Your People team uses AI inconsistently. What should you do first?

Separate four different problems

Access. Does the team have a suitable, approved tool? A demonstration of something colleagues cannot use at work does not resolve that barrier.

Guidance. Do people know what information they may use, where human review is needed and who can answer a boundary question? A confident user can still be unclear about the rules.

Practice. Can people brief an AI, check the answer and improve a weak result? This is where focused practice may help. Ask for a demonstration on suitable material, rather than assuming confidence means capability.

The workflow itself. Is the source material current? Is the task well defined? Are people repeatedly correcting the same inputs? Sometimes clearer guidance, a simpler process or existing software is the useful next step.

Ask about work before asking about tools

“Which AI should we use?” can become a long discussion without identifying anything worth improving. These questions give you a more concrete starting point:

• Which task takes more effort than it should?
• What goes wrong, and who has to correct it?
• What would a useful improvement look like?
• What information and human judgement does the task need?

Keep the example bounded. “Improve onboarding” is broad. “Reduce the corrections needed before we send an answer to a recurring onboarding question” gives the team something it can examine.

Read the answers without turning them into a score

If several people say they are unsure what they can put into AI, clarify the guidance before pushing for more usage. If people understand the boundaries but struggle to check an answer, a practice session may be useful.

A readiness survey can reveal these reported experiences. It cannot prove someone’s capability, establish that every colleague feels the same way or calculate savings. Written answers and small-team context may identify people, so ask participants to leave out personal and confidential details and keep the results link private.

Discuss what the answers suggest. Check that interpretation with the team. Someone who uses AI rarely may have good reasons for doing so.

Choose one next step you can check

For the onboarding example, first check that the approved source is clear and current. Agree what a useful answer must include, what it must not invent and when a person should take over.

If AI is suitable and permitted, test it with fictional questions and known answers. Compare the result with the standard you agreed. Record what failed as well as what worked. A decision to improve the source material instead can be a good outcome.

That makes training easier to scope: there is a task, a boundary and something participants should be able to demonstrate afterwards. It also gives you a better follow-up question than “Did everyone enjoy the session?”