Your AI policy has been approved. The announcement went out. Employees acknowledged it.
Now give three managers the same request: “Can my team use AI to summarize customer correspondence and prepare a response?”
Ask them to answer independently. Which information can the employee enter? Which tool can they use? Who must approve the work? Who reviews the result? What would make them stop?
If the answers depend on which manager happens to be available, leadership has work to do.
I would not call that implementation complete.
Approval is a decision. Application is a leadership responsibility.
A policy can say “protect confidential information” and “ensure human oversight.” Both sound reasonable. Neither, by itself, tells an employee what to do with the customer email sitting in front of them.
Does it contain information they must remove? Is the proposed tool approved for this task? Can they use the summary internally? Can they send the drafted response? What exactly must the reviewer check?
When those questions remain unanswered, the employee has to fill the gap. A manager may approve the task, another may prohibit it, and a third may send it to someone who has no clear responsibility for deciding.
That is an expensive place to leave judgment: at the point of action, without usable instructions.
I use AI. That is why I ask harder questions.
I use multiple AI assistants in my own businesses. My position comes from working with these tools, not watching them from a distance.
Using them has made me more deliberate about the task I assign, the information I provide and the output I am willing to rely on. An answer can be fluent and still fail to address the decision in front of me.
That distinction matters even more when a leader expects an entire team to use AI. My personal familiarity with a tool does not establish what another person understands, what they are authorized to do or how their work will be reviewed.
“I know how to use it” is not a management system.
Try the three-manager exercise.
Use a fictional scenario. No customer records are needed. Give each manager the same task, the same employee role and the same relevant policy language.
Ask for five short answers:
- Information: What may the employee provide to the tool, and what must they leave out?
- Permission: Is this tool approved for this task, and who can authorize any exception?
- Review: Who checks the output, and what must that person verify before it is used?
- Boundaries: What can the employee do independently, and what requires another decision?
- Escalation: When must work stop, and who receives the question?
This is an illustrative discussion exercise, not a validated assessment. Its value is in making assumptions visible.
Different answers are not automatically wrong. Different roles can require different authority. The question is whether the difference has a clear reason in the policy and the work, or whether each manager is inventing a rule.
Do not settle the disagreement by choosing the most confident person in the room. Identify the missing instruction or decision owner.
Then ask the employee to explain the next move.
Once the managers agree on the instructions, give the scenario to someone expected to follow them. Ask that person to explain what they would do next and why.
A signed acknowledgment tells you someone acknowledged the policy. Attendance tells you someone attended a session. Neither shows you how that person will handle this request.
Their explanation gives you something more useful: a chance to hear where the instruction is unclear before real work depends on it.
Start with one task. Write the instructions. Name the decision owner and reviewer. Test the explanation. Revise what remains ambiguous. Repeat when the task, tool or responsibilities change.
Leadership has to create the conditions for sound judgment. Publishing a document is only one part of that work.
The CEO’s question
At your next leadership meeting, ask: “Show me one AI task our people can carry out with clear permissions, review and escalation.”
If the answer is another presentation about the policy, keep asking.
You should be able to see how the policy changes a decision. Your managers should be able to explain it. Your employees should know their next move.
That is the standard I believe leaders should set.
When to consider AI policy consulting
If managers interpret the same policy differently, define the decision the advisory work should address: permitted uses, approval authority, review or escalation. Ask an advisor to distinguish a policy document from the organizational work needed to apply it. Policy development, assessment and implementation need an agreed scope; they are not included automatically in an educational briefing.
What to clarify when comparing AI compliance consulting
Start by identifying the uses, locations, responsibilities and requirements your organization needs to examine. Ask which evidence will be reviewed, what the advisor will deliver and which questions need separate legal or technical expertise. A general AI discussion cannot establish that an organization complies with every applicable requirement. For teams in Puerto Rico and Florida, describe where the work and participants are located before agreeing on scope.
Explore AI governance, risk and compliance advisory, or use our guide to hiring a business consultant in Puerto Rico or Florida to prepare the conversation.