You Can Buy the Transistor. You Can't Buy the Radio.

In August 1953, a small, near-broke Tokyo company called Tokyo Tsushin Kogyo sent one of its founders, Akio Morita, to New York with a single mission: to buy a licence for the transistor. Western Electric was willing to sell. The price was $25,000. Morita paid it. So, it should be said, could almost anyone: the licence was on open offer to any firm with the fee.

What happened next is the part everyone forgets. The licence was not enough. Neither was the famous Bell Labs "cookbook" of instructions that came with it. When the engineers back in Tokyo tried to build a transistor good enough for a commercial radio, they found the recipe simply did not reproduce a working production line. They had to pull apart what they thought they knew and rebuild it from first principles. But they had one thing the fee could not buy: a base of hard-won knowledge in magnetic tape and precision manufacturing that let them make sense of what they had bought. That company renamed itself Sony.

The lesson is not that Sony was clever. It is that the technology was available to everyone, and the capacity to absorb it was not.

What absorptive capacity actually is

The term comes from a 1990 paper by Wesley Cohen and Daniel Levinthal: a firm's ability to recognise the value of new information, assimilate it, and apply it to commercial ends. Their most important finding was almost unsettling in its simplicity — this ability depends overwhelmingly on prior related knowledge. You can only absorb what you are already, partly, equipped to understand. Knowledge is not poured into an empty vessel; it is grafted onto what is already growing.

A later refinement (Zahra & George, 2002) broke the capability into four moves that map cleanly onto a factory floor:

Notice what none of these four are: none of them is buying the technology. Buying is the easy 5%. The other 95% is human work — and it is the work that decides whether your investment joins the winners or the graveyard.

Why this is your problem, specifically

Across the economy, roughly one-fifth of firms are capturing something like three-quarters of the measurable value from AI (PwC, 2026). In manufacturing specifically, the World Economic Forum's long-running study finds that more than 70% of companies investing in advanced analytics, AI or digital solutions never move a single project beyond the pilot (WEF & McKinsey, 2025). The industry has a name for it: pilot purgatory.

This is not a technology-access story. The 80% who stall have access to exactly the same tools as the 20% who soar. AI's productivity gains are conditional on complementary capabilities — the skills, routines and organisational learning that let a firm absorb the tool (Li & Liu, 2026). The technology is necessary and nowhere near sufficient.

If you run a brownfield plant in South Africa — or Manchester, or Ohio — this should land as good news and hard news at once. Good, because the thing that separates the winners is not a bigger cheque or a better vendor; it is a capability you can deliberately build. Hard, because you cannot buy it.

The systems view: absorptive capacity is System 4, working

In Stafford Beer's Viable System Model, System 4 is the outward-and-forward-looking function — the organ that scans the environment and brings new knowledge in. Absorptive capacity is, quite precisely, System 4 doing its job well.

But here is the trap, and it explains pilot purgatory exactly. A plant can have a lively System 4 — an innovation team, a data-science pilot, an eager engineer — and still fail completely, because System 4 on its own only recognises and acquires. It does not integrate. Beer's model turns on the homeostat between System 4 and System 3 (the running of today's operation), held in balance by System 2's coordination. A pilot that never touches System 3 is a pilot that never becomes how the work is done. It wins an award and dies.

This is why the newest research is so people-centred. Studying a connected-worker programme across four factories between 2023 and 2025, van Dun, Weritz and Kumar (2026) found absorptive capacity was not built by the technology, the data platform or the IT function. It was built — or starved — by leadership and social-integration mechanisms: the concrete, human ways knowledge actually moves between people. The absorbing is done by people, through other people. It is a leadership act, not an IT deployment.

The plant that treats AI as a purchase is asking System 4 to do the whole job. The plant that treats it as a capability to build is wiring System 4 into System 3 — through its people. Only the second one scales.

What works, and what doesn't — on real floors

What doesn't work: buying the capability. The canonical failure is the plant that treats AI as procurement — signs the contract, takes delivery of a model built by a consultancy, and hands it to a floor that had no part in making it. It is the Bell Labs cookbook without Sony's prior knowledge: technically fine and organisationally orphaned. Within a quarter it is a screen everyone has learned to ignore. Worse is deploying onto a hollowed-out floor, where the process knowledge has already eroded. You cannot absorb what you were never equipped to understand.

What works: building the capability deliberately.

• Upskill the frontline first. Fully 75% of the WEF's Lighthouse factories made frontline upskilling a priority — because absorptive capacity lives in prior knowledge, and prior knowledge lives in people. They built the base before, not after.
• "Assetise" every win. The Lighthouses turn each successful pilot into a reusable, documented playbook that travels to the next line, the next site, the next region — the difference between a pilot that dies a hero and a capability that compounds.
• One South African plant, one un-heroic habit. The most effective thing I have watched a plant manager do with an AI quality model was not technical at all. He refused to let the vendor "hand it over." He insisted the two operators who knew the line best sat inside the build, questioning every alert until they could explain it in their own words — then made those two responsible for teaching the next shift. His model was no better than anyone else's. His absorption of it was.

Five moves to build absorptive capacity

• Audit the prior knowledge before you buy the tech. Do we have people who can even recognise what this system is telling them? If not, that — not the software — is your first project.
• Never accept a black box. Pair every deployment with the operators who must live with it, inside the build, until they can explain it in their own words.
• Fund the absorption, not just the licence. A licence with no assimilation budget is a $25,000 cookbook.
• Assetise every win. The moment a pilot works, give someone the job of carrying it to the next line. The spread is the return.
• Make leaders the carriers. Name, each quarter, who is responsible for moving knowledge between people — and hold them to it as you would any operational target.

The bottom line

Everyone could buy the transistor for $25,000. The licence was never the moat. AI is the transistor of our moment: astonishingly powerful, increasingly cheap to licence, and worthless without the organisational capability to absorb it. That capability is not on the vendor's price list. It lives in your people, it is built by how they work together, and it is the one thing your competitors cannot copy off the shelf. You can buy the transistor. Only your people can build the radio.