Organisations are pouring billions into AI transformation, yet many are struggling to translate that investment into material business impact. While enterprise adoption is extremely high, recent data points to a growing gap between activity and impact.
According to MIT’s 2025 NANDA research study, despite US enterprises investing $35-40 billion in generative AI initiatives over two years, 95% report zero measurable P&L impact. The study, based on 150 executive interviews, 350 employee surveys, and analysis of 300 public AI deployments, reveals only 5% of custom enterprise AI solutions reach production deployment.
This is not a technology problem. The challenge is that AI is being deployed on structurally ambiguous foundations, leading to a lack of contextual learning and misalignment with day-to-day operations.
You Cannot Model What You Haven’t Mapped
When we are faced with complex information, the first thing we instinctively do is create a mental map. We draw diagrams, sketch structures, and lay things out visually so we can see how elements connect. While there is a tendency to think of this as a presentational exercise, it is an essential part of reasoning. It’s how we impose order on complexity and turn information into understanding.
What we now know is that we need to do the same for machines. Without an explicit view of context, in terms of how obligations, dependencies, and risk interact, AI has nothing stable to reason over; it lacks necessary relevance. It may accelerate execution, but it cannot support strategic decision-making.
A well-constructed information architecture provides the missing context. It shows how entities relate to one another, where dependencies sit, how obligations cascade, and which elements matter in a given situation. That context allows both humans and machines to reason, rather than simply process.
This is why information architecture must come before the implementation of artificial intelligence. If we want AI to reason rather than report, we need to imbue it with context, connections, and navigable pathways — what we call ‘Structural Intelligence’ — so it can determine what is relevant, what is not, and adjust its reasoning accordingly.
What AI Can and Cannot Do without Structure
Without explicit structural foundations, AI is limited to surface-level tasks. In 2021, Zillow shut down its ‘iBuying program’ after its AI-powered Zestimate algorithm overvalued properties by $500+ million, forcing the company to sell 7,000 homes at a loss. The algorithm processed millions of home sales but failed to account for post-pandemic market volatility, contractor availability, and regional variations — contextual factors that weren’t represented in its data structure.
Competitors like Opendoor and Offerpad, using algorithms with better structural intelligence about market relationships, weathered the same volatility. The difference wasn’t computational power, it was structural context.
It can read documents, summarise information, and generate outputs. But it cannot reason reliably. When structural intelligence is in place, the picture changes fundamentally.
Mapped relationships allow AI to move beyond isolated analysis and begin reasoning about impact. It becomes possible to ask questions like, “What are the unintended consequences of this decision?” or “Where are the hidden risks or dependencies we’re not currently seeing?”
Importantly, this does not replace human judgement. It strengthens it. Visual structure allows professionals to interrogate AI-generated outputs, validate them against source data, and apply contextual experience. It is the judgement that comes from knowing when something looks right, and when it does not.
Where Human Judgement Meets Machine Intelligence
When structure is made explicit, AI and human expertise can operate together. What Ethan Mollick refers to as ‘co-intelligence’, each compensating for the other’s limitations. Machines excel at navigating scale, surfacing patterns, and testing assumptions. Humans remain the arbiters of context, and retain the ability to take action.
Context is shaped by experience, domain knowledge, and judgement built up over time. It is how professionals recognise when something feels wrong, even if it appears correct on the surface. Structural intelligence gives humans the ability to interrogate AI outputs with discipline, tracing conclusions back to source data and testing them against lived experience.
This is where the relationship becomes symbiotic. AI augments human reasoning by extending reach and speed, while humans anchor AI in reality, relevance, and responsibility. This is not a future in which people are replaced. It is one in which they become bionic, where they are better equipped to reason, decide, and act in complex environments than either could alone.
Law as a Model for Context-Driven Intelligence
Nowhere is the importance of context clearer than in the practice of law. Everything in law is fact-pattern based. Facts are the raw material. They allow legal issues to be identified, laws to be applied, and liability to be determined. Without a coherent fact pattern, there is nothing for the law to attach itself to.
Context is not supplementary in law, it is foundational. Without it the law cannot be applied. If machines are to assist meaningfully in legal reasoning, rather than simply accelerating administrative tasks, they must be given access to that context in a form they can navigate, interrogate, and reason over.
This is where information architecture and structural intelligence become essential. They are the means by which legal context is made explicit for humans and machines alike.
From Elite Dealmaking to Alternative Finance and Tax
The relevance of structural intelligence extends far beyond law firms. Alternative finance, private equity, hedge funds, and tax advisory all operate in environments where structure is foundational.
These organisations manage hundreds, sometimes thousands of interconnected entities, where structures evolve constantly, becoming more complex as capital is raised, investments are made, and jurisdictions shift. However, structural information is typically fragmented across systems, documents, and individuals, while decisions must be made quickly and defensibly.
Human pattern recognition does not scale to this level of complexity. Static diagrams fall out of date. Spreadsheets fracture under change. In these domains, making structure explicit becomes the prerequisite for safe automation and the delivery of competitive advantage. When that happens, AI becomes more powerful not because it is smarter in isolation, but because it is grounded in a structural reality it can navigate reliably.
What Professionals Have Been Doing All Along
In reality, professionals have been creating information architectures for centuries. Lawyers, dealmakers, and advisors have always drawn structure charts, timelines, and flow diagrams to make sense of complexity. These artefacts encode relationships, obligations, and dependencies in visual form. They are domain knowledge externalised.
Those structures exist in people’s heads first. Drawing them is both an act of understanding and an act of control. The problem is that this clarity has historically been ephemeral. Created in PowerPoint, spreadsheets, or on paper, it quickly becomes outdated as deals evolve. Meanwhile, the underlying data remains fragmented across documents, systems of record, and individual expertise.
Optimising AI for Competitive Advantage
As AI adoption accelerates, organisations face a choice. They can continue to deploy tools that optimise execution speed, or they can invest in the foundations that allow AI to support strategic decision-making.
Information architecture, and the structural intelligence that results from it, is not an optional enhancement to AI. It is the layer that makes intelligence usable in complex, real-world environments.
You cannot reason safely over ambiguity. If you want AI to help you make better decisions, not just faster ones, you have to give it structure first. In a world racing toward automation, the organisations that recognise this distinction early will be the ones that build durable advantage, not short-term acceleration.
