I read a piece this week arguing that AI adoption sticks when it is hands-on. Give people a real working day on their own problems. Let them fail a bit. Skip the weeks of online modules. Use AI to write the prompts so nobody feels locked out by "prompt engineering". Know what the tool cannot do so you do not waste effort on it. All of it sensible. I agreed with most of it.

Then I sat with the thing it left thin.

Hands-on training with the people who do the work is good. A lot of programmes never get that far. They arrive with a product or a services pitch wrapped in transformation language, promise a new system that will do this and that, and never get into the detail with the people on the ground of what is actually broken and what needs fixing. That is the miss. Not the training. The training is the right kind of contact: it includes the people who carry the work, takes their problems seriously, and brings them onto the journey instead of doing change to them. What has to sit under it is knowing how the work runs today, what better should look like, and which kind of transformation you are even running.

"Transformation" gets used as if it is a single thing. It is not. I count at least four, and they differ in kind, not degree. Get those wrong and a brilliant hands-on workshop just makes people quicker at the wrong work.

One: same business, same shape, cheaper

You do exactly what you do today. You just do it for less. Digitise the paper, automate the manual steps, consolidate overlapping systems, take the cost out. This is where a lot of so-called AI transformation actually sits, and there is nothing wrong with that. It is real money. But it is tuning, not transformation. The business is unchanged. Same machine, running leaner.

Two: same business, completely different way

Netflix rented films. So did Blockbuster. You drove to a shop, or you waited for a DVD in the post, watched it, sent it back, waited for the next one. Netflix stayed in the business of renting films and threw the whole mechanism away. Streaming, subscription, a platform instead of a building. Same job for the customer, which is watch a film tonight. A completely different way of delivering it.

This is not cost-out. It is a reinvention of how the value reaches the customer. Uber did it to taxis. Spotify did it to music. The thing you sell is recognisably the same. Everything about how you make it and move it is new.

None of those moves were possible without leveraging a technology that was already changing the ground under the old model. Netflix needed the internet and enough bandwidth to stream. Uber and Spotify needed smartphones in people's pockets. The business stayed recognisable. The delivery changed because the technology finally let it.

Three: change what the business is

Apple sold desktops and laptops. Then it stopped being a computer company and became an ecosystem. iPod and iTunes, then the iPhone, the iPad, the watch, the store, every device talking to the others. What you were buying stopped being a machine and started being a place to live. That is not a cheaper computer, and it is not a better-delivered computer. It is a different business wearing the old name.

That is also why this is not a tech story dressed up as strategy. Apple did not become an ecosystem by starting with the technology, listing what it could do, and then hunting for a way to sell it. They started with the customer experience they wanted people to live in, and worked back to the technology that made that experience hold together. The products matter. The experience is what the business became.

Four: turn what you built for yourself into the product

There is a fourth kind, and it is easy to miss because it looks like operations from the outside. You build something to run your own business. Then you realise the thing you built is valuable to other people too. Amazon built infrastructure to run its own shop and sold it back as AWS. That is not becoming a new gadget company the way Apple did. It is turning your own plumbing into someone else's platform.

Same pattern shows up whenever an internal capability becomes the offer: the logistics network, the data platform, the operating engine. The business you started with may still exist. A second business has grown out of how you run it.

Before any of it: where does the truth live?

All four kinds share a floor, and it is the part the training rarely names. AI is only as good as the source it reasons over. Point it at a clean, authoritative record and it flies. Point it at a business whose truth is scattered across spreadsheets, inboxes, personal drives and shared folders, and it will answer confidently from whichever fragment it happened to find. Fast, fluent, and wrong.

So the first question is not which prompt to write. It is where the record lives, and whether it can be trusted. Which system is the master for a customer, an order, a price, a contract? Who owns it? In some businesses that question is not always clear, because the answer depends on who you ask. Until it is settled, AI has no firm ground. It just gets faster at guessing.

This is also where people confuse a useful technique with the destination. Retrieval over a pile of documents can help you find things faster than a human ever could. That is real. It does not decide the master. It does not put the fact in the right store at the right time with clear ownership and a line back to where it came from. If the starting move is "leave the mess, put a model on top, and call that AI-first", you have the order wrong. Fit-for-purpose stores and owned masters come first. Then the model has somewhere honest to stand.

Systems thinking earns its place here too. You cannot transform one part in isolation, because the parts are wired to each other. Speed up one step and you usually move the bottleneck downstream, or break something that depended on the old pace. AI makes that worse when it accelerates a single step while staying blind to the whole. Naming where the truth lives, who owns it, and how the pieces connect is the unglamorous work before the workshop. Skip it and the hands-on day teaches people to draw on a source that was never solid.

Why any of this matters for the training

Because AI means something different in each kind.

In the first, AI is mostly automation. You point it at expensive manual steps and you measure cost and cycle time. In the second, AI helps rebuild how the value reaches the customer, and you measure the new experience, not only the old cost. In the third and fourth, AI may sit inside what you now sell. The thing on offer did not exist in that shape before, and you measure value that could not have existed a year ago.

Same technology. Four different intents, with different owners, different measures, and different definitions of done.

So when a workshop treats "adopt AI" as one skill, people can leave better at using the tool and no clearer on what they are trying to change, why, what good looks like, or which kind of transformation they are in. The skill still matters. What it needs is somewhere to attach: a named change, a trusted record, and a view of how the moving parts fit together, not a tidy improvement to one step that ignores what sits upstream and down.

Hands-on is right. It is just not the first move. Get clear what you are trying to change and why, what good will look like, which of these transformations you are running, and where the truth lives. Bring the people who do the work into that, get their buy-in, and let them own their part of it. Then the tool has somewhere to land.

If you want a structured way into that conversation, start with the AI-First Operations Scorecard. It will not choose the tool for you. It will show where you stand on the ground the tool needs: people, process, technology, data, strategy, operating model, and governance.