Context Is King. But the King Needs a Queen.

It’s been nearly a month since Big Data LDN 2026, and I’ve taken a little longer than usual to sit down and write about it. Partly because there’s been plenty else going on, but also because I wanted some time to reflect on what I’d taken away from the event rather than rush to publish my thoughts.

One of the things I enjoy most about attending these events is reconnecting with people I’ve met throughout my career. Hearing what others are working on, how their thinking has evolved and how different organisations are approaching familiar challenges is always valuable. The presentations are interesting, of course, but I’ve often found that the conversations between sessions can be just as thought provoking. There’s also something to be said for stepping away from the demands of day-to-day delivery and giving yourself the space to consider where the industry is heading.

This year, unsurprisingly, artificial intelligence dominated much of the conversation. What interested me, however, wasn’t so much the technology itself but the increasing attention being given to the knowledge and organisational understanding required to make it useful. Data governance, semantic models, ontologies and enterprise context layers featured prominently, with some interesting ideas about how these might develop as organisations begin deploying more sophisticated AI systems.

One presentation in particular, from Atlan, explored the concept of an enterprise context layer, bringing together business definitions, relationships, rules and operational expertise so that AI systems can interpret information more consistently. I could see the value in what was being proposed, but the more I listened, the more I found myself questioning the underlying narrative.

Context is king, apparently. But hasn’t it always been?

A familiar problem with a new name

Anyone who has spent time working in business intelligence and analytics will recognise the importance of context. The structure of data, the relationships between entities, the meaning of individual attributes and the rules governing how information is interpreted have always been fundamental to producing useful intelligence.

Consider two departments reporting different figures for the same performance measure. Both may have reliable data pipelines and technically correct calculations, yet their results differ because they’ve applied different reporting periods, exclusions or interpretations of the underlying business activity. Sometimes those differences are legitimate, but often the definitions have simply evolved independently, without anyone taking responsibility for reconciling them.

We’ve spent decades developing approaches intended to address these challenges, from dimensional modelling and master data management to business glossaries, metadata catalogues, semantic layers and governance frameworks. Each has contributed something useful, and it would be unfair to suggest the industry has made no progress. Nevertheless, the persistence of conflicting metrics, duplicated business logic and poorly understood data suggests that technological advancement hasn’t always translated into a comparable improvement in organisational understanding.

Part of the difficulty, I suspect, is that we’ve become extraordinarily capable at moving and processing information without becoming equally disciplined about establishing what it means. Cloud platforms have transformed data infrastructure, modern engineering tools have improved the reliability of pipelines, and reporting technologies have made information more accessible than ever. These are considerable achievements, but none guarantees that the resulting information is meaningful.

A measure can be calculated accurately while representing something different from what its audience believes it represents. A dashboard can be technically sound while encouraging a misleading interpretation of performance. Indeed, the more polished and accessible our reporting becomes, the easier it may be to overlook the assumptions beneath it.

This distinction between technical correctness and meaningful interpretation is something I’ve become increasingly conscious of in my own work. Establishing a calculation is often relatively straightforward compared with reaching agreement about what should be measured, why it matters, what circumstances should be excluded and how the result ought to influence decisions. Those conversations require an understanding of the business that cannot be derived from database structures alone.

What AI changes, and what it doesn’t

I don’t think the current enthusiasm for enterprise context is entirely another exercise in repackaging established ideas. There is something genuinely different about the way AI systems are beginning to interact with organisational information, even if the foundations they require are familiar.

Historically, much of the interpretation of analytical information has remained the responsibility of people. An analyst investigating an unexpected result might question a definition, examine the underlying records or speak to an operational colleague who understood the circumstances behind the figures. These processes are far from infallible, but they provide opportunities for ambiguity to be recognised and challenged.

Increasingly, however, we’re asking AI systems to do more than retrieve information or produce summaries. We want them to investigate performance, identify patterns, recommend interventions and potentially take action through connected business systems. That creates different consequences when the information available to them is incomplete or poorly defined.

Imagine an organisation that has never fully reconciled its definitions of customer retention. In a conventional reporting environment, the disagreement might result in competing dashboards and periodic arguments about whose figures should be trusted. An AI agent operating across those systems could encounter the same ambiguity while being expected to identify customers at risk and recommend interventions.

The underlying problem hasn’t changed, but our expectations of the technology have. We’re moving from systems that primarily present information for human interpretation towards systems that increasingly participate in the interpretation and application of that information.

Human decision makers have always been capable of acting on misunderstood data, so AI doesn’t introduce that risk from nowhere. What it potentially changes is the speed and scale at which an interpretation can be applied, particularly where automated actions are involved.

If we haven’t consistently established trusted business meaning for relatively familiar analytical systems, what gives us confidence that we’ll achieve it simply because the next generation of technology depends upon it?

From governance as documentation to governance as practice

I suspect one reason we’ve struggled is that data governance has too often been treated as something we document rather than something we practise.

We create glossaries, establish standards, assign ownership and publish policies, all of which can be valuable. But documenting a definition doesn’t ensure that the same definition is used in the calculations underpinning business reporting. Naming a data owner doesn’t necessarily establish who has responsibility for resolving disagreements, and publishing a governance framework doesn’t guarantee anyone will follow it when the pressure is on to deliver something quickly.

There’s frequently a considerable distance between the way an organisation describes its information and the way that information is actually implemented and used.

