A short case for people who approve budgets
The twenty-dollar
chief of staff
You are not short of information. You are short of unbroken hours—and the free plan runs out at exactly the moment the work stops being simple.
Run the hour ledger$20Pro, per month, or $200 a year
$100 / $200Max 5x and Max 20x, per month
5 hoursthe session window before limits reset
1M tokenscontext on the newest models, paid plans
01 · THE ACTUAL BOTTLENECK
Nobody at your level is blocked by a lack of answers.
You are blocked by fragmentation. Eleven-minute gaps between meetings. A board pack that lands the night before. A model somebody else built, an assumption nobody has questioned in four quarters, and a decision that is yours whether or not you had time to read the appendix.
The free plan is a real product and it handles the first two questions of any task well. It tends to stop somewhere around the third hour—the point where a task turns into work: the long document, the fourth revision of a narrative, the model you are still interrogating at 23:00.
A paid seat does not buy a cleverer Claude. It buys the same Claude, available for the whole length of the actual task.
02 · THE HOUR LEDGER
Price it the way you would price a hire.
Set what one hour of your attention costs the organisation, then how many hours a week you would realistically hand over. The grid is one week: 24 columns, 7 rows, 168 hours.
4 hours handed over164 hours untouched
Hours back a year184
That time, valued$22,080
Pro breaks even at10 min/mo
Max 5x breaks even at50 min/mo
Assumes 46 working weeks and list prices of $20 and $100 a month. Break-even is the amount of time the subscription has to save you per month to pay for itself. The lit squares are the whole argument: you are buying back a sliver of a week, and the sliver is cheap.
03 · WHAT A PAID SEAT ACTUALLY CHANGES
Four differences you notice in the first fortnight.
It stops stopping
Pro gives at least five times the free plan’s usage per session, plus priority access when traffic is high. Sessions reset every five hours and a weekly limit applies on top. The change is not raw volume—it is that the limit stops arriving in the middle of things.
Anthropic Help CentreIt stops starting from zero
Paid plans get memory and project summaries, so context builds instead of being retyped. Put the strategy, the org chart and the last three board packs into a project once; documents there are cached, so reusing them costs you less of your limit each time.
Usage limit best practicesIt works inside the workbook
Claude for Excel is available on Pro, Max, Team and Enterprise. Ask questions about the workbook that is open in front of you, get answers with cell-level citations, change an assumption while formula relationships hold, and trace an error to its root cause.
Use Claude for ExcelIt holds the whole pack
On paid plans the newest models support a context window of up to one million tokens; others run at 500K or 200K. In practice: the ninety-page pack, the appendix and last quarter’s version can sit in one conversation, so the comparison happens in the model rather than in your memory.
Usage and length limits04 · WHICH SEAT
Four honest answers. One of them is “stay free.”
Capacity is shown relative, not absolute—Anthropic publishes multiples, not message counts. Pick the row that describes your week rather than the one that describes your ambition.
CapacityAt least 5x the free plan per session
This is you ifClaude is part of your working week—reading, drafting, modelling, deciding—several days out of five.
The tell that you have outgrown itYou hit a limit in the middle of something that mattered, or you have started splitting one task across two days.
VerdictThe default for a senior individual. Twenty dollars a month, or two hundred a year, and it also covers Claude Code and Cowork.
Buying four personal seats for a team is the expensive mistake. Shared context, admin control, central billing and identity management live in Team and Enterprise plans, which is also where your data policy conversation belongs.
05 · MONDAY, NOT THEORY
Four hours from a real week.
The inputNinety pages at 22:40, for a 09:00 meeting where your only job is to ask three questions worth the room’s time.
What the seat changesThe whole pack fits in one conversation: a structured read, the numbers that moved, and the places where the narrative and the appendix quietly disagree.
What stays yoursDeciding which disagreement matters, and being willing to raise it in front of the person who wrote it.
You are not outsourcing the judgement. You are arriving with it already formed.06 · WHERE THIS ARGUMENT STOPS
Four reasons not to upgrade.
- 01You use it twice a month.Then the free plan is the right answer and the upgrade is a tax on good intentions. Wait until it interrupts you.
- 02It can be confidently wrong.Fluent output is not verified output. The higher the stakes, the more the number gets checked outside the chat—ask for the cells, the source and the working.
- 03Confidentiality is a policy question.Board material, personal data and client files are governed by your organisation’s rules, not by a subscription tier. If Claude is going to hold real company context, involve whoever owns that policy and buy Team or Enterprise.
- 04The limits are still real.Sessions reset on a five-hour window, weekly caps apply, and usage across the app, Claude Code and Desktop draws on the same pool. Max moves the ceiling; it does not remove it.
THE DECISION RULE
Two questions, thirty seconds.
Twenty dollars was never the question. The question is whether the next ninety-page pack gets read by you at midnight, or with you at nine.