The Adoption Advantage

The adoption race

The value of general-purpose technologies rarely goes to their inventors.

In 2014, Google bought DeepMind, the most celebrated artificial intelligence company the UK has ever produced, for around half a billion dollars, a fraction of what a frontier laboratory commands today. Arm, the most important chip company the UK ever built, listed in New York rather than London. Graphcore, once the UK's great hope in AI silicon, went to SoftBank in 2024. The companies now inventing this technology are American and Chinese, funded by private investment running at a scale no single European government can match.

Based on that, it would be easy to think that the UK has already lost the AI race and will not catch up. While that argument may be true, it is irrelevant to the long-term success of the economy.

This pattern runs through two centuries of technology development. In 1845, at the peak of railway mania, UK investors were promised dividends of 10 per cent. They were earning 1.83 per cent by 1850, and share prices had collapsed, while the economy the trains served captured most of the value.

Warren Buffett observed that the money made by every American airline in history added up, as of 1992, to zero; the value of powered flight went past the manufacturers and airlines to the passengers, exporters and the hub cities they served.

Malcom McLean invented container shipping in 1956 and was bankrupted by container economics in 1986. The ports of Rotterdam and Singapore, who invented nothing but adopted this new model fastest, became two of the world's largest hubs. Telecoms firms spent billions of dollars laying the internet's fibre in the 1990s. The companies that eventually got rich on the internet were built on top of that infrastructure, while the ones laying cables lost more than $2 trillion collectively.

William Nordhaus examined the American economy across half a century and calculated that innovators keep about 2.2 per cent of the value their innovations create. The other 97.8 per cent flows through to the people and organisations that use the technology. Imitation is fast, patents expire, and better successors arrive.

Complementary assets

Complementary assets are key to value.

In 1971 EMI invented the CT scanner, one of the great medical advances of the century, and the work won a Nobel Prize. Within a decade EMI was out of the scanner business, while GE and Siemens, who invented nothing, took the market. The adopters owned what the economist David Teece called the complementary assets: the factories, the sales forces, the hospital relationships and the trained engineers the invention was useless without. Teece's work became some of the most cited in management research, and its lesson is simple: the question is never only who invents, but who owns the assets the invention depends on to create value.

Adoption is not automatic, and the one thing that reliably delays the payoff is skill. When electric power was commercialised in the 1880s, American factory productivity did not move for roughly 40 years, because factories installed dynamos and kept their steam-age layouts and steam-age habits. The economist Paul David, who documented the lag, noted that until managers redesigned the work and trained the people to run it, the dynamo was everywhere but the productivity statistics.

The value of a general-purpose technology flows to its adopters, but only to the adopters whose people can actually use it. We call that the Adoption Advantage, and it is a race the UK is uniquely positioned to win.

The AI race

AI fits the pattern, and it is moving faster than anything before it.

Apply this same law to artificial intelligence. The inventors are the frontier laboratories and the cloud platforms that operate their infrastructure. The complementary assets, the capabilities the invention is useless without, sit inside every organisation that uses the technology to do something else. Two centuries of evidence says that is where the value will ultimately go.

Every disadvantage in the opening case attaches to the invention race. A missing frontier laboratory stops mattering when the frontier is available to anyone with a credit card: the models are rented by the token, the price of a given level of capability has fallen more than two hundred-fold in eighteen months, and openly published models trail the frontier by months. Expensive electricity stops mattering when the computation happens in someone else's datacentre. The dynamos are shipped. The wiring is done. The question is whether the workforce can run what is connected to the wall.

Speed is a further complication. ChatGPT reached 100 million users within about two months of launch. The container took decades to conquer the ports and the dynamo took 40 years to reach the factory floor; this technology reached a meaningful share of the world's desks in months. Access spread. The skill did not, and the gap between the two is now one of the most measurable facts in working life.

The skill gap

Everyone has the tool and almost nobody has been taught.

In the largest global study of its kind, covering 48,000 people across 47 countries, two-thirds of workers said they now use AI regularly. Two-thirds also admitted they rely on its output without checking whether it is accurate, over half said they had already made mistakes at work because of it, more than half hide their use of AI from their employers, and fewer than half had received any training at all.

Untrained use has a texture, and by 2025 it had acquired a name. Researchers at Stanford and BetterUp surveyed over a thousand American desk workers and found that 40 per cent had recently received what they called workslop: AI-generated work that looks polished and says nothing, shifted onto a colleague who must then untangle it. Each instance cost its recipient nearly two hours, an invisible tax the researchers priced at $186 per person per month, and it cost the sender colleagues' trust.

