What the weekly growth meeting is for
The experimentation cadence guide sets out the rituals of a growth programme, from the weekly read-out and prioritisation to the monthly economics review and the quarterly model review, and notes that small teams merge several of them into one weekly hour. This tutorial runs that merged hour. It does not restate the cadence guide's rituals or rules; it turns them into an agenda, a pre-read and a sequence of moves, and each step links the page that owns the rule it applies.
The meeting has three outputs and only three: decisions on finished tests, the next tests started with owners and capacity confirmed, and entries in the learning logA single, searchable record of every experiment, losers and inconclusive results included, with the hypothesis, design, result, decision and what it taught.. Anything else, including a status update that could have been read, a debate about a live test's early numbers or a tour of every chart in the dashboard, takes time from those outputs. The growth marketing pillar describes the weekly meeting in the context of the growth team's roles and its monthly and quarterly rhythm, and the growth fundamentals hub explains the model the meeting serves: a north star, an input tree beneath it and loops whose leaks the experiments target. The academy capstone on a growth operating system for a vintage clothing resale app practises designing the weekly meeting that turns a year's model, tests and two-sided readings into decisions.
Two neighbouring meetings are worth separating from this one. A dedicated testing programme, as the experimentation programme playbook runs it, keeps a weekly stand-up for tests in build and quality assurance and a separate read-out for finished tests; a team large enough to run that programme can keep its meetings apart and use this tutorial for the growth review that sits above them. And a business-to-business team that needs marketing and sales in the same room about stuck deals runs the weekly pipeline review, whose subject is opportunities rather than experiments.
Before you start
- A north star and its input tree, written down and agreed. The north star and OKRs framework covers choosing them; the meeting reads them, it does not debate them.
- A ranked backlog of experiments, each written with a hypothesis, a primary metric and a decision rule. The experiment hypothesis template gives each entry that shape, and the guide to prioritising tests covers the ranking.
- A learning log, in one place, in the format the cadence guide describes, with read-outs drawn up on the experiment results report template.
- A growth accounting bridge or similar weekly view of how the active base moved: the growth accounting tutorial builds one by period, and its movements are the clearest way to show where a change in the north star came from.
- Someone who is not the test's builder to call each result, as the cadence guide requires.
The steps
Name the owner, the attendees and a fixed hour
One person owns the meeting, usually the growth lead. The owner sends the pre-read, runs the agenda, keeps live results out of the room and makes sure decisions are written down. Attendees are the people needed to make the meeting's three outputs: the analyst who reads results, the owners of tests that are finishing or starting, and the product and engineering partners whose capacity the next tests need. Everyone else reads the notes. Fix the slot in the calendar and hold it whether or not the week feels eventful, because a meeting that happens only when something is exciting reverts to the heroics the cadence guide warns against.
Set a standing agenda
Use the same agenda every week, in the same order, so that nobody has to ask what the meeting is for. The table below gives the shape: the north star and inputs first, then finished tests, then live tests, then the backlog and capacity, then decisions and the log. Put the time for each item in your own agenda from your own experience of how long the team needs, and protect the last item: decisions and log entries are written in the meeting, not after it. A standing agenda also makes absences survivable, because a deputy can run it without preparation.
Build the pre-read from the north star, its inputs and the live tests
The pre-read is short and fixed in format. It shows the north star and each input beneath it for the week, with the prior weeks for comparison and a note of any known event that touched them. It shows the active base's movement split into new, retained, resurrected and churned customers where the team keeps a bridge. It lists the tests that finished, each with its read-out already drafted against its decision rule. And it lists the live tests with their progress towards the planned sample and their health checks (traffic split, tracking, errors) but not their primary metric. Leaving the live results out of the pre-read is what makes the later steps possible.
Send it the day before
Send the pre-read early enough to be read, and expect it to have been read. Questions of fact go to its author before the meeting, so the hour is spent on decisions rather than on finding out what a chart shows. A pre-read that arrives as the meeting starts turns the meeting back into a presentation. If the read-out for a finished test is not ready in time, the test moves to next week's agenda rather than being read out live from raw numbers.
Open on the north star and its inputs
Start with the business, not the tests. Read the north star, then the inputs, and ask which input moved and whether the movement is real. Weekly numbers are small, and the guide to reading small numbers explains why a movement of a few customers is usually noise; read the trend over several weeks and compare like with like. When an input has clearly moved, the guide to explaining a change in a marketing metric covers separating a real change from a mix shift, and the growth accounting bridge shows which movement in the active base changed. Opening here keeps the experiments honest: a team that sees its activation rateThe share of new users or sign-ups who complete the activation event, the action marking first real value, within a set window. falling before it hears about a winning test on a minor page asks the right question about priorities.
Call finished tests against their decision rules
For each test that reached its planned sample and end date, the analyst presents the drafted read-out: the primary metric, the guardrail metricsA metric an experiment must not worsen beyond an agreed bound, watched beside the primary metric that decides the test., and the call against the decision rule written before launch. The call is win, loss or inconclusive, and an inconclusive result is recorded as inconclusive, not argued into a narrow win. The test pitfalls guide lists what to check before trusting a result, from a sample split that differs from the plan to segments found after the fact. A win that will ship gets an owner and a date, and the tutorial on shipping a winning variant covers the work between the call and the change being live for everyone. Read wins from tests on new customers by CohortA group of customers who share a starting event and period, tracked together over time. later, because the cadence guide revisits shipped wins against cohort data a quarter on.