This is why I’m increasingly interested in treating business meaning as something that needs to be engineered and maintained. If a metric is important enough to influence organisational decisions, its definition should be explicit, its implementation consistent and testable, and changes managed with an understanding of their wider consequences.

Many of the practices required already exist within software and data engineering. Version control, automated testing, documented decisions and dependency management all provide ways of making systems more reliable. There’s no obvious reason why the business definitions governing those systems should be treated with less discipline.

AI may even help with some of this work. Existing SQL, transformation models, metadata and documentation contain evidence of how organisations currently interpret their information. AI could help identify repeated calculations, highlight inconsistencies and suggest where definitions are missing. It could also expose gaps through the questions people ask and the circumstances existing knowledge fails to address.

But discovering how an organisation currently interprets information isn’t the same as establishing how it ought to interpret it. A calculation repeated across fifty reports isn’t necessarily correct, and a convincing AI generated description of a business process isn’t automatically authoritative. The responsibility for establishing meaning and resolving disagreement still belongs to the organisation.

The real opportunity is to use AI to make good governance more achievable, rather than imagining it can replace the need for governance altogether.

Context is King. But the King needs a Queen.

There’s another dimension to this conversation that I’ve found myself reflecting on recently, although it comes from somewhere quite different from data engineering.

I’ve been thinking about the relationship between the masculine and feminine, particularly the idea that they represent complementary principles rather than opposing forces from which one must prevail. I’m not referring to fixed characteristics belonging to men and women, but to a symbolic understanding of qualities that can coexist within individuals and the wider world.

In that symbolic language, we might associate the masculine with structure, differentiation, order and directed action, while the feminine represents relationship, receptivity, intuition and integration. These associations are neither universal nor exhaustive, but I’ve found them useful as a way of considering how different forms of understanding contribute to a more complete picture.

What interests me isn’t simply the idea of balance, as though the objective were to maintain equal quantities of two opposing qualities. It’s the possibility that each becomes more meaningful through its relationship with the other. Structure provides something within which understanding can develop, while understanding gives structure purpose and relevance.

The more I’ve considered this, the more I’ve recognised a parallel with the way we approach information.

The data industry has become highly accomplished at imposing structure. We define entities, establish relationships, develop measures and introduce controls intended to ensure consistency. These activities are essential, but there’s another aspect of understanding that is less easily captured in a technical specification. It involves recognising the circumstances behind an observation, appreciating the limitations of a measure and remaining receptive to explanations that the available data may not immediately reveal.

Consider a measure of participant engagement within an employment support service. We might establish a precise definition based on recorded appointments or interactions, determine the reporting period and consistently apply the appropriate exclusions. The resulting metric could be technically sound and useful for understanding patterns across a service.

Yet an absence of recorded engagement doesn’t necessarily tell us why an individual hasn’t participated. There may be financial difficulties, caring responsibilities, health circumstances or problems with the way support is being delivered. Some of those factors may be recorded elsewhere, while others may not be captured at all.

The metric isn’t necessarily wrong. The mistake would be to assume that it provides a complete explanation of the person or circumstances behind it.

We need the discipline of measurement to establish what we can reliably observe, but we also need the judgement to recognise what those observations cannot tell us. Understanding the limitations of a measure is part of understanding its meaning.

The masculine and feminine provide a philosophical analogy for this relationship, rather than a technical model or a claim that one mode of understanding belongs exclusively to either sex. The point isn’t to elevate intuition above evidence or to suggest that structure and interpretation exist in separate worlds. It’s to recognise that precision and understanding depend upon one another.

We can be remarkably precise about something we have fundamentally misunderstood, just as we can have an intuitive understanding of a situation without the evidence needed to establish whether that interpretation is justified.

Perhaps that’s the imbalance we should be paying more attention to. We’ve devoted enormous effort to developing the systems through which information is structured, processed and distributed, but haven’t always given the same attention to the relationships and human understanding that make it useful.

Beyond the buzzwords

I came away from BDL 2026 encouraged by the growing attention being given to organisational knowledge and business meaning, but also somewhat frustrated that so many of the underlying challenges remain familiar despite years of technological progress.

Perhaps AI will provide the urgency organisations need to address them more seriously. It may also provide useful tools for discovering inconsistencies, maintaining definitions and identifying gaps that previously required considerable manual effort. Both possibilities are worth exploring.

But I’d like to see us become more interested in practical evidence of progress than in the terminology used to describe it. How many of our important business measures have definitions that are genuinely agreed and consistently implemented? How confidently can we explain their assumptions and limitations? When operational colleagues challenge a metric, do we have a process for investigating their concerns? And when a definition changes, can we identify the systems that depend upon it?

These questions are less exciting than announcing a new enterprise architecture, but answering them would represent meaningful progress.

Ultimately, I think the most interesting opportunity presented by AI isn’t simply giving machines access to more organisational context. It’s the possibility that the demands of AI might finally encourage us to become more rigorous about how organisations establish and maintain their own understanding.

That will require more than technology. It will require the difficult conversations about definitions, accountability and the limitations of what we measure, alongside a willingness to value understanding the business as highly as building the systems that describe it.

Context has always mattered. What remains to be seen is whether this latest wave of technological development will finally persuade us to treat it with the seriousness it deserves.

As for the King and Queen, I’ve increasingly come to think that neither has much of a kingdom without the other.