The same studies document the prize. An analysis of more than 1.3 billion job postings found that roles asking for AI skills paid a premium of 28 per cent, nearly $18,000 a year, and that half of those postings were outside IT entirely: lawyers, marketers, accountants and analysts. Employers have worked out that the skill is scarce and valuable. What they cannot yet do is tell who has it.

At the level of whole organisations, the results confirm where the gap bites. A 2025 study from MIT found that around 95 per cent of company AI pilots had shown no measurable effect on profit, and RAND Corporation's interviews with experienced practitioners put the failure rate of AI projects above 80 per cent, roughly twice that of ordinary technology projects. By the end of 2025, 25 per cent of UK businesses were using AI, up from fewer than one in ten two years earlier, but adoption at scale had not yet translated into productivity gains at scale.

Why certification

Competence can be defined, taught and proven.

Using AI well is a workplace capability like any other. A competent person can select the right tool for a task, direct it the way they would brief a capable colleague, check its output in proportion to what a mistake would cost, and operate within the rules that keep an organisation safe, while protecting data and staying accountable for the outcome. A capability that can be defined that clearly can be taught, and a capability that can be taught can be tested and certified.

We have made workplace capabilities certifiable before, and the most familiar example tracks this moment closely. When the motor car arrived, the UK spent three decades handing the technology to anyone who wanted it: from 1903 a driving licence could simply be bought over a post office counter, with no test attached. By 1934 more than 7,000 people were dying on UK roads every year, with a fraction of today's traffic. The compulsory driving test arrived in 1935, and it did not slow adoption; it made it sustainable. Certifying the capability did not reduce the number of drivers. It changed what kind of drivers there were.

The closest precedent for desk work comes from the last general-purpose technology to land on one. In the 1990s, as computers arrived in workplaces ahead of the skills to use them, Europe created the European Computer Driving Licence, a plain, portable certification of basic computing competence. It went on to certify some 17 million people through more than 70 million tests in over 100 countries. AI has spread faster than the computer did, and no equivalent yet exists.

The gap is becoming legally awkward as well as commercially expensive. Since February 2025, European law has required organisations that deploy AI to ensure their staff have a sufficient level of AI literacy, a duty that reaches any business whose systems or output touch the European market, with enforcement beginning in 2026.

The UK government has set a target of upskilling 10 million workers in AI by 2030. There are standards for organisations, notably ISO/IEC 42001, which certifies that an organisation has an AI management system in place. What does not yet exist is a plain, portable, independently assessed certification of what an individual worker can actually do with AI in their job. That is what the AI Skill Centre was built to provide.

What we built

The AI Skill Centre is that missing infrastructure, built.

The AI Skill Centre is an independent, industry-funded certification body, and it supplies the three things a competence system needs.

We built this first for mid-size organisations of roughly 50 to 1,000 employees, because that is where the gap between AI exposure and AI support is widest: they face the same data exposure, AI-enabled fraud and regulatory duty as large enterprises, with none of the in-house capability to manage them.

The five certification levels

Our mission

The mission is 10,000 people and 150 organisations in the first year.

That target is deliberately public. Every certificate on the register is one more person who can use the most capable tool they have ever been handed, safely and accountably, and every accredited organisation sets the benchmark its sector gets measured against. Our longer-term ambition is that verified AI competence becomes as ordinary, as expected and as portable as a driving licence, in every sector where this technology now sits, which is fast becoming every sector there is.

The history in this paper suggests the stakes. Singapore and Rotterdam were not the great ports of the world before the container arrived; the league table was redrawn by adoption, and the ports that hesitated never got their places back. General-purpose technologies redraw the table once in a working generation, and this one moves faster than any of its predecessors. We believe the organisations that win will not be the ones that bought AI first, and they will certainly not be the ones waiting for the invention race to resolve itself. They will be the ones whose people can actually use it: safely, accountably, and at scale.

The Adoption Advantage is real, it is measurable, and it is available to any organisation whose workforce can demonstrate genuine competence with the tools they already have.

Anyone can start claiming it today. The five-minute self-assessment is free at aiskillcentre.com, the standard is published in full at aiskillcentre.com/standard, and organisations can arrange a free accreditation consultation at hello@aiskillcentre.com.