Check live tests for problems, not results
Go through each live A/B TestControlled comparison of two variants. and ask only whether it is healthy: is traffic splitting as planned, is tracking firing in both arms, has a variant broken on any device, has a promotion or a release disturbed it? Problems like these can justify stopping and restarting a test. An encouraging or worrying early result cannot. Evan Miller's essay on how not to run an A/B test, as the site's test pitfalls guide and the experimentation loop framework set it out, shows why: checking a test repeatedly and stopping when it first looks significant inflates false positives, so the sample and the rule are fixed before launch and the result is read once, when the test has finished. The owner's job in this step is to stop anyone asking how a live test is doing.
Choose what starts next from the ranked backlog
Take the next tests from the top of the ranked backlog rather than from whoever argues hardest in the room. If the opening review of the inputs has changed what matters most, re-rank in the meeting with the method the team already uses: the impact-effort matrix for a quick sort, or the scoring and traffic constraints in the guide to prioritising tests. Each test pulled must claim a path to the north star through a loop stage, as the growth loops guide describes leaks, and must have its hypothesis and decision rule written before it is scheduled. Ideas raised during the meeting go into the backlog to be written up and ranked, not straight into the plan.
Confirm capacity and owners
A test chosen without the people to build it is a wish. For each test that will start, confirm who builds it, who checks it before launch, who owns its decision and when it can launch, with the product and engineering partners in the room agreeing to the capacity. If capacity is short, choose fewer tests rather than spreading people thin, and say which test waits. Record any dependency, such as a release the test must not overlap, so the next pre-read can show whether it held.
Write decisions and log entries before the hour ends
Close by reading the decisions back: each finished test's call and what happens next, each new test with its owner and launch date, and any change to the backlog's order. Then write the learning-log entries for the tests called, in the log itself rather than in meeting notes, including the interpretation: what the team now believes that it did not before. Send the decisions to everyone who was not in the room the same day. The experimentation programme playbook describes the wider log and how decisions follow from it. A meeting whose decisions are written up a day later tends to find that people remember them differently.
A standing agenda: a table
| Agenda item | Question it answers | Who leads | What goes in the pre-read | Output |
|---|---|---|---|---|
| North star and inputs | What moved in the business this week, and is it real? | Growth lead or analyst | The north star, each input, prior weeks and known events | A note on which input moved, or that nothing did |
| Active base movement | Where did the change in active customers come from? | Analyst | New, retained, resurrected and churned for the period | An owner asked to look at any movement that changed |
| Finished tests | What did the tests that finished show? | Analyst, not the builder | Drafted read-outs against each decision rule | Each test called win, loss or inconclusive |
| Live tests | Is each running test healthy? | Test owners | Progress to planned sample; split, tracking and error checks; no primary metric | Tests left alone, or stopped and restarted for a fault |
| Backlog and next tests | What starts next, and does it still rank first? | Growth lead | The top of the ranked backlog with hypotheses and rules | Next tests chosen |
| Capacity and owners | Who builds, checks and owns each next test? | Product and engineering partners | Capacity for the coming weeks; known releases | Owners and launch dates |
| Decisions and log | What did we decide and learn? | Meeting owner | None | Decisions read back; log entries written; notes sent |
The agenda does not include a tour of the dashboard. A KPIA metric tied directly to an objective. that is not the north star or one of its inputs belongs in the monthly or quarterly review that the cadence guide describes, unless it has moved far enough to change a decision this week.
A bike servicing firm's meeting in its second month (illustrative)
Tazzimor Bikes turns a dashboard hour into decisions
Tazzimor Bikes, an invented bike servicing subscription that riders book online, used to spend its weekly growth hour walking through fourteen dashboard charts. The north star, services completed for subscribed riders each month, was mentioned near the end if at all, and decisions were rarely written down.
In the second month of the new format, the growth lead sends a one-page pre-read the afternoon before. It shows the north star at 1,180 services for the month to date against 1,150 at the same point last month, and its three inputs: 214 new subscribers in the week, first-service booking within a month at 44% for the latest cohort, and rebooking after a first service at 61%. The bridge for the month so far shows churn up by 9 subscribers, which the analyst marks as within the usual weekly swing.
Two tests have finished. The first, showing the next free workshop slot on the booking page, reached its planned sample of 1,600 new subscribers per arm: first-service booking rose from 41% to 46%, the guardrail of first services cancelled was flat, and the analyst calls it a win against the rule. The operations lead owns the change and its release date. The second, a reminder before the second service was due, reached its sample with rebooking at 60% against 61%, and is called inconclusive; the idea returns to the backlog to be rewritten with a bolder change.
A third test, an annual-plan prompt on the booking confirmation page, is at 58% of its planned sample. The pre-read shows its traffic split at 50.4% to 49.6% and tracking firing in both arms, and nobody asks how it is doing.
The next test on the ranked backlog, photos of the finished work in the service report, is confirmed with the workshop app developer's capacity from the following Monday, and the product lead owns it.
The meeting ends with three decisions (ship the slot display, return the reminder to the backlog, start the photo test), two new owners and four log entries: the two calls, the live test's health note, and a note that the north star's rise came from first services rather than rebooking.
Common mistakes
- A tour of the dashboard. An hour spent on every chart leaves no time for decisions; the pre-read carries the numbers and the meeting carries the choices.
- Opening on the tests. Starting with the north star and its inputs keeps the meeting on the business, not on whichever test has the best story.
- Reading live results. A weekly look at an unfinished test, followed by stopping when it first looks good, is the peeking the test pitfalls guide warns about.
- The builder calls the result. The person who wanted the variant to win is the wrong person to call it.
- Starting tests without capacity. A test with no builder confirmed slips silently and the backlog loses its meaning.
- Ideas from the room go straight into the plan. New ideas join the backlog and are ranked like every other entry.
- Decisions written up later. Write them and the log entries before the meeting ends, and send them the same day.
- Cancelling quiet weeks. A week with nothing to call still needs the inputs read, the live tests checked and the next tests resourced.
