पाठशाला Pathshala · संचालन Sanchālan, Operations · Lesson 01 · Start

The weekly metrics review: one page, one hour

Eight numbers on a single page, read in silence, explained by their owners, acted on before the hour ends. The meeting most small companies replace with a status call, and the one that actually runs the business.

Pathshala, The Founder Library · 11 October 2026 · 8 min read

Most ten-person companies have a Monday meeting. It goes round the table, everyone says what they did, nobody looks at a number and it ends when the hour does. The meeting the company needs is the opposite of that: eight numbers on one page, read before anyone speaks, and an hour spent on why they moved. The first kind feels like management. The second kind is.

This lesson sets up the page and the hour. Which eight numbers, how to lay them out so a change is visible without a chart, how the sixty minutes are spent, and the handful of mistakes that turn a metrics review back into a status meeting within a quarter.

Why one page, and why eight

Sachin Rekhi, who ran product at LinkedIn and SurveyMonkey, wrote the most useful short guide to metrics reviews and his first rule is about size. A dashboard needs enough metrics to see the business and few enough that it is not noise; he calls it a goldilocks principle. Detailed breakdowns live on separate dashboards. Low-volume metrics that bounce around are left off. What remains is the set the team is actively trying to move, the set that tends to move on its own and must be watched, and a temporary metric or two for the current crisis.

Eight is not a magic number. It is roughly what fits on one printed page with a trend beside each figure, and roughly what eight people can hold in their heads from one Monday to the next. Fewer than six and the page cannot tell revenue from the thing that drives revenue. More than ten and the meeting reads the page instead of discussing it. Amazon’s version of this meeting, the Weekly Business Review, is described in Working Backwards as the single most important tactical meeting the company conducts, and the point its authors keep returning to is that output metrics are not enough: the review is built on the controllable inputs that drive them.

Choosing the eight

Build the page in two rows. Row one is value and its causes. The first tile is the north star, the one number that captures the value the company delivered this week: active paid seats, repeat orders, completed transactions, accepted deliverables. The next three are its inputs, the three or four factors that most directly move it and that the team can act on in a week, which [the next lesson](/library/north-star-metric-and-its-input-tree) builds into a tree. A SaaS company might carry paying accounts, seats per account and the share of seats active this week; multiply the three and you get the north star, which is the test that the inputs are the right ones.

Row two keeps the company honest. Two money numbers: revenue or MRR, and either cash in the bank or [runway](/library/runway-how-many-months-you-really-have) at current net burn. One cost-of-growth number, such as blended CAC or contribution margin. One quality number that gets worse before anything else does: churn, return rate, disputed orders, rework. And one people number, because a company of ten has no HR function and the first sign of trouble is usually an open role that stays open or a second resignation in a quarter.

Define each number once, in writing, with the exact query or sheet formula beside it. The a16z note on 16 startup metrics exists because the same word means different things in different companies: bookings are not revenue, GMV is not revenue, an active user is whatever you have decided it is. A metric whose definition drifts cannot be reviewed because last week’s figure no longer means what this week’s does.

The page itself

Each tile carries six things: the number this week, the number last week, the week-on-week change as a percentage, the same week last year where the company is old enough, a six-week trendline, and the owner’s name. Rekhi’s layout rules are worth copying exactly: lay the metrics out in the order of the business model so that one flows visibly into the next; show week-on-week and year-on-year change colour-coded for direction; put a trendline beside the figure so a single bad week can be seen against the run. The trendline matters more than the percentage. A number that fell four per cent after rising for five weeks is a different fact from a number that fell four per cent for the third week running.

One person owns the page. In a ten-person company that is usually the founder who is closest to the data or the first analyst, and the job is to have it in everyone’s inbox the evening before the meeting with the numbers already checked. A page that arrives at the start of the meeting is read in the meeting, and the hour is gone. Each metric also has an owner, named on the tile, who is the person who will speak to it. Rekhi is firm on this: every metric needs a clear owner, and the meeting is run by the most senior person in the room, ideally the CEO, because a review the chief executive skips is a review the company will skip.

