Handing a capable model to an untrained team does not produce much.
One person uses it to write an introductory email. Another uses the same model, on the same licence, to turn thousands of customer feedback points into a product strategy. That is roughly a tenfold difference in what comes out, and none of it is the tool.
What the difference actually is
It is the prompt, and treating a model like a search box is the expensive version of the mistake.
Getting a professional result means giving it context on purpose. Assign it a role. State the constraints of the task. Say what the deliverable should look like. Give it your objectives and who the work is for before you ask for a strategy, and it stops guessing at the things you already knew.
That precision is what removes the loop of rewriting mediocre drafts, which is where most of the time actually goes.
Why this is a business problem
- Weak prompting burns compute and the employee time the tool was bought to give back
- Good prompting spreads one person's way of thinking across a whole team
- Neither happens on its own, so somebody has to teach it
Prompting is closer to a business skill than a technical one at this point. The organisations getting real returns are not the ones with better models. They are the ones that trained their people.
What to do about it
Treat it the way you would any other tool that costs money and needs a competent operator. We run this with teams who want their AI spend to show up in the work rather than on the invoice.