A zero-sales week means nothing on its own

You sold two things last week. This week you sold none. Something has changed, and the obvious move is to work out what: which post underperformed, which room went quiet, which part of the funnel broke over seven days.

Nothing has changed. At the volumes most independent products actually run at, a zero week and a two-sale week are the same event seen twice.

The arithmetic, briefly

Suppose your product genuinely sells at a rate of two customers a week, and suppose that rate is perfectly stable, with no seasonality, no post that landed better than the others, no change in anything. Sales still do not arrive two per week on a schedule. They arrive when individual people, each making an independent decision, happen to decide.

Count that out over a year. With an underlying rate of two per week, a week with zero sales turns up roughly one week in seven. Not as a warning sign. As the ordinary behaviour of a small number. A week with five sales turns up about as often, and it means exactly as little.

Now put those two weeks side by side, which is what every dashboard does by default. One is coloured red and one is coloured green. Both are the same underlying rate. You have been handed a 200% swing and a story to explain it, and the story will be about whatever you happened to change that week.

This is why the product refuses to draw a bar chart of weekly revenue. There is no honest way to render two data points from a small sample so that the difference between them reads as noise. The eye will not do it. The bar is taller, so something happened.

What a zero week can tell you

The count is useless on its own. Where the funnel broke is the whole message, and unlike the count, it is legible at low volume, because a broken stage shows up as a ratio across everything that came before it, not as one week's total.

What you see	What it means	Whose problem
No clicks	Wrong room, or your posts aren't landing	marketing
Clicks, no signups	The page doesn't deliver what the post promised	messaging
Signups, no activation	Onboarding	product
Activated, no payment	Value or pricing	product
Every stage clears	Scale it	—

Four of those five rows are not marketing problems. That matters, because the instinctive response to a quiet week is to post more, which is the correct response to exactly one of them. If clicks are healthy and signups aren't, another twenty comments in the same room will produce more clicks that do not convert, and you will conclude the room is dead.

Why ratios survive what counts don't

It is worth being precise about why the funnel is readable when the weekly total isn't, because the two look like the same kind of number.

A weekly sales count is one draw from a small distribution. It carries almost no information, and the information it does carry is swamped by the variance. But the funnel is not a count. It is a set of ratios, each measured against everything that came before it, accumulating from the day you entered the room. Four hundred clicks and eleven signups is a rate of 2.7%, and it stays 2.7% whether you look at it on a Monday or a Thursday. It moves when the underlying thing moves.

That is also why the stage thresholds are stated as ratios rather than as counts. Under 5% of clicks becoming signups says the page is not delivering what the post promised. Under 40% of signups reaching first value says onboarding. Under 5% of activated users paying says value or pricing. Each of those is a claim about a proportion, and a proportion measured over a few hundred events is a far steadier thing than a total measured over seven days.

There is one guard on top: below ten signups, the product declines to read the stages past the first, because a 40% activation rate computed from four people is one person changing their mind. The ratio is more robust than the count, not immune to the same problem.

The windows are not interchangeable

Two of the numbers on a community's page are measured over different periods, deliberately.

Effort is the last seven days. It is a description of what you have been doing lately, and it should move when your behaviour moves.

The funnel is cumulative since that community's entry date, never rolling. This is the one that surprises people, and it follows directly from the confidence rules. The thresholds below are cumulative counts. A rolling 30-day window would subtract evidence at the same rate it arrives, so a slow community would sit permanently just under the line and the product would say "not enough data yet" forever. The entry date is the denominator, which is also why it locks after 48 hours: nothing can reconstruct it later, and every conclusion drawn for the next six months cites it.

One comparison is worse than useless, and it is the one everyone makes: this week's effort against this week's revenue. The gap between a good comment and the subscription it eventually produces runs one to three weeks. Comparing them week-for-week blames the wrong week's work every single time, and it does so consistently enough to look like a finding.

The price of a defensible read

Three conditions have to hold before this product will say anything at all about a community:

• Three weeks since you entered the room. The lag window. Below this, the revenue from your first week's work has not arrived yet.
• Thirty cumulative clicks. Below this, a signup rate is one or two people and a percentage is theatre.
• Twelve contributions, or 120 minutes recorded. Not a productivity target, but a data-quality floor. A community you visited twice has not been tested.

All three, not two of three. Until they hold, the community's page shows the meters and what is still missing ("22 more clicks and 5 more days") and no conclusion. That is not the product being coy. It is the only honest thing to render, and it is the reason the numbers on the page can be trusted when they do appear.

The errors are not symmetric

Suppose you get it wrong. There are two ways to.

Keeping a community that was never going to work costs you a few mornings. Annoying, recoverable, and you will notice eventually. Cutting a community that was working costs you the channel that would have produced your customers, and you will never find out, because the counterfactual does not get logged anywhere.

Those two mistakes are not the same size, so the evidence required for them is not the same either. A recommendation to keep going needs one clean read. A recommendation to cut needs two consecutive reads failing at the same stage, dated, with both attached. Any improvement, or a different stage failing the second time, resets the count.

The practical effect is that you will wait about a week longer than necessary to cut something that is already dead. That is the trade, stated plainly, and it is the right way round.

The failure this prevents, concretely

Here is the sequence the confidence rules exist to interrupt, and it is worth recognising because it feels like diligence from the inside.

You enter a subreddit and spend three weeks being genuinely useful. Week four, you post something that mentions what you built. Twelve people click. Nobody signs up. Week five, eight people click, one signs up, they never come back. You conclude the room does not want this, rotate away, and spend the next two months somewhere else.

What actually happened: twenty clicks is not enough to distinguish a 2% signup rate from a 10% one, and either way the number that mattered, that the one person who did sign up never reached first value, is an onboarding problem you took with you to the next room. You will reproduce it there, conclude that room is also wrong, and eventually conclude that community marketing does not work for your product.

Every step of that is a reasonable inference from the evidence in front of you. The evidence was just too thin to support any inference at all, and nothing on the screen said so.

What to do with a zero week

Nothing. That is the honest answer, and it is why the review happens on Sunday rather than daily. There are two questions worth asking that often:

1. Which community produced a paying customer, and what did it cost?
2. Is any project past week six with nothing paid?

Everything else is checking, and daily checking is how you get discouraged by noise into abandoning the thing that was working.