The hour, minute by minute

Minutes zero to five: silence. Everyone reads the page, including those who read it last night. Nobody presents. The page is the deck; there are no slides. Minutes five to thirty-five: owners speak. In page order, each owner takes two or three minutes on what moved, why they believe it moved, and whether the trend is noise or signal. The discipline Amazon imposes here is the one to borrow: the owner is expected to know what normal variation looks like for their number and to say so, rather than inventing a story for every wobble. A metric that moved within its usual range gets one sentence. A metric that broke its range gets the middle of the meeting.

Minutes thirty-five to fifty: the one anomaly. The meeting picks the single most important thing on the page that is not understood and works it. Not the three most important things; one. The question is always the same: what changed in the inputs, in the product, in the market or in the measurement? Half the time the answer is the measurement, which is why the page’s owner is in the room. If the answer cannot be found in fifteen minutes it becomes an action item with an owner and a date, which is Rekhi’s rule for unanswered questions, and it comes back next Monday with an answer. Minutes fifty to sixty: decisions. Each decision is written as a sentence with a name and a date. Nothing is discussed in these ten minutes; it is recorded. The notes go out within the hour.

A status meeting asks what everyone did. A metrics review asks what the numbers did, and the difference is the whole of operating a company.

Five ways the review decays

It becomes monthly. Rekhi’s warning is that monthly or quarterly reviews make the metrics easy to ignore, and the arithmetic agrees: a company growing ten per cent a month that waits a month to notice a stall has lost a tenth of its year. Weekly, in a fixed slot, even when the founder is travelling. Charts that cannot go down. The a16z note is blunt about cumulative charts: by definition they go up and to the right for any business showing any activity, and they can go up and to the right even when a business is shrinking. Nothing on the page is cumulative. Percentages without absolutes. A forty per cent rise from five customers to seven is written as five to seven. Both figures, always.

Delegation without inspection. Rekhi’s example is a marketing site redesign that hurt revenue for months before anyone connected the two, because the leader had handed off the metric and stopped looking. The founder reads every tile every week. That is not distrust; it is the job. Definitions that move. When a metric’s definition changes the old series is restated or the metric is retired; it is never quietly redefined so that this week compares well with last.

A worked page

A B2B SaaS company in Pune with eleven people and ₹18.4 lakh of MRR. Row one: 1,240 weekly active paid seats; 186 paying accounts; 9.2 active seats per account; 72 per cent of seats active this week. Row two: ₹18.4 lakh MRR; 14.1 months of runway at current net burn; 2.1 per cent monthly logo churn; nine of eleven open roles filled. On a Monday in the third week of the quarter the north star is down three per cent, accounts and seats per account are flat, and the share of seats active is down from 76 to 72. The owner of that tile says the drop is concentrated in accounts onboarded in the last sixty days. The fifteen-minute anomaly slot establishes that a product change two weeks earlier moved a feature those accounts were using to a new menu. The decision: the product lead reverts the change by Wednesday and the tile is re-read next Monday. Total time, one hour. Without the page, the company finds out at the quarterly churn number, three months and perhaps twelve accounts later.

Setting it up: the first four Mondays

Monday one. Choose the eight. Write the definition and query for each. Name the page owner and the eight tile owners. Put a recurring sixty-minute slot in every calendar for the same hour each week; in a ten-person company Monday at ten is usual and nothing else is booked against it. Monday two. Run the review on a page that is probably half wrong. Spend the anomaly slot on the measurement errors, which is what it is for. Monday three. Add the trendlines and the year-ago column where it exists. Retire anything nobody spoke to in two weeks. Monday four. Read the decisions from the first three Mondays and check which happened. From here the page changes only when the strategy does, and the slot does not move for anyone.

Then check three things each quarter: whether the eight still multiply up to the business, whether every decision taken in the hour was actually done, and whether the meeting has started to drift back towards people describing their week. If it has, go back to minute zero and read in silence.

The figures in the page above are illustrative. The sources are below; Rekhi’s guide is the one to read first and takes ten minutes.