Every large organisation I talk to has run its AI pilot. Some have run a dozen. Very few have changed how anyone actually works on a Wednesday lunchtime.
This is all too familiar for me – the gap between a thing that works and a thing people use. Long before AI, I was handed a near-impossible brief: launch 80 videos in 10 weeks for NHS UK, the new national health platform.
I still remember the scramble beneath the spreadsheets – editing a three-minute film on earache while waiting for videos on chemotherapy and shingles to come in (talking doctors, talking patients). What stopped it from feeling like a sausage machine was the quality and generosity of the interviewees, and of my colleagues.
The earache film was three minutes because we’d worked out that a parent at 2am will not watch five.
The mad scramble paid off. The site wasn’t perfect, but it got better, and fast. Before we knew it, it had become Europe’s largest public health platform. Fifty million visits a month.
But here’s the part I still think about.
It didn’t work because the technology was clever, or because we had more videos than anyone else. It worked because someone could finally understand something about their own health – plainly, in the moment they needed it. The 2am search from someone frightened about a symptom – their own, their child’s, a loved one’s – and, for once, a clear answer at the end of it.
That lesson has held through every job since – a bank serving 30 million customers, a healthcare business covering 141 million members. It never changes.
People don’t adopt something because it exists. They adopt it because someone made it make sense.
And that’s still the whole game with AI. The pilot was never the hard part. The people who carry it forward are.
Which means the questions most organisations are asking about AI are the wrong ones.
Not “does the model work” – it almost certainly does. Not “what’s the use case” – there are forty, and that’s the problem. The question is: who, specifically, is going to change how they work on a Wednesday lunchtime, and what do they need to believe before they will?
That’s a communications problem wearing a technology costume. And it’s usually owned by nobody. The pilot has a sponsor, a budget and a deadline. The adoption has a slide.
The NHS videos worked because someone had decided, in advance, what a frightened person at 2am actually needed – and then held everything to that. Not what we could produce. What they could use.
Most AI programmes have never had that conversation. They have a capability and a hope.
So here’s the test I’d apply. Can someone three levels down tell you what changes for them, in a sentence, without using the word “AI”? If not, you don’t have an adoption problem yet. You have a story problem, and it will become an adoption problem in about six months.