The principal of a small engineering firm writes every proposal. He is genuinely better at it than anyone else there, the win rate says so, and each one takes him four or five hours.
He is right about the quality gap. He is running the wrong calculation.
The comparison people actually make
The instinctive comparison is between two versions of one document: the proposal he would write and the proposal his senior engineer would write. On that comparison he wins, and the decision to keep doing it himself follows automatically.
The comparison that governs is different. It is between two futures. In one, he writes every proposal for the next six years at four hours each, and nobody else ever learns. In the other, the next eight proposals are somewhat worse, after which he has a person who writes them at ninety percent of his standard and he has recovered four hours a week permanently.
The second future is better by an enormous margin, and it becomes available only by accepting eight worse proposals. The cost is real and concentrated at the front, which is why so few people pay it.
Two variables decide it
How often the task recurs. Anything happening weekly or monthly should almost always be handed over, because the arithmetic compounds. Anything genuinely one-off carries no future stream and the development argument does not apply to it.
Whether the gap will close. This depends on whether the task offers usable feedback. Proposals do: they are won or lost, and the person finds out. Work whose quality is never assessed by anything external will not improve regardless of how many repetitions somebody gets, and delegating it produces a permanent quality reduction rather than a temporary one.
Where the task recurs and the feedback is real, hand it over. Where it is a one-time irreversible matter, do it yourself and do not dress the decision up as protecting the client.
The gap is smaller than you think
Leaders systematically overestimate it, and the error has a consistent structure.
You are comparing your current output, after two hundred repetitions, against their first attempt. The honest comparison is against their third or fourth, and the honest reference point for yours is what you produced at that stage, which you have almost certainly forgotten and which was worse than you remember.
There is also a quality you cannot see. Your version carries assumptions you no longer notice, shortcuts that work because you know when they fail, and habits that were correct four years ago. Someone doing it for the first time occasionally produces something better precisely because they lack your accumulated adjustments.
What doing it yourself actually costs
The hours are the visible cost and the smaller one.
Every time you take back the work you are best at, you make a claim about the organization: that this cannot be done without you. The claim is then true, because you have prevented the evidence that would contradict it. Firms built this way are worth less, are harder to sell, cannot survive an illness, and will not promote anyone into the work because nobody has ever done it.
There is also what it does to the person. A senior engineer who has been at the firm nine years and has never written a proposal has been told something about his prospects that no conversation about development will override.
Closing the gap deliberately
Handing the work over is not the whole intervention, and delegating without any mechanism for improvement is how the eight worse proposals become forty.
Give feedback on the output rather than rewriting it, which is the distinction between teaching and quietly resuming the task. Be specific about what would have made it stronger, in terms the person can apply next time rather than corrections you have already made for them. And let some of the imperfect ones go out, because the ones you intercept teach nothing and the client will usually not notice the difference you are agonising over.
Set the threshold honestly before you start. Some work genuinely requires your standard and should be identified in advance rather than reclaimed halfway through on grounds you invent at the time. Most work does not, and the test is whether the shortfall would change any outcome that matters, rather than whether you can see it.
Edited by Patrick J. Wolf, PhD