Enterprise AI is entering a new phase.
The initial wave focused on making data accessible, deploying large language models and experimenting with generative AI. The next wave will be about enabling AI to understand the business, interpret context, navigate relationships and take meaningful action.
This is where the resurgence of semantic architecture becomes particularly interesting.
As highlighted in Corinium Intelligence’s article, The Semantic Comeback: Why Enterprise AI Is Forcing a Rethink of Data Architecture, semantic layers, ontologies and knowledge graphs are moving back into the spotlight.
And for good reason.
The problem was never just about accessing data. It was always about understanding what the data means.
Consider a seemingly simple question: Who is our most valuable customer?
Sales might define value through revenue. Finance might look at profitability. Customer Success might consider lifetime value and retention. Operations might factor in service complexity.
All four perspectives could be correct within their respective contexts.
But what happens when an AI agent is expected to make a recommendation or take action based on these conflicting interpretations?
Without shared business context, AI can become very good at producing answers that are technically plausible but operationally wrong.
The next competitive advantage in Enterprise AI may not come from bigger models or more data, but from how effectively organisations capture, connect, govern and operationalise their institutional knowledge.
Data architecture to Knowledge architecture
Traditional enterprise data architecture has largely focused on collecting, integrating, storing and distributing data.
That foundation remains important. But it is no longer sufficient for an AI-driven enterprise.
We need an additional architectural capability that connects three dimensions:
Semantic layers that establish consistent business definitions, metrics and calculations.
Ontologies that describe business concepts, their properties, relationships and rules.
Knowledge graphs that connect entities and relationships across fragmented enterprise systems.
Together, these capabilities can help AI move beyond retrieving information to interpreting it within the context of the business.
This is a fundamental shift.
The ambition should be to move from an enterprise that has data to an enterprise that can make its institutional knowledge accessible and actionable.
Five steps to bring this to life
This transformation does not require enterprises to rip and replace their existing data platforms. It requires a deliberate evolution of the architecture and operating model.
1. Start with business decisions, not tech
Identify high-value business decisions and workflows where fragmented information, inconsistent definitions or disconnected systems create friction.
Start with a tangible use case rather than attempting to model the entire enterprise.
2. Establish a shared business vocabulary
Bring business, data and technology teams together to define critical business concepts, metrics, relationships and rules.
Create domain-owned semantic models with a common enterprise foundation. Consistency should not mean imposing one definition on every business context.
3. Connect, don’t consolidate
Build semantic capabilities over existing data warehouses, lakehouses, APIs and operational systems.
The objective is to create a connected knowledge fabric, not another centralised technology silo.
4. Make governance intrinsic to the architecture
Embed data ownership, lineage, access controls, quality, versioning and accountability into the knowledge layer.
As AI agents become more autonomous, understanding what they can access, what they are permitted to do and why they made a decision becomes just as important as understanding the underlying data.
5. Prove value and progressively scale
Choose one business domain, establish measurable outcomes and demonstrate improvements in decision quality, AI accuracy, process efficiency and trust.
Use the learning to expand into adjacent domains and progressively build an enterprise-wide knowledge architecture.
Now, Who owns enterprise knowledge?
There is another dimension that deserves attention.
Enterprise knowledge is not confined to databases. It is embedded in business rules, operational processes, policies, contracts, employee experience and decades of institutional decision-making.
Much of this knowledge is undocumented, fragmented or dependent on individuals.
Building a semantic layer without addressing this institutional knowledge gap risks creating a technically elegant architecture that still lacks business context.
This calls for a stronger partnership between business leaders, enterprise architects, data leaders and AI teams.
It also raises an important organisational question: Should enterprise knowledge become an explicitly governed asset, with clear ownership and lifecycle management, much like enterprise data?
I believe it should.
A different kind of competitive advantage
As foundational AI models become increasingly accessible, differentiation will increasingly depend on what enterprises bring to those models.
Their proprietary knowledge. Their business context. Their operating intelligence. Their ability to connect information across organisational boundaries.
The organisations that get this right will potentially build AI systems that are not just smarter, but more relevant, reliable and aligned with how their businesses actually operate.
Perhaps the next evolution of enterprise architecture is not simply about building AI-ready data platforms.
It is about building knowledge-driven enterprises that are genuinely AI-ready.
And that leads to one question worth debating:
Are enterprises investing enough in making their institutional knowledge understandable to AI, or are we still primarily focused on making more data available to it?
I’d be interested in hearing how others are approaching this challenge.


