Research Paper No. 001
Why the biggest AI opportunity is not building models. It is deploying them. And why most organisations are failing at it.
There is a widespread misunderstanding at the heart of how organisations think about artificial intelligence. They believe that access to a powerful model is access to AI capability. It is not.
A foundation model is infrastructure. It is the electricity grid, not the appliance. The organisations that are generating real value from AI right now are not doing so because they have access to better models than their competitors. GPT-4, Claude, Gemini, and Llama are all broadly accessible. The differentiation is not in the model. It is in the deployment.
Deployment means understanding which specific workflows within a specific organisation can be transformed by AI, and then building the systems, processes, and human capability to actually make that transformation happen. It means connecting a general-purpose model to a specific business context with specific data, specific integrations, and specific people who know how to use it.
This is not a technical problem. It is an organisational one. And it is the problem that almost no one has solved at scale.
The numbers are stark. McKinsey estimates that AI could deliver $4.4 trillion in annual productivity value globally. The World Economic Forum projects that 85 million jobs will be transformed by AI by 2030, while 97 million new AI-era roles will emerge to replace them. The IMF finds that one in four jobs will be fundamentally changed. PwC reports that 74% of CEOs identify the AI skills gap as their single greatest threat.
These numbers describe enormous potential. They do not describe enormous achievement. The reason the gap between potential and achievement exists is the deployment problem.
Across the organisations we have studied and worked with, the deployment failure follows a consistent pattern. It is not random. The same five barriers appear in almost every case.
Understanding these barriers matters because organisations that recognise them can address them directly. Those that do not tend to invest in more powerful models, more expensive consultants, or more elaborate technology stacks, none of which solve the underlying problem.
The distance between where organisations believe they are on AI adoption and where they actually are is one of the most consistent findings in applied AI work. The gap between aspiration and reality is larger than most leadership teams realise.
Government and public sector organisations sit at the bottom of every AI readiness measure. This is not because AI is less applicable to government operations. It is because the barriers to deployment are compounded by procurement requirements, data sovereignty obligations, compliance frameworks, and a structural shortage of technical talent willing to work in the public sector at public sector salaries.
The irony is that government is where AI can deliver some of its most significant social impact. Faster citizen services, more efficient public administration, better resource allocation, and more responsive policy implementation are all achievable with current AI capability. The constraint is deployment, not technology.
The same pattern holds for SMBs. Small and medium businesses represent the majority of economic activity in most countries, but they have the least access to the AI deployment expertise that larger enterprises can afford. The result is a compounding disadvantage as AI-enabled enterprises increase their productivity gap over AI-excluded SMBs.
In 2026, OpenAI acquired Tomoro AI. Anthropic announced a $1.5 billion deployment joint venture. Microsoft deepened its enterprise AI deployment partnership network. Every major frontier lab is making the same move: buying or building deployment capability.
This is not a coincidence. The frontier labs have solved the model problem. They have not solved the deployment problem. And they know it. The models exist. The infrastructure exists. What does not exist at scale is the human and organisational capability to put those models to work inside real institutions with real data and real constraints.
The acquisition of Tomoro by OpenAI's deployment company is the clearest possible signal of where the value has shifted. Tomoro did not build a better model than OpenAI. They built better deployment capability. That is what OpenAI paid for.
This pattern will repeat. Every frontier lab needs deployment partners, deployment companies, and deployment infrastructure across every geography and every sector. The companies being built right now that focus on applied AI deployment, particularly in underserved geographies and sectors like government, SMB, and workforce development, are building toward an acquisition market that is only going to grow.
The question for any organisation thinking about AI is no longer whether to engage with it. The question is whether to engage with it now, while the deployment infrastructure is being built and the competitive positions are being established, or later, when those positions are locked.
There is a version of the AI deployment problem that is purely technical. Build the integrations. Deploy the models. Automate the workflows. That version is real, but it is the smaller half of the challenge.
The larger half is human. The World Economic Forum projects that 85 million jobs will be displaced by AI by 2030. The same research projects 97 million new roles will emerge. The net number is positive. The transition is not automatic.
Those 97 million new roles require skills that the current workforce does not have in sufficient quantity. The educational systems that would typically supply those skills operate on 3 to 5 year curriculum cycles. The job market is moving faster than the training pipeline.
This creates a specific and urgent problem for governments, employment agencies, and workforce development institutions. The workers exist. The jobs are emerging. The pathway between them is missing.
Building that pathway is not just a social good. It is a significant economic opportunity. Governments in Norway, the UK, Germany, and across the Nordics are actively seeking organisations that can deliver structured, measurable AI workforce development at scale. The contracts are real. The funding is available. The programme designs do not yet exist in sufficient numbers to meet the demand.
The deployment gap is real, but it is not permanent. Organisations that act systematically in the next 12 months will establish advantages that compound for the following decade. The following recommendations are drawn from field experience deploying AI across enterprise, government, and SMB contexts.