AI Can’t Fix Bad Data: Why Quality Still Defines Value

As organisations accelerate their investment in AI, the conversation is often dominated by models, platforms, and capabilities. But beneath that momentum lies a more fundamental challenge – the quality of the data those systems depend on.

For many enterprises, data is fragmented, incomplete, or simply out of date. And as AI adoption grows, those underlying issues are becoming harder to ignore. Rather than solving data problems, AI can amplify them.

In this interview, Stuart Harvey, CEO at Datactics, shares insights from over three decades of working across real-time systems, big data, and digital transformation. Drawing on experience supporting financial institutions, government bodies, and national-scale data initiatives, he explains why data quality remains one of the most persistent – and underestimated – challenges in the AI era.

The Reality of Messy Data

Across sectors, the pattern is familiar and increasingly unavoidable. Organisations are sitting on vast amounts of data, but much of it is incomplete, inconsistent, or simply out of date. The problem is not confined to one industry. Whether it is financial services, policing, or healthcare, the same structural issues appear again and again.

“Everybody has a data quality problem,” said Harvey. “The more data that we have, the more the problem sort of reveals itself.”

The reality is that most enterprise data is not clean or neatly structured. It sits somewhere in between — identifiers combined with free-text descriptions, often entered manually and under pressure.

“It’s in that intersection between the standard identifier and the wider textual description… that you get all sorts of data quality issues,” added Harvey.

Whether the data is a customer record or a police report entered on the move, the conditions under which data is created introduce friction, inconsistency, and error.

Why AI Raises the Stakes

While these issues are not new, AI changes their impact. Complexity is not the only risk here; there is also timing. Data that appears complete may already be obsolete, and in an AI-driven environment, that becomes a governance issue.

“You could be training your models on very out-of-date information,” explained Harvey. “From a governance perspective, you need to understand just how current that information is.”

Combined with bias and incomplete datasets, this creates a scenario where AI does not just inherit problems – it amplifies them.

From Overwhelm to Practical Progress

Despite growing awareness, many organisations still struggle to act. In some cases, the problem feels too large to tackle. Data leaders understand the risks, but progress stalls under the weight of complexity.

“For the Chief Data Officer, it has just become an unsolvable problem… and for the business users, they’re bored of hearing 100 reasons why it can’t be fixed,” said Harvey.

The solution, he suggests, isn’t scale; it’s focus.

“How do you eat an elephant? Do one bite at a time.”

Rather than attempting wholesale transformation, organisations need to identify small, high-impact problems and solve them incrementally, building momentum as they go.

Why Data Quality Fails in Practice

The new challenge is behavioural, rather than technical. In many organisations, data quality processes exist but fail to drive action. Reports are generated, issues are logged, but fixes are delayed or ignored.

“It’s probably the least interesting thing to do in your week… so it gets kicked into next Friday, and then it builds up to a set of items they just can’t manage,” said Harvey.

This creates a cycle where problems accumulate faster than they are resolved.

To break that cycle, some organisations are changing how they engage teams,  introducing visibility, accountability, and even gamification to make progress tangible.

In one example, a large UK asset manager tackled the issue head-on by making data quality visible across the organisation. Rather than sending static reports, they created a central dashboard showing all regional offices and effectively a league table of performance, tracking who had resolved issues, when, and whether those actions had improved outcomes. The result was a noticeable change in engagement, with teams responding more positively to clear, shared visibility than to traditional reporting alone.

Elsewhere, organisations are taking a more structured approach by building internal centres of excellence. These teams develop reusable patterns and expertise that can be applied across different parts of the business, turning data quality from a reactive task into a scalable capability.

In more complex environments, such as the NHS, the approach goes further, treating data quality as an ongoing service rather than a one-off fix. Teams work closely with stakeholders to understand specific requirements and build solutions tailored to real-world use cases, recognising that standard metrics alone are rarely enough to drive meaningful improvement

The Challenge of Validation in an Open Data World

As data ecosystems expand, validation also becomes more complex. Organisations increasingly rely on external datasets to verify and enrich their own information — from company registries to government data sources. But open data introduces its own challenges, particularly around accuracy and trust.

“It’s a work in progress,” Harvey noted, reflecting the ongoing evolution of how data is curated, validated, and maintained.

The balance between accessibility and reliability remains a moving target.

AI: Transformation or Enhancement?

All of this leads to a broader question about AI itself.

Not long ago, there was a belief that AI might solve many of these challenges — that better models could compensate for imperfect data.

“There was this idea that we might not need data quality anymore… that the machine would just solve all of this,” said Harvey.

That view is now changing.

“I think we’re starting to settle down now and see how AI is really going to have an impact.”

Rather than replacing data infrastructure, AI is exposing its limitations. The debate now is whether AI represents a fundamental paradigm shift or a ‘feature extension’ of existing capabilities.

Harvey’s view is measured: the benefits are real, but the fundamentals remain.

Conclusion: Foundations Still Matter

As organisations push forward with AI, the focus is repositioning from experimentation to reliability.

The challenge now involves not just creating intelligent systems but also ensuring that the data powering them is accurate, timely, and trustworthy.

AI may accelerate insight and automate processes, but it does not eliminate the need for strong data foundations.

If anything, it makes them more critical than ever.