Skip to content
There’s a pattern we see again and again with organisations embarking on their AI journey. They invest in the tools, brief the team, and wait for the transformation to arrive. Then, a few months in, the results are disappointing. The model isn’t behaving as expected. The outputs feel unreliable. The project quietly stalls.
The problem isn’t AI. The problem is almost always the data underneath it.
The Uncomfortable Numbers
Recent research from BARC, which surveyed over 400 organisations on their AI experiences, found that data quality issues more than doubled as the top obstacle to AI success, jumping from 19% of organisations in 2024 to 44% in 2025. That’s not a small uptick. That’s a signal.
Gartner found that at least half of all generative AI projects were abandoned after the proof-of-concept stage. And according to the same BARC research, only 20% of organisations have actually built the proper foundations needed to support AI in production.
These numbers are striking, but they shouldn’t be surprising. AI doesn’t create insight from thin air. It finds patterns in the data it’s given. If that data is inconsistent, incomplete, or siloed across a dozen different systems, the model will faithfully reflect that mess right back at you.
Garbage In, Garbage Out. At Scale.
We’ve worked with clients who came to us excited about AI, only to discover that their biggest challenge wasn’t which model to use. It was that their customer data lived in three separate systems with no common key, or that the same metric was being calculated differently across departments, or that nobody quite knew which version of a dataset was the authoritative one.
These aren’t exotic problems. They’re the everyday reality of most data environments. And while they might be manageable when a human analyst is doing the work, AI amplifies them. A model trained on messy data doesn’t produce mediocre outputs. It produces confidently wrong ones.
So What Does “Discipline” Actually Mean?
It means treating data as infrastructure, not an afterthought. Here’s what we see separating organisations that succeed with AI from those that struggle:
Data governance that’s actually practised. Not a policy document that lives in a shared drive, but agreed definitions, clear ownership, and processes that people follow day to day.
Data quality monitoring. Knowing when something has gone wrong before a model makes a decision based on it. This means checks, alerts, and accountability.
Connected data. AI works best when it can draw on a full picture. That means investing in integration, whether through a data warehouse, a lakehouse, or well-managed pipelines, so models aren’t working from a partial view.
A clear business question. The most disciplined AI projects start with “what decision are we trying to improve?” not “what can AI do for us?” Starting with the use case keeps data work focused and makes it easier to measure success.
The Organisations Getting This Right
The BARC research found that organisations with mature data and AI foundations were nearly twice as likely to have five or more AI projects running in production compared to those without. The discipline compounds. Once the foundations are in place, deploying AI for new use cases becomes faster and cheaper.
This is the virtuous cycle we try to help our clients build. It’s not glamorous work. Getting data governance right rarely makes headlines. But it’s the difference between AI that actually changes how your business operates and AI that stays stuck in a pilot forever.
Where to Start
If you’re wondering whether your data is AI-ready, a good first question is: can your team agree on the definition of your top five business metrics? If the answer is “it depends who you ask,” that’s your starting point.
The good news is that building these foundations doesn’t have to take years. With the right focus, organisations can make meaningful progress quickly, and the payoff extends well beyond AI.
If this resonates with where your organisation is right now, we’d love to have a conversation. At Idiro, we help businesses get their data in shape for the future. Sometimes that means AI. Always, it means better decisions.
