Most reviews spend the hour on outputs nobody can act on. Twenty minutes, input numbers only, and one dated decision the tool will not let you skip.
A meeting that reports, and decides nothing. The numbers are read out, everybody nods, and the week that follows looks exactly like the one before it. At a thirty-person software company the decision from last week is not on the agenda, because nobody wrote it down.
By hand today
“You are not allowed to discuss output metrics during the WBR, except in a reporting sense”
Cedric Chin, on the Amazon weekly business review, Commoncog, 2 July 2024
That rule is the whole difference between a review and a report. Revenue, signups and pipeline value are outputs. They are yesterday, and nobody in the room can act on them. The things a room can act on are the inputs: how many companies were added, how many messages went out, how many replies were answered the same day.
Most weekly reviews do the opposite. They spend the hour on the outputs, because the outputs are what the founder is anxious about, and they finish without naming a single thing that will be done differently. A fortnight later the same slide appears with a worse number on it.
Reading the numbers is a separate skill and the outbound book covers it in full, including how to find the stage that leaked. This chapter assumes the numbers are already on the screen from the previous chapter. It is about the twenty minutes after that, and whether anything changes.
What it costs you each week
Twenty minutes, the length of the weekly ritual on the ArkOne site: read the numbers, write the decision, stop. Anything longer has become a meeting about the numbers.
Source: the ArkOne SEO runbook, line 27, read 8 September 2026
That ritual is written down as roughly twenty minutes in docs/seo-runbook.md line 27, and the length is the discipline. A short slot forces the room to skip the commentary and reach the decision, because there is no time to argue again with a number that is already on the screen.
Compare that with the hour most companies give it, and the comparison is not really about time. A long review is a symptom: the numbers were not trusted, so the meeting became an argument about measurement instead of a decision about work. If the three chapters before this one hold, that argument has nowhere to start.
The build
Codex holds one decision, written against a number. The review opens on the decision from last week, and asks whether the number it aimed at moved. It closes with one new decision, dated, with the number beside it.
Work in the folder that holds the weekly screen and your CRM. Type:
Every Monday, open the review. First show the decision I wrote last
week and the input number it aimed at, this week beside last week, and
say plainly whether it moved. Then show only the input numbers from the
screen, never the revenue. Ask me for one decision, and store what I
write with the date, the number it aims at and the stage it belongs to.
Turn the decision into work: name the chapter it reopens and put the
first task on the lead or the list it touches. Refuse to close the
review with no decision written.
What comes back is a page in three parts. Last week, and whether the change worked. This week, in input numbers only. And an empty line waiting for one decision, which the tool will not let you skip.
The last part is what makes the loop close. A decision that stays in a document is a wish. A decision that lands as work in the CRM, against a lead or a list, is the reason next week has a starting point that nobody had to invent on Monday morning.
Write it verbatim, and write it dated. On the ArkOne platform, every weekly check is recorded as a dated note in the words that were used, at docs/seo-runbook.md line 35; every lesson from the ad work is one dated entry carrying the evidence that taught it, at plans/linkedin-ads-build-day/LEARNINGS.md line 10; and decisions are appended to a log at the end of every working session, never rewritten in place, at CLAUDE.md line 45 and docs/state.md. Copy that shape and build your own. There is no demo to download.
One rule about the shape of the decision. It has to be small enough to be finished before the next review and specific enough to be judged. Send the second message a day earlier is a decision. Improve the messaging is not.
What it feeds
Next week, already written. The decision names the stage to reopen, so Monday starts on work rather than on a discussion. This is the last chapter, and what it feeds is the first: a decision about who you sell to sends you back to chapter one, and a decision about replies sends you back to chapter seven. The list for next week is written by the number that moved, not by whoever spoke loudest on Monday.
Three numbers prove this stage. Weeks with at least one decision logged against a number, which should be every week. The input number that last week aimed at, moved or not. And the minutes the review took, which should stay under thirty.
What stays with a person
The decision, and what to stop. A screen can name the stage that fell. Only a person chooses what to do about it.
Everything else in this book can be built. This cannot, and it is the reason the rest exists. The engine narrows twelve stages down to two or three things worth arguing about, and then a person decides.
The decision nobody makes is the decision to stop. A channel that has produced nothing for two months, a sequence nobody answers, a report that no longer changes anything: each one carries on because stopping it feels like admitting the last three months. Put one thing to stop in the review every month, and let the number decide it rather than the sunk cost.
The tools you already pay for
Nothing new to buy.
The screen from the previous chapter supplies the numbers.
A text file, decisions.md, holds one dated entry a week.
Your calendar gives it one twenty minute slot, weekly.
CRM is where the decision lands as real work.
No meeting product, no deck, and no analyst to prepare it.
One number, one decision, written down. Then next week starts itself.
Ask: What did you decide in your last weekly review, and did the number move?
This is chapter 12 of twelve, after the whole engine on one screen; the engine then runs again from the one sentence that decides who you sell to. The book, Building a real GTM engine with Claude Code and Codex, end to end, runs the whole engine in order, one chapter per stage. A free starter edition for firms under fifty people is coming through the 100k programme.