What an admissions season lets you change, and what it has already decided for you

Every piece on this subject tells you to prepare for the rush, and anyone who has run one season already knows the rush is coming. The harder question is which of the things you would like to fix are still fixable in week three, and most coaching advice quietly assumes you have weeks to work with.

A note on evidence before we start. The public numbers in this field describe university admissions offices, not the sales teams at education companies. Nobody publishes the second kind. The structure holds across both, and where it might not, this piece says so.

The intake calendar is published a year out, and almost nothing else in sales is

UCAS, which runs the admissions system it reports on and is therefore counting rather than sampling, published that 600,660 people had applied by its January 2025 deadline, and states that this deadline normally accounts for approximately 80% of applicants in any given cycle. That 80% is the operator's description of a typical cycle rather than a measurement of one year, and the calendar is British. By results day on 13 August 2026, 262,820 UK eighteen-year-olds had been accepted, and 446,690 applicants of all ages and all domiciles in total, both of them running totals as at that date rather than counts of what happened on it.

What follows is narrow. If the date is published a year out, then every staffing, scripting and rehearsal decision for that fortnight can be made in a quiet month, and the only thing the season itself has to supply is the calls. The test is cheap: ask your team today which two weeks of next year's calendar will carry the volume. If nobody can name them without opening a file, the planning has not happened yet. That gap is the whole argument behind working on admissions conversations before the season rather than during it. Which two weeks of last season actually carried the calls?

A weekly loop is longer than the window it is meant to fix

The loop as usually described takes three weeks to close, by its own arithmetic. You see a pattern in this week's calls, you raise it in next week's one-on-one, and you check whether it moved the week after. That cadence works, and the weekly plan built from call data is the right default for eleven months of the year. It is also three weeks long, and a results day lasts one.

So inside the season the unit of work changes. It becomes one sentence the team has no agreed answer to, settled on the day you hear it and checked against the next hour of calls. One-on-ones move to the other side of the deadline, where there is time for them to work.

Which row disappears first when the week gets loud? The last one, and it is the only row with a payoff after the deadline has passed, because next year's team will not have met this year's. The same question of what a conversation has to establish out loud shows up after a property viewing, where the follow-up is short and the window is hours.

Your bench was already new before you hired for the season

Hiring for the season is the first thing anybody blames, and the numbers say the bench was already new in May. A 2023 research brief on the higher education admissions workforce, run and paid for by the association representing higher education HR offices, covered 12,042 admissions employees at 940 US institutions as of November 2021 and found that counselors and coordinators had a median of two years in post, with 71% at three years or less and a median age of 30. The equivalent share in 2017 was 74%, so this is the shape of the role rather than a recent shock. Years in position there means time in that post at that institution, which the report states in its own footnote, and the data is around five years old.

The practical consequence lands on memory rather than on training. If the only place last cycle's working phrasing lives is in the heads of the people who ran it, and those people turn over on a two-year median, every season genuinely does start from nothing.

The weekly plan makes a related point about a teammate's recording settling an argument this week. The seasonal version is colder. The teammate will not be here next cycle, so the recording has to survive a handover to somebody who never met them, which means the sentence has to be marked at the time rather than searched for a year later. Practicing against those recordings is a separate question from keeping them, and keeping them is the part that cannot wait for a decision about tooling. Last season's wording is still in last season's calls

The triage rule gets written either in February or by the queue in August

Capacity is the constraint that makes triage unavoidable. A 2019 report on the state of college admission, run by the professional association for admissions officers, put the average at 1,035 applications per admission officer at public institutions and 461 at private ones, from a survey of 67 and 133 institutions respectively across 2018 and 2019 in the United States. At that ratio, a personal conversation with everyone is arithmetic that does not work.

The report carries an internal inconsistency worth naming: its prose refers to one admission cycle and its table heading to another. That, plus its age, is why the ratio belongs in an argument as an order of magnitude. It is enough to make the point, which is about who decides.

Written before the season, the rule says these inquiries get a call because only a conversation settles them, and the rest get a good message with a way to ask for one. Left unwritten, the queue decides instead, sorting by whoever rang most recently and whoever complained loudest. Neither of those correlates with who was about to enroll.

We trusted the model's own score, and then came a deadline week

The roleplay simulator used to take its final score from the model that ran the conversation. In August a client running accounting courses hit the case that broke it: two parts scored five out of ten at equal weight, which is fifty by the formula, and the field came back with twenty-four. The learner saw twenty-four, and the pass decision rode on that number. Now the score is computed from its parts rather than taken on the model's word.

The seasonal version of that lesson is about who is left to notice. The person who caught the formula was not reviewing scores at the time; they were doing something else and happened to add two numbers in their head. In the week of a deadline nobody is doing something else, and a number that looks official is simply used. Whatever you configure before a season runs through the season unattended, which is an argument for configuring less of it and checking the little that remains.

What to do next

Take the two weeks around last season's deadline and pull the calls. Count how many of them contain the same question from the applicant answered three different ways. That count is the size of the thing a same-day correction would have fixed, and it is usually larger than anyone expects.

Then write the triage rule now, while the queue is short and nobody is under pressure. One page: which inquiries get a call, which get a message, and who decides when the two lists disagree. Put a date on it and the name of the person who signs off a deviation during the season, because that is the part the queue overrides first.

Last season is already recorded. Send us the calls from the two weeks either side of your deadline and we will mark where the same question got three different answers, and which of those calls needed a voice rather than a message. Two weeks either side of one deadline.