Most executives I talk to still think AI is a technology challenge. It’s not. It’s a data driven problem.
AI investment succeeds or fails based on what you feed it. But here’s what makes this tricky: AI consumes two different flavours of data. And businesses only think about one of them.
Training: Building the Brain
Training data is what builds AI models. You teach the machine to recognize patterns, make connections, and understand context. This needs huge volumes of high-quality, well-labelled and most importantly, relevant data.
The good news? most organisations won’t be doing this. You’re not Google or OpenAI building foundation models. You’re what I call an “AI Consumer” – using pre-built models that someone else trained.
Inference: Feeding the Machine
This is where most businesses (including yours) lives. Inference is the act of throwing your data at a pre-trained model. And, most of the time, getting back something useful.
Ask your phone to find photos of your dog: that’s inference. An agentic workflow that processes customer data to generate insights: that’s inference too. The model already knows how to think. You’re just providing it your data to think about.
The Formula That Matters
Both flavours follow the same formula: Data + AI Black Box = Output You Hope Is Useful.
Notice “hope” in there? That’s the problem we need to fix.
If half of the formula is making sure that the data going into our black box is correct and relevant and not junk, then we need to talk about how you organizing that data today. AI will have a go a processing the data you feed it, even it complete garbage.
The Question Nobody’s Asking
“We’ve spent the last two years obsessing over which AI model to implement. Studying the benchmarks weekly. Agonising over public or private deployment”
We’re asking the wrong question. The question shouldn’t be which AI to buy – it’s whether you have the right data to feed it.
AI is only as good as your ability to get relevant data to it quickly. Not all of your data. Not just your newest data. The right data for the task at hand.
What This Means for Your Strategy
Most companies are still thinking about AI back-to-front. They’re buying the models first, then finding out they don’t have the right data to make the models useful.
You wouldn’t hire a brilliant analyst and then bury them in a room with unsorted filing cabinets. That’s what we’re in danger of doing with AI.
The businesses that will win with AI will be the ones who organise their data so well that any model can access what it needs, when it needs it.
Where to Start
Before you invest another penny in AI capabilities, ask one question: can you find the right data for a specific task in less than a minute?
If the answer is no, you don’t have an AI problem. You have a data management problem that AI will only make more expensive.