The single biggest spend of the year lands on one predictable peak, and the evidence for a cut-or-scale call needs to be back with runway before that peak, or the answer becomes a note for next year instead of a decision for this one. The window is known every year, and the honest read on the account's own history is that the test has started too late more than once, leaving the same open question for another twelve months. Every extra week of internal deliberation before a design gets committed subtracts from runway already known to be tight.
The design gets timed backward from the peak, not forward from whenever sign-off happens, so the read is back with room to act on it instead of room to regret it. The year's largest commitment gets made on an answer instead of a standing suspicion. And the exact question reopened every year at the same point on the calendar finally closes.
A recommendation keeps pointing toward increasing a campaign because its recent numbers look strong, right up until an external event, a funding decision, an exam-window change, a program cancellation, cuts the demand behind it overnight. The model cannot see that kind of shock coming, so it keeps recommending more of what just stopped working, and the only option left is to override it on judgment alone, in a company small enough that a single name sits behind every call like this.
Following the recommendation while the shock is in motion makes the loss read as a deliberate choice; overriding it and turning out wrong lands the blame the same way. The calendar shock gets built into the read itself, a known constraint rather than a surprise the model has to be forgiven for missing. A recommendation that redistributes budget toward what is still alive replaces one pointing at what just died, and a record separates a bad outcome the calendar caused from a bad call actually made.
It means setting the schedule of a cut-or-scale read backward from a predictable demand peak or exam-driven event, rather than forward from whenever a report happens to be finished. The goal is for the answer to exist while there is still budget left to act on it, not as a note for the following year.
By starting the read early enough that it can finish with runway before the lock, sized to the time actually available rather than a standard test length. A read that finishes after the peak has already been spent answers a question the business no longer has time to act on.
By treating known calendar risks, such as a funding decision or an exam-window change, as a constraint built into the read rather than an exception handled after the fact. When the shock lands, budget still pointed at the campaign that just lost its demand can be redistributed toward what is still alive.
When there is no history of a prior peak or shock to calibrate against, when the decision needed is a same-day pause rather than a budget commitment, or when the calendar event itself needs to be predicted rather than read once it happens. In those cases the honest answer is a wider range, not a confident call.
The cut-or-scale decision stops arriving after the peak it was meant to inform, and a calendar shock stops looking like a personal misjudgment after the fact. Both get separated from ordinary reporting delay and tied instead to the actual date the money has to move.