Why AI Projects Fail in Production
Reflections from DataVision EMEA 2026: the missing data operating system
The main obstacle to getting AI into production is usually not the model but the condition of the data beneath it. AI amplifies whatever it sits on: strong data speeds up good decisions, weak data speeds up poor ones. The lasting advantage will come from data that is trusted, well defined, reusable and lawfully accessible.
I recently attended DataVision EMEA 2026 at UBS in London. The conference brought together senior figures from banking, market infrastructure, technology and data management under the theme “Gold In, Gold Out: Building Trusted Data to Optimise AI.”
Beyond “garbage in, garbage out”
That wording was deliberate. It moved the discussion beyond the familiar warning about “garbage in, garbage out” and towards a more useful question: what does good data actually look like when an organisation wants to use AI properly, at scale, and inside live business processes?
The clearest lesson I took away was that the main obstacle to successful AI is often not the model. It is the condition of the data sitting beneath it.
The gap between a demo and a working system
It has become remarkably easy to produce an impressive AI demonstration. A small team can take a carefully selected dataset, define a narrow problem, and show a convincing result in a short period of time.
Putting that same system into production is a different matter entirely. Production requires working with old and new technology side by side, inconsistent definitions, incomplete metadata and local workarounds. It also requires privacy, confidentiality, security, regulatory and contractual obligations to be addressed properly. A demonstration proves a model can do something interesting; moving it to production requires the organisation to prove it can do that repeatably, lawfully and at scale.
Quoted at DataVision EMEA 2026
Digital sandcastles
The additional cost of reaching production is not simply the model; it is the work needed to make an organisation's data dependable enough for that model to use.
A panellist offered a useful image: children building a sandcastle on a beach. It is possible to build something impressive quickly, but when the tide comes in it destroys what has been built, because there were no solid foundations, just sand. Many AI pilots are digital sandcastles: convincing in a protected environment, but unable to survive contact with the operating business.
AI amplifies what is already there
James Dallas of UBS described AI as an amplifier sitting on top of data. Where the data is strong, AI can accelerate sound analysis and better decisions. Where the data is weak, it accelerates poor decisions instead.
AI is also exposing weaknesses that have, until now, remained hidden, largely because experienced staff have quietly fixed the errors and kept imperfect systems working. An experienced team usually knows which source of information is reliable, which field to treat with caution, and which exceptions need checking before a figure is relied upon in a report or spreadsheet. Those protections are not present once the processing is carried out by an AI system, because the AI has none of the local knowledge a team builds up over years of working with the data and the systems that hold it.
If an organisation has no coherent way of identifying its important data, tracing where it came from, assessing its reliability, and controlling how (and by whom) it may be used, then it has no supporting structure. And without that structure, AI has nothing dependable to stand on.
AI strategy is data strategy
The strongest lesson I took from DataVision was not that organisations need less ambition around AI, but that they need stronger foundations beneath that ambition.
The winners are unlikely to be determined solely by who has access to the largest or most advanced model. Models will keep improving, and access to them will keep broadening. The more lasting advantage will come from the ability to provide those models with data that is trusted, well defined, reusable and lawfully accessible.
In Summary
An AI strategy without a data strategy may still produce useful experiments. On its own, however, it will not produce lasting transformation.
Key Takeaways
- The main obstacle to production AI is usually data quality and governance, not the model itself.
- A convincing AI demo proves little about whether a system can run repeatably, lawfully and at scale.
- Weaknesses in data are often masked by experienced staff; AI has no equivalent instinct for what to distrust.
- Durable AI advantage will come from trusted, well-governed, reusable data, not from access to the biggest model.
Frequently Asked Questions
The main obstacle is often not the model but the condition of the data beneath it. Production means working with old and new technology side by side, inconsistent definitions, incomplete metadata and local workarounds, while also meeting privacy, confidentiality, security, regulatory and contractual obligations.
A demo can be built quickly from a carefully selected dataset and a narrow problem. It proves a model can do something interesting. Moving to production requires the organisation to prove it can do that repeatably, lawfully and at scale.
It was the theme of DataVision EMEA 2026: Building Trusted Data to Optimise AI. It moves beyond the familiar warning about 'garbage in, garbage out' towards a more useful question: what good data looks like when an organisation wants to use AI properly, at scale and inside live business processes.
AI sits on top of data like an amplifier. Where the data is strong, it can accelerate sound analysis and better decisions; where the data is weak, it accelerates poor decisions instead. It also exposes weaknesses that experienced staff have quietly been fixing, because an AI system has none of their local knowledge.
A coherent way of identifying its important data, tracing where it came from, assessing its reliability, and controlling how, and by whom, it may be used. Without that structure, AI has nothing dependable to stand on.
Probably not solely from access to the largest or most advanced model, as models will keep improving and access will keep broadening. The more lasting advantage will come from providing those models with data that is trusted, well defined, reusable and lawfully accessible.
This article is for general information only and does not constitute legal advice. You should take specific advice on your own circumstances before acting.
