Do you need Claude training? Four questions first

Before you buy Claude AI training, from me or from anybody else, four questions. They are the same ones I ask on a first call. If the answer to all four is yes, you do not need training. You can shut this page and keep the budget, and I would rather you did that than paid for a day that changes nothing.

The four questions

• Is more than a third of your team getting useful work out of it every week? Not signed in, and not curious. Useful work, weekly.
• Can the people who are good at it explain what they do differently? If your best user shrugs and says they have a knack for it, the knack is not going anywhere.
• Has anyone written down what you will not use it for? One page is enough.
• Did the last workflow anyone built survive a month? Plenty of good ideas die in week three.

There is nothing I would add to four yeses. Most teams I speak to manage one, occasionally two.

What the questions are testing

They test whether skill is spreading on its own. Some of it does, given long enough. Of the 9,700 people Anthropic surveyed for its Economic Index, 57% felt their skills had risen in market value, and the ones doing the more automated work came out more optimistic about pay and job security, not less.

Nobody prices what that slow curve costs while it runs. Left alone, a team takes half a year to learn by accident what a few deliberate sessions would settle in a fortnight, and it learns unevenly, through whoever happened to persist. That is the whole argument for training, and it is narrower than most people selling it will make.

What good training looks like, and what to avoid

Avoid anything sold on a certificate. A certificate says a person attended, which nobody was in doubt about. Avoid generic exercises, because prompting a fictional coffee chain teaches your legal counsel nothing about a supplier contract. Above all, avoid the one-off workshop with no follow-through: enthusiasm on the day, silence by the third week, and an invoice that already cleared.

Good training runs on your documents and your deadlines. It leaves behind prompts your team wrote themselves, a written note of what is out of bounds, and a return visit once the workflows have met a few real weeks and started to creak. The follow-through is the part that separates a training spend from a training result.

When the answer is still no

Four noes does not automatically mean you should book something. Three situations come up often. Nobody has agreed what the tool is for, so training sharpens twelve people on a question no one has set. The leadership team does not use it and does not plan to, which no amount of teaching survives. Or nobody has settled what may safely be typed into it. Each of those is free to fix, and none of them is a training problem.

If your answers were mostly no, start with the two that are yours. Write the one page on what this will not be used for, which takes an afternoon and which nobody has ever regretted having. Then put your two best users in front of everyone else and make them narrate a real task out loud, start to finish, including the bits they would normally skip. Both are free. Neither needs me.

The other two do not fix themselves. A knack nobody can describe does not spread, and a workflow that died in week three will die in week three again. Those are the two I get called about, and they are the two that are still costing you something in November.

So: three lines about which of the four questions you failed. I will tell you whether a programme moves it. If the honest answer is that you have an agreement problem rather than a skills problem, I will say that instead, and you will have spent four minutes finding out.