When Knowledge Drives AI
What makes intelligence competitive?
Two media companies can use the same AI models, drawing on increasingly capable technologies available to almost any organisation. Yet the value they create from those technologies need not be the same.
That's because competitive advantage has never depended on technology alone.
For media businesses, it comes from a combination of intellectual property, creativity, audience understanding, commercial relationships, operational expertise and experience accumulated over many years.
AI creates an opportunity to make more of that knowledge operational—and potentially extend its influence well beyond the people and processes through which it has traditionally been applied.
I think this deserves more attention in discussions about AI strategy.
The opportunity isn't simply to give AI access to organisational information. It's to engineer the knowledge that makes an organisation distinctive into the intelligence that helps drive its business.
From what we know to how we think
Organisational knowledge is much more than the information captured in documents, databases or operational systems.
Consider an experienced advertising operations specialist investigating an unexpected fall in revenue.
Their expertise isn't simply knowing how the advertising systems work. It includes recognising the significance of particular symptoms, deciding what to investigate, interpreting relationships between observations, testing alternative explanations and determining which actions might resolve the problem.
Much of that expertise comes from experience. It may be partially documented, but significant aspects of how it is applied remain embedded in people's working practices and judgement.
Now consider the possibility of engineering that knowledge into something an AI-enabled system can apply.
Not merely retrieving the expert's documentation, but capturing and engineering the business understanding, diagnostic reasoning, decision criteria and accumulated experience that guide their actions.
This doesn't mean capturing every aspect of human judgement or prescribing every step an AI system must take.
It means making relevant expertise sufficiently explicit that it can help guide how AI interprets situations, evaluates evidence and approaches problems.
Knowledge becomes more than something AI can access. It becomes something that helps direct how AI operates.
A lesson from developing with AI
I've been exploring this relationship through my own return to hands-on development, working on an AI-enabled diagnostic capability in streaming monetisation.
One recurring challenge has been the distinction between a plausible explanation and a conclusion justified by the available evidence.
AI models can be remarkably capable of interpreting technical observations and proposing explanations. But those explanations don't necessarily reflect the full business context, nor are they always supported by what has actually been established.
The challenge isn't always insufficient information or an inadequate model.
It can be the knowledge underpinning the interpretation: understanding what events mean, how they relate to business outcomes, how an experienced specialist would approach the investigation, and what evidence justifies a particular conclusion.
As I experimented with different AI models, I was also progressively developing the knowledge framework intended to support their reasoning.
This became an iterative process. Different models interpreted and applied the knowledge with varying degrees of success. Understanding those differences helped us assess both the capabilities of the models and how effectively the knowledge was represented and applied.
One of the more important lessons has been that developing intelligence for a particular business purpose involves engineering more than the AI model itself.
It also involves engineering the knowledge that gives the model's capabilities direction, meaning and criteria against which its conclusions can be evaluated.
An established discipline, with new possibilities
None of this suggests that knowledge engineering is new.
The discipline has existed for decades, particularly through expert systems and knowledge-based computing. Its underlying ambition was to represent specialist expertise in forms that computer systems could apply.
Knowledge has also long been recognised as fundamental to competitive advantage. Robert Grant's knowledge-based theory of the firm, published in 1996, explored how organisations develop capabilities through the integration and application of specialised knowledge.
What contemporary AI potentially changes is the practicality and economics of applying that knowledge.
Traditional approaches often required extensive efforts to formalise expertise and prescribe how systems should respond. Modern AI models can interpret information and reason with greater flexibility, potentially allowing explicitly engineered knowledge and adaptable AI capabilities to complement one another.
There is also evidence of the value of spreading expertise. A 2025 study of generative AI in customer support found particularly substantial productivity improvements among less-experienced workers, suggesting that AI can help disseminate practices associated with more experienced colleagues.
That doesn't establish a general competitive advantage from knowledge engineering. But it illustrates the potential value of making expertise more widely applicable.
The opportunity is therefore not simply to preserve what organisations know. It is to make more of that knowledge available for intelligent application, at a cost and scale that can deliver meaningful business value.
Engineering competitive advantage
For media companies, the implications extend across content investment, audience development, distribution, advertising and commercial operations.
Consider two organisations with access to comparable AI technologies.
One uses AI primarily to improve productivity and automate existing activities.
The other also makes a deliberate effort to capture and engineer the specialist knowledge underlying its most valuable decisions and operational capabilities.
Over time, the second organisation may be able to apply more of its accumulated expertise across a wider range of situations, improve the consistency of decisions and build on lessons from previous outcomes.
The distinction isn't necessarily the sophistication of the AI technology. It's the organisational intelligence being developed around it.
But engineering knowledge isn't automatically valuable. The resulting capabilities must improve decisions, operations or commercial outcomes sufficiently to justify the effort required to develop and maintain them.
And knowledge doesn't remain useful simply because it has been captured. Business circumstances change, experts learn and established practices need revision. The knowledge underpinning AI-enabled capabilities must evolve accordingly.
This makes knowledge engineering more than a technology discipline. It becomes an organisational capability, bringing together domain expertise, business leadership and technology.
It also raises an important question of ownership. Organisations should be able to maintain and develop their engineered knowledge as an asset in its own right, with the flexibility to apply it through different AI technologies as those technologies evolve.
Portability is useful, but the greater value lies in accumulating, refining and applying expertise that reflects the organisation's particular strengths.
AI technology providers will continue to develop increasingly sophisticated capabilities. Organisations should benefit from that progress.
Their opportunity is to develop the distinctive knowledge that makes those capabilities increasingly valuable to their own businesses.
Knowledge has always contributed to competitiveness. In the AI era, how effectively an organisation engineers and applies that knowledge could become an increasingly important source of advantage.
The models may be widely available. The organisational intelligence developed around them need not be the same.
