Programs measure the wrong thing about their timeline. They ask whether decisions went out by the date on the website. The question that matters is what happened to the applicant pool while they waited.
Strong applicants apply to several things at once. They are strong precisely because other people also want them, which means they are always holding more than one open option. When your decision takes six weeks and someone else takes two, you are not competing on program quality. You are competing on whether the applicant is still available. The pool you decide from is not the pool that applied.
This produces a failure that never shows up in any report, because the applicants you lose look identical to the applicants who declined for other reasons. You see acceptances that do not convert and you conclude the offer was not compelling. Often the offer was fine and simply arrived after the decision had already been made elsewhere.
The delay is rarely caused by the deciding. It is caused by the queueing. Applications sit until someone has a clear block of hours. Reviewers wait on each other. A round of chasing happens because two reviewers interpreted a criterion differently and their scores cannot be reconciled. The actual judgment, when it finally happens, takes minutes per applicant. The waiting takes weeks.
So attack the queue, not the judgment. Three things move the number most.
First, let the mechanical work happen as applications arrive rather than after the deadline. Eligibility checks, completeness, extracting the passages that answer each criterion, reading the attached document or linked project. None of this needs a human and none of it needs the round to be closed. By deadline day, a reviewer should be opening a prepared file, not a raw form response.
Second, fix the criteria before the first application, and calibrate on five applications together before anyone reviews alone. Almost every mid-round stall traces back to reviewers who discovered on day nine that they were scoring different things. That discovery is cheap on day zero and expensive after two hundred applications have been scored under two different standards.
Third, decide in one sitting per stage. A shortlist assembled across three weeks is scored against a standard that drifted, which means it will be revisited, which is another week. Twenty applications in a block, a decision at the end of the block, then stop.
There is a limit worth naming. Fast is not the goal, and a process that gets fast by reading less has traded one failure for a worse one. The aim is to remove the waiting that adds nothing, so the reading and the deciding get the time they deserve. A well run round is mostly judgment with very little queue, and it feels slower to the reviewer even as it gets dramatically faster to the applicant.
Measure it directly. Time from the last submission to the first decision email, and time from deadline to the final one. Write both numbers down each cohort. They are the clearest available signal about whether your best applicants will still be there when you choose them.