The 250-year pattern your AI strategy ignores

I started the Oxford Artificial Intelligence Programme recently. The first question I faced sounded like economic history, but turned out to be a marketing strategy question: “Why do some technologies take decades to pay off when the technology itself works on day one?”

Steam engine, electricity, the mobile phone. Every general purpose technology of the last 250 years followed the same curve. The technology arrived, but the value wasn’t visible until businesses rebuilt themselves around it. Which brings me to the question for this issue: why do we keep expecting AI to be the exception?

The data says we do expect it, and we are being disappointed. 95% of enterprise integrated AI pilots are delivering no measurable return on the P&L (MIT NANDA, The GenAI Divide, 2025). Half of generative AI budgets are going into sales and marketing. We are the biggest buyers, but not the biggest winners.

AI is not just another tool to buy. It is a general purpose technology, in the same category as the steam engine or electricity (Bresnahan and Trajtenberg, 1995). A GPT spreads everywhere, keeps improving, and sparks more innovation around itself; most of its value comes from the complementary innovations built on top of it.

The 250-year pattern

Electricity: 20 years of flat productivity. Factories used electric motors for roughly two decades before they gained real productivity advantage. The gains only arrived once managers redesigned the factory floor around the motor instead of fitting it into the old layout.

Mobile: the handset wasn’t the value. By 2023, mobile technologies contributed about US$5.7 trillion to global GDP, 5.4% of the total (GSMA, 2024). That number is not the phone itself; it is the networks, app stores, payment rails and business models built around the phone over twenty years.

Brynjolfsson, Rock and Syverson gave this pattern a name: the productivity J-curve (American Economic Journal, 2021). Early in a general purpose technology’s life, measured productivity looks flat or even negative, because organisations are investing into intangible complements — data, processes, skills, new ways of working — that do not show up on the books yet. In time, the curve goes up as the investment pays off.

Today, most AI investments are at the bottom of the curve. The tools are bought, the complementary work is skipped, and teams conclude that AI does not work for them.

The McKinsey data says the same thing. 88% of organisations report using AI in at least one function, but only 6% are capturing real enterprise value from it (The State of AI, 2025). BCG puts numbers on where the value actually comes from: 10% algorithms, 20% technology and data, 70% people and processes. That 70% is the J-curve we need to uncover.

What the complementary investment is

Before you plan for your next AI tool, here are the four questions that need answering.

• Data. By the end of 2025, over 50% of generative AI projects were dropped after proof of concept, with poor data quality named first among the causes (Gartner, 2026). Do you have usable brand style guides, content taxonomies and a basic customer data model that feed your AI tool?
• Workflow. The high performers are 2.8 times more likely to have fundamentally redesigned how work happens (55% of them, against 20% of everyone else), not just added AI into the existing process. Are you redesigning the workflow around AI, or simply dropping AI into an old process?
• Governance. Who owns the output, who signs off, who escalates when the model is wrong. AI fails in large organisations when accountability is unclear. For each AI use case, can you name a single accountable owner and a clear escalation path when the system fails?
• Talent. Someone has to do the redesign, run the human-in-the-loop review, and hold the standard. That capability is the complementary investment that takes longest to build. Have you redesigned your operating model with the right human-agent mix, and do you have a concrete plan to upskill the people who need to use these tools?

All of the above is the reason behind the 6%. Every general purpose technology has rewarded the ones who rebuilt the work around it — the factory owners who redesigned the floor, the businesses that grew up around the mobile phone, the companies that reorganised around the internet. Everyone has access to the same tools, but the advantage goes to whoever does the slow complementary work to earn the payoff.

← All insights