Nasr Khan

14 AI

AI staffing check: before you cut, and if you stop

We plan a smaller team because AI now does part of the work. Can the people left keep things running, today and on the day we have to switch the AI off?

For CEOs, CFOs and operations leaders planning a team change. Takes about five minutes. Opens with an example.

The work, the team and the stop drill

1The work the team shares

Up to five workflows. "Promised" is the share of items the AI was expected to handle without a person. "Measured" is what it handles today.

2The team

After meetings, training and breaks.

People who no longer do this work every day are slower at it.

3The stop drill
Why the AI stops

The answer

Cut less

People needed, against the plan

The stop drill with the planned team

Must continueCan wait

Each workflow

The drill plan

    Take the result with you

    A PDF report with your inputs, results, charts and method. Or the same as plain text for an email.

    How this tool works

    What it does

    Team plans are often sized on the share of work an AI was promised to handle. This tool checks two things before you cut.

    First, the share it handles today. Second, what happens if you have to switch it off: because it's down, or because it's making mistakes. Work piles up while people switch to doing it by hand. The tool finds the smallest team that keeps urgent work on time and clears the backlog.

    What you enter

    • The work. Up to five workflows that share one team: weekly volume, minutes by hand, the AI share promised and measured, and whether each must continue or can wait.
    • The team. People today and after the plan, real working hours, and checking time.
    • The stop drill. How long the AI is off, how long switching takes, overtime, temporary help, and re-checks if it was making mistakes.

    What you get

    • People needed at the promised and the measured AI share.
    • The staffing floor for the stop you describe, and the backlog day by day.
    • A drill plan with owners, and a PDF.

    Try it: switch "Why the AI stops" to "It's making mistakes". Re-checking its recent work adds to the pile.

    The maths, assumptions and sources

    People needed = (items by hand × minutes + items the AI handles × checking minutes) ÷ 60 ÷ hours per person. The stop drill runs day by day over working days. Capacity = people × daily hours, plus overtime and any temporary people from the day they start. Urgent work goes first, oldest item first. Items the AI normally handles take longer by hand at the speed you enter, and wait until the manual process is running. An item is late when it's still waiting after its deadline. The floor is the smallest whole team that keeps every must-continue item on time and clears the backlog within your limit.

    This is a capacity plan, not a service guarantee. Treat a fallback as untested until you've rehearsed it.

    • AWS Well-Architected, Agentic AI Lens, AGENTREL02-BP05: tier human oversight, and set timeouts and fallbacks so work doesn't stall when reviewers are unavailable. The lens also asks teams to keep the skills to do the work themselves.
    See answer