Human judgement & leadership
Video · 5 min read
The short answer
Watch: As AI becomes more capable, where do humans and leaders still have a role?
Watch the explanation, then explore the practical implications below.
Human involvement should have a clear purpose
As more work is delegated, it becomes important to define where leadership attention is needed. Reviewing every action can create a bottleneck. Leaving the boundary undefined can allow consequential choices to pass without meaningful oversight.
Identify the decisions that require authority or contextual judgement, and the routine activities that can proceed within agreed limits. The appropriate boundary depends on the task, the evidence available and the consequences of an error.
Make supervision more deliberate
Define the signals a reviewer should see: an unresolved contradiction, a commitment outside an agreed limit, missing approval or an exception the workflow cannot handle. State who receives the issue and what information they need to decide.
Also review a sample of work that did not trigger an exception. A quiet queue does not prove that everything is working. It may indicate that important problems are not being detected. Supervision needs a way to test the rules themselves as well as resolve individual cases.
Better information access is not complete understanding
A connected system may help bring together records that previously required several reports. The value depends on whether those records are current, comparable and appropriate for the question. Missing context can remain missing even when the answer is easy to read.
Ask where the figures came from, which period they cover and what is absent. Preserve the opportunity for the people closest to the work to explain developments that have not reached the records. Treat the output as evidence to interpret, not a complete account of responsibility.
An example: a project that appears to be losing money
Suppose an AI-assisted summary highlights falling project margin. That should prompt a focused investigation. Check whether costs and revenue use the same reporting period, whether approved changes have been recorded and whether the forecast includes commitments that have not yet been invoiced.
Then speak with the project and commercial owners. The problem may require a delivery decision, a correction to the records or both. The summary alone does not establish the cause or who is at fault. Leadership adds value by bringing the right people together and resolving the consequential choice.
Use visibility to support people rather than bypass them
Direct access to operational information can tempt a leader to intervene in more details. If every new insight creates another instruction outside the agreed process, local teams may lose clarity about who decides and which priorities take precedence.
Agree how leaders will raise questions and how decisions return to the workflow. Make the intervention useful: remove a constraint, clarify a trade-off or authorise an exception. Keep the person responsible for delivery involved instead of creating a parallel chain of direction.
How CSM would approach leadership oversight
We would map decisions and information needs with the business owner and the people doing the work. That includes what the system can prepare, which uncertainty it must surface and what remains with an authorised person.
We would test the workflow with incomplete and conflicting information, then agree review, monitoring and support responsibilities. Measure whether decisions become better informed and obstacles are resolved sooner, alongside review effort and avoidable escalation. The objective is leadership attention that improves outcomes while enabling others to act within clear boundaries.
Questions for your leadership team
Where does leadership judgement materially affect the result?
Use real cases to identify the decisions that need contextual knowledge or authority.
Which signals should require attention?
Define specific exceptions and the evidence needed to assess them.
What could be missing from the information?
Check source quality, timing and the knowledge held by people close to the work.
Are we enabling the team or adding another approval queue?
Review how interventions affect ownership, speed and clarity.
How will we detect problems that are not escalated?
Sample routine work and test whether the exception rules identify consequential issues.
Full video transcript
The full video transcript is included below for reference.
One thing is that being involved can’t always mean approving every action. If you have systems carrying out several pieces of work at once, you can’t personally inspect every step and still get much benefit from giving them the work.
So you have to think more carefully about what you need to see. What decisions should come to you? What uncertainty should be surfaced? What would tell you that the work is moving in the wrong direction?
We already do versions of this with people. You don’t ask every team member to get approval for every email. You agree what they’re responsible for and what needs to be escalated. The difficulty with AI is making those boundaries explicit enough that they actually work.
For leaders, there’s also a real opportunity to get closer to the information behind a decision. You may be able to ask questions across operational records, customer feedback and project information without waiting for several layers of reporting.
But I’d be careful about confusing that access with complete understanding. The information still has gaps. Some of it may be out of date. The person running the project may know something important that hasn’t been recorded anywhere.
So if the AI shows you that a project appears to be losing money, that gives you a better question to ask. It doesn’t automatically give you the full explanation or tell you who is responsible.
That’s where leadership still matters. Deciding what deserves attention, bringing the right people into the conversation and making a judgement with the consequences in mind.
I think the risk is that leaders use better access to information to intervene in more details, and accidentally make everyone less able to act. The opportunity is to understand the business better so you can give people clearer direction and remove problems they couldn’t solve on their own.
Put these questions to work in your business
Bring a real decision or workflow to a conversation with CSM. We can explore the evidence, responsibilities and practical steps needed to make AI useful for your team.