L&M Consulting GroupFind where AI is already being used.
Begin with an inventory. Ask employees which tools and built-in features they use, what tasks they use them for, what information they enter, and where they are unsure about policy or review. Include personal tools and informal workarounds, not just platforms bought by the organization.
People are more likely to give an accurate account when the exercise is framed as discovery rather than enforcement. Without that view, leaders may write rules for the wrong risks.
Separate information rules from decision authority.
Two questions are easily blurred: what information may enter a tool, and what decisions a tool may influence. An approved platform does not automatically make every input appropriate, and a permitted input does not make every output suitable for autonomous use.
Teams need plain guidance for confidential information, approved tools, human review, and escalation. Name the person accountable for the final decision.
Build with the people performing the task.
The people closest to a workflow know its exceptions, informal handoffs, and failure signals. Their participation shows whether an assisted step saves more time than it adds in review.
A small pilot should define the task, approved inputs, expected output, reviewer, success measure, and stopping condition before the tool is introduced.
Put policy into daily practice.
An AI workflow needs approved inputs, written steps, a review point, an exception path, and documentation someone else can follow. Training should use the same workflow people will use afterward.
Assign an owner and review the workflow after launch. Track where people hesitate, which exceptions recur, and whether the assisted step still saves time after the novelty wears off.
Change the guidance, training, ownership, or tool when the evidence changes.