Delegate by level of decision rather than by task. Decide what someone else can own outright, what they can do with your approval, and what only you can hold, then brief each one differently and review at a fixed cadence rather than continuously.
Delegation rarely fails because the wrong person was hired. It fails because the task moved and the decision did not. The assistant books the meeting but cannot decide whether it should happen, so the question comes straight back and the leader is still in the loop, now with an extra message to read.
The fix is to delegate at the level of the decision. That means being explicit about what someone can settle themselves, what they can prepare for your approval, and what genuinely needs you.
Almost every piece of work fits one of three levels. Naming the level when you hand something over removes the most common source of friction, which is a mismatch between how much authority you think you gave and how much the other person heard.
The point of the framework is movement. Work that starts in Draft should be sitting in Own within a quarter, because the pattern has been observed enough times to be predictable.
A simple promotion rule keeps it honest: once five consecutive drafts have been approved without change, the work moves up a level. If a decision is later reversed, it moves back down and the reason is written into the decision log so the pattern is captured rather than remembered.
A brief that produces good work is short and specific about four things: the outcome, the constraints, the deadline and the level. Anything longer usually means the task has not been thought through, and anything shorter guarantees a follow-up question.
The most common omission is constraints. People describe what they want and leave out what must not happen, which is where most rework comes from.
An AI-trained assistant shifts the boundary of what is worth delegating. Work that used to be too fiddly to explain, such as pulling a weekly summary from three systems, becomes worth handing over once because it will be automated shortly afterwards.
It also raises the standard for briefs. A model needs the same context a person does, so the discipline of writing the outcome and constraints down pays twice: once in the delegation, and again in the automation built on top of it.
Continuous checking is the tax people pay for an unclear framework. Once levels are set, reviews can move to a fixed cadence: a short daily check-in for the first fortnight, twice weekly through the first month, then weekly.
The weekly review is the one worth protecting. Fifteen minutes covering what moved, what slipped, what needs a decision and what should be promoted a level keeps delegation improving rather than drifting.
Every delegation framework works for about three weeks. What kills them is not disagreement with the idea but the absence of a moment in the week where the framework is actually used, so it quietly reverts to ad hoc requests in a chat thread.
Attach it to something that already happens. The weekly review is the natural home: run through what was owned, what was drafted, what came back and why, and decide whether anything should move a level. Ten minutes is enough, and the discipline of doing it out loud is what keeps the levels meaningful.
The second failure mode is silent scope creep in the other direction. As trust grows, leaders often stop stating the level altogether, which works until the first genuinely consequential task is misread as routine. Keeping the one word label on every handover costs nothing and prevents the only expensive mistake this system is designed to avoid.
The third is a decision log that nobody maintains. It does not need to be elaborate: a single shared document, one line per decision, with the reasoning attached. Its value is that it lets an assistant answer questions on your behalf accurately, which is the point at which delegation stops feeling like supervision.
What should I delegate first? Work that is high frequency and low consequence: scheduling, expenses, meeting notes and routine supplier email. It builds context quickly and mistakes are cheap to correct.
How do I stop work coming back to me? State the level when you hand something over. Most returned work is caused by the other person not knowing whether they were allowed to decide.
How long before delegation feels comfortable? Usually four to six weeks. The turning point is when a written decision log means you stop explaining the same context twice.
More insights | Alvara Partners