AI did not suddenly get clever. Three things that had been missing for fifty years finally turned up at the same time. That is the whole of "why now", and it matters, because it tells you this wave is not magic. It is mechanics. Mechanics can be learned, hired for, and pointed at a commercial problem.
The hype wants you to believe otherwise. Magic is easier to sell, and easier to be scared of. Both reactions cost you money.
The idea itself is old. Researchers named the field in 1956, then spent decades overpromising and underdelivering, through two long funding droughts the industry still calls the AI winters. The theory mostly worked. The world around it had not caught up.
But AI was never just theory. Machine learning, the engine under most of what we call AI, has quietly run parts of your life for decades. The spam filter on your inbox. The fraud check that clears your card at the till. The recommendations on Amazon and Netflix. The credit score behind your mortgage. All machine learning, much of it working away since the 1990s, almost all of it invisible. Arthur Samuel had a computer teaching itself to play draughts back in 1959. None of this is new.
What was missing was range. Each system did one narrow job and nothing else. The fraud model could not write you an email. The recommendation engine could not hold a conversation. Then in 2017 a team at Google published a paper called "Attention Is All You Need" and introduced the transformer, the architecture that now sits under almost every model you have heard of. It took AI from narrow, single-purpose tools to systems that can do things most of us thought were years away, or flatly impossible: hold a real conversation, write working code, generate an image from a sentence. The science was decades old. The transformer changed what it could do.
Five years later ChatGPT put it in front of everyone, and AI stopped being a lab curiosity and became a tool your customers use daily.
What happened next is not subtle. Corporate AI investment hit $252bn in 2024, more than thirteen times the level of a decade earlier. The share of organisations using AI jumped to 78% in a single year, up from 55%. This is not a fad waiting to pass. It is infrastructure being laid.
Here is the part the hype skips. Adoption has raced ahead of return. In the same research, most companies seeing any financial impact report cost savings under 10% and revenue gains under 5%. Marketing and sales is where people report the most upside, 71% of users saw revenue gains, and even there the most common increase is below 5%. The technology is real. The results, for most, are still thin.
That gap is not a technology problem. It is a people problem, and it is the most useful thing I have learned about AI.
I consider myself an early adopter. It interested me enough that I left the corporate world and spent a solid year studying it. The clearest lesson: put a commercial, strategic or marketing mind next to AI and it gets really interesting. Put a pure technologist next to the same tools and it often fails on impact. The science is necessary. On its own it is not enough.
A small example. Last year I asked a developer to build me a chatbot. Thirteen weeks in, the headline update was that it was going to take a long time. He was, in his own way, closed to the tools. Once I understood the science well enough to brief it properly, I built it myself in two weeks. Same technology, very different result. The difference was not coding talent. It was knowing what to ask, and why.
So where does this go. Not, I think, to AI quietly replacing your marketing team. It goes to a widening split between businesses whose leaders understand this and the ones who outsource the understanding and hope for the best.
AI is logic. It is brilliant at the measurable, the repeatable, the optimisable. But the things that actually move a brand, why someone pays more, stays loyal, or picks you over a cheaper rival, are not fully logical. They are closer to what Rory Sutherland calls alchemy, the parts of human behaviour that never show up in the data. Hand those to an optimisation engine on its own and you get efficient mediocrity. The version that works is human plus machine: the machine for scale and speed, the human for judgement, meaning and taste.
Which is why I would push every business leader to do the unfashionable thing and get properly close to this. Not to become an engineer. To understand the science well enough to aim it. The leaders who treat AI as someone else's department will keep paying for tools that return 5%. The ones who become a little obsessed with it will build businesses that are hard to catch.
AI is not magic. It never was. It is the most capable tool we have built, and like every tool before it, it rewards the people who bother to understand what it is actually doing. The advantage was never going to sit in the machine. It sits with the people who know what to ask it.
Sources
- Vaswani et al., Attention Is All You Need, 2017 (the transformer architecture behind modern models).
- Stanford HAI, 2025 AI Index Report, Economy chapter ($252.3bn corporate AI investment in 2024, more than thirteenfold growth since 2014, organisational AI use up to 78% from 55%, most financial impact under 10% cost savings and under 5% revenue gains, 71% in marketing and sales reporting revenue gains).
- Rory Sutherland, Alchemy: The Dark Art and Curious Science of Creating Magic in Brands, Business and Life (2019), via Chay's own paper, Behavioural Economics and Artificial Intelligence: Logic versus the Alchemy of Human Decision-Making.
- Background, not sourced to a link: the 1956 naming of the field, Arthur Samuel's 1959 self-learning draughts program, the AI winters and the November 2022 launch of ChatGPT are well-established history.
Ready to move the numbers.
Book a 30-minute impact discovery, no obligation. A clear read on where brand, e-commerce and AI can move growth.