The across-the-board cut is the easiest to decide and the most expensive to pay for.

Ten percent everywhere sounds fair, executes fast and forces nobody to defend a position. The problem shows up six months later: the savings are in the P&L, and so is the slowness, the errors somebody used to catch, and the departure of the two or three people who quietly held together a process nobody had documented.

Capacity got cut along with cost, because the whole payroll was treated as if it were the same thing.

The distinction that changes the decision

Inside any administrative function, two kinds of work live side by side, and from the outside they look almost identical:

Work that requires judgment. Deciding whether to extend this customer credit. Noticing that an order looks off. Knowing that the invoice for that particular payer has to be assembled differently or it comes back rejected. That knowledge is rarely written down; it lives in the people who have been doing it for years.

Work that only requires repetition. Copying a figure from one system to another. Consolidating twelve files that arrive the same way every month. Retyping what another system already holds. Attach, rename, send, confirm.

The repetitive part gets automated. Judgment is preserved and supported with better information.

The across-the-board cut makes no such distinction, and that's why it destroys capacity: it removes people, not tasks. And the people who leave tend to take the judgment with them, while the repetitive task stays — spread among those who remain.

A case where cost fell without anyone leaving

In the early nineties, Colombian banking had a cost-to-serve problem: every teller transaction cost what a person costs during office hours, and a branch could only grow by renting more square meters and hiring more tellers.

What we built was a touchscreen self-service machine, running online rather than in batch, inside the bank lobby, available 24x7. The customer deposited their check, got a printed receipt, and left — no queue, no dependence on opening hours.

The numbers of that operation: more than 1,100 machines built, installed and supported, reaching close to seventy percent of Colombian banking; more than two million transactions a month; over 300 million across fifteen years. Cost per transaction fell to a third, and one sixth of the retail floor previously dedicated to serving customers was freed up.

What matters for this question is why the cost fell. Not because tellers were cut. It fell because the labor of the simple transaction was supplied by the customer, who preferred it that way — no line, any hour. The tellers stayed, handling what actually requires a person: the odd case, the complaint, the operation that doesn't fit the form, the sale.

The bank didn't reduce capacity. It reassigned it. And it extended hours and coverage on top, which is the opposite of what a cut produces.

The test worth applying

Before cutting anything, three questions are worth answering about each administrative activity:

What happens if this isn't done tomorrow? If the answer is "nothing for a couple of weeks", it probably exists out of inertia. Plenty of reports nobody reads fall in this category.

Does it require someone to decide something, or only to execute? If there's no decision, there's a candidate for automation. If there is a decision, the question becomes whether that person has the information to decide well.

Does it exist because the business needs it, or because a system demands it? This is the one that surprises. A good share of the administrative work in a mid-market company exists to feed or reconcile systems that don't talk to each other. That isn't the business's work: it's a tax levied by the architecture.

Where the real savings are

Almost never in payroll. They're in the time people spend on tasks that produce nothing: moving data, reconciling versions, rebuilding by hand what a system already holds, and chasing information to assemble an answer.

That time appears on no budget line, and so it goes unmanaged. But it is measurable: follow one complete process end to end and count how many steps create value and how many only move it.

Once that count exists, the conversation changes register. It's no longer how much to cut, but which part of the work can stop existing without anyone missing it — and which people are freed up for the work that actually requires judgment.

A question to close on: of everything your administrative team does this week, what share would have to be redone by hand if your systems talked to each other?

---