Table of Contents

• The data most companies never use
• Most AI never reaches scale
• The limits of training everyone
• What “truly usable” actually means
• Designing around the person

Most organisations already have more data than they can use and more AI tools than they can operate. The dashboards are built, and the pipelines run every night. And still, the person who needs an answer on a Tuesday morning cannot get one without filing a request and waiting days for an analyst. The complexity did not go away when the technology arrived. It moved closer to the people who were supposed to benefit from it.

This is the quiet failure at the centre of the current AI wave. Scale was supposed to solve it. Companies collected more data and trained bigger models, and both of those things worked on their own terms. What scale did not fix is the distance between that capability and the people meant to use it. The systems we build still ask people to think like engineers before they are allowed to think like decision makers.

The data most companies never use

Start with the raw material. Analysts have estimated for years that the majority of enterprise data is “dark”, collected and stored during normal operations and then left untouched. Gartner popularised the term, comparing dark data to dark matter because it makes up more than half of what most organisations hold. Seagate’s Rethink Data Report put a figure on it, finding that only around 32% of available data is ever put to work and roughly two-thirds sits idle. Splunk’s research pointed the same way, with most surveyed leaders reporting that more than half of their data is dark and a large majority admitting they struggle to find and use it.

None of this is a storage problem. The data exists. It is backed up and paid for. What is missing is a practical way for an ordinary person to ask it a question and get something useful back. Volume was never the hard part. Access was.

Most AI never reaches scale

The same pattern shows up one layer higher, in AI itself. McKinsey’s 2025 global survey found that 88% of organisations now use AI regularly in at least one function, up from 78% the year before. By that measure, adoption is close to universal. Then the numbers turn. Only about a third of organisations report scaling AI across the enterprise, and just 39% see any impact on EBIT at the enterprise level. Most companies are running pilots that never graduate to production.

McKinsey is direct about why. Scaling depends on clean, well-governed data and on workflows that have been redesigned around the technology. Bolting a model onto an unchanged process produces a demo, not a result. We covered these findings in more detail in an earlier analysis of the report. The gap between using AI and getting value from it is, in large part, a gap in usability.

The limits of training everyone

There is a tempting response to all of this: teach every employee to work with data. The evidence suggests that approach only goes so far. An Accenture study of more than 9,000 workers found that only 21% felt confident in their data literacy skills, and Gartner’s surveys of chief data officers repeatedly list poor data literacy among the top barriers to becoming a data-driven organisation.

Training helps, and it should continue. But expecting every marketer and operations manager to become fluent in query languages and statistical caveats is not a plan. It is a bottleneck dressed up as an aspiration. A usable system carries that load so the person using it does not have to.

What “truly usable” actually means

Usability is easy to claim and hard to design. In practice, it comes down to a few commitments that shape every decision.

Meet people in their own language. A usable system accepts a plain question the way a colleague would, and returns a plain answer. Natural language is the interface most people already know. When someone can type “how did youth unemployment change in Spain over the last five years” and receive a clear answer with the chart already drawn, the complexity has been absorbed by the system rather than the user.

Turn output into answers. A wall of rows is not an answer. A usable system does the interpretation, surfaces the figure that matters and puts it in context, so the person can make a decision instead of starting a second job.

Hide the machinery. Keep the reasoning visible. The pipelines and model calls should be invisible. The logic behind an answer should not be. A system earns trust by showing where a number came from, and by saying clearly when it does not know rather than presenting a confident guess.

Respect the boundaries of the data. A usable system is honest about coverage. If a question reaches past what the data can support, it says so instead of inventing a tidy response. That restraint is what makes the useful answers believable.

None of these principles are about making AI more powerful. They are about pointing existing power at the person who needs it.

Designing around the person

This is the principle behind how we build at Neodata. Across video, documents and structured data, the aim stays the same. Take a genuinely complex system and give people a simple way in.

equa, our natural language assistant for European statistics, is a working example. Eurostat and ISTAT hold an enormous amount of public data across 27 EU countries, and almost none of it is reachable for a non-specialist in a hurry. equa lets anyone ask a question in plain language and get a sourced answer with a chart, without learning a single dataset code. The complexity of the underlying data has not been removed. It has been moved to where it belongs, inside the system and away from the person asking.

The end of data complexity does not mean simpler data. Data will keep growing and keep getting messier. It means the complexity stops being the user’s problem. That is the standard worth designing to, and it is what separates an organisation that owns a great deal of data from one that can actually use it.

If you are trying to close the gap between the data you hold and the decisions you make, we would be glad to talk about what usable AI could look like for your team.