Find who is already paying for ChatGPT before you post a req.

Find who is already paying for ChatGPT before you post a req.

You need people who can build, people who can say the output is wrong, and people who can ask a model a real question. Start by asking what the team already uses. Shadow AI is a map. One shop found departments buying their own ChatGPT seats. That was cheaper intel than a consultant: demand, and a bill to consolidate.

You do not need a data scientist on every project. You need one in the first few, to check whether the model is lying and which tool fits. Developers still own security and scale. Someone has to translate the business question into a prompt the model can fail honestly on.

Train the curious. Policy first, then the 30 minutes.

Hiring experts is expensive. Give the curious people time and a two-page rule: what they can use, what they cannot paste, what counts as sensitive. One company gave people 30 minutes a day. In six months those people were the internal bench. Conferences are not that.

The skill that matters is context: who you are, what you need, why, an example, a limit. First output is a draft. A marketing team cut article time by breaking the ask into outline, research, draft, edit. Hire people who can split a messy problem into those steps. Fast learners beat a resume that says AI.

Write down what worked

Keep the prompts and the misses in one place. That file is the playbook. The shops that pull ahead are not the ones with the biggest budget. They are the ones who already have a team while everyone else waits for a perfect hire.