That is when AI shifts out of pilot mode and starts behaving like infrastructure the business can rely on.
Share