Only 10 of 431 cases disclose churn. At 3% a month the ceiling is 33x new MRR, at 10% it is 10x

When we wrote how far one person can go in July 2026, we settled the question of ceilings with hours worked. Counting 121 solo and side businesses, what separated the orders of magnitude in monthly revenue was not the hours but whether the revenue had been decoupled from the owner’s own labour. We stand by that count. What we failed to count was what happens after the decoupling.

The standard way to decouple revenue from labour is a monthly subscription. Once someone signs, money arrives in months when you do nothing. In exchange, a second ceiling arrives, and it has nothing to do with labour. Every month, some percentage of the contracts disappears.

We searched the full text of all 431 published cases for churn, retention and cancellation rates. Ten cases state an actual figure. Of those ten, only five can be laid side by side on the same measure.

Those five run from under 3% a month to under 10% a month. As numbers they look like rounding error. Convert them into ceiling multiples and they become 33x and 10x. For the same volume of new business, the point at which revenue goes flat differs by more than three times.

Ten cases out of 431 state a rate, and only five use the same measure

The bottom five cannot sit in the same column as the top five. Ghost reports net churn, which already has expansion revenue subtracted out, and even when we wrote that article the split was unreadable from outside. The Lab offers annual billing only, so a monthly churn rate does not exist there. Social Burro runs on six to twelve month retainers rather than monthly subscriptions, and 75% is renewal on those contracts. The 18.8% at IsTalk counts people who opened the app again the following week, not people who kept paying. Noosa Labs gives the improvement with no starting level.

One word, churn, pointed at six different numbers across ten cases. At that point, taking an industry average and comparing your own figure against it stops meaning very much.

At 3% a month the ceiling is 33x, at 10% it is 10x

Where subscription revenue finally stops can be approximated from two numbers.

Steady-state MRR is roughly new MRR each month divided by monthly churn

The formula comes from the case study behind the ScreenshotOne article, which states it as Max MRR equals new MRR divided by churn, and we used the same formula for an estimate in the Data Fetcher article. The multiples below, 1 divided by churn, are our own arithmetic rather than anything the operators published.

The median across the five is 7% a month, a multiple of 14x. Two of the five are stated as “under” a value, which we treated as upper bounds, so the real median may sit lower.

What makes the formula unkind is that improvement does not arrive in a straight line. Going from 10% to 9% a month moves the ceiling from 10x to 11x. Going from 4% to 3% moves it from 25x to 33x. A single point weighs more the lower you already are. A business starting from the top and working downward feels nothing for a long time, and that feeling is the shape of the fraction rather than a failure of effort.

One of them had actually hit the ceiling

ScreenshotOne is the only one of the ten to publish the numerator as well. The screenshot API that Dmytro Krasun runs alone out of Ukraine takes in roughly $3,000 of new MRR a month against 9% monthly churn. Put those in the formula and the answer is $33,333. The MRR it actually reached after four years was $32,000, and there it stopped.

Among our published SaaS cases, the 106 that disclose monthly revenue have a median of ¥6.86 million a month, so $32,000 sits unremarkably inside that distribution. The part that matters is that the place where this business would stop was computable four years earlier.

The case study puts the cause of that 9% not in product quality but in how long the job lasts. Once the launch graphic is made, once the scraping project ends, customers leave satisfied. Four years of collecting more than a thousand customers has not yet turned up a segment with a reason to keep using it.

We read those four years not as a record of failure but as a record of growing the numerator while leaving the denominator untouched.

The three that brought it down never touched the product

Look at the cases that lowered churn or hold it low, and they are all doing the same thing.

Data Fetcher sat flat at $6,000 MRR for five months from August 2023. In December the founder did two things. He put a 50% discount on the annual plan, and he switched the default tab on the pricing page from monthly to annual. Monthly churn fell from 10 to 11% down to 7%, and the stalled growth restarted. In multiples, that is a move from 9x to 14x. The product itself was left alone.

The Lab has no monthly plan at all. The paid community bills annually only, and the stated reason is that a monthly membership is a setup that fails in a place built on peer relationships. Retention at first renewal is 94% and stays above 90% after that. Membership is capped at 200 people.

IsTalk comes at it from the other end. Weekly usage retention is 18.8%, so most users never open the app again the following week. The design assumes that brevity and converts people in their hottest first week, at a 37.5% conversion rate, onto a ¥2,500 annual plan. The ¥2.5 million a month follows from that.

None of the three changed how the product gets used. What they changed was how often the customer gets to decide. A monthly plan hands out twelve decisions a year. An annual plan hands out one.

Annual billing does not plug the hole. It cuts the number of decisions

Put plainly, calling what those three did an improvement in churn is a little generous. As the ScreenshotOne source points out, what drives the rate up is the lifespan of the job. Moving to annual billing does not make a customer who finished their launch graphic start hitting the API again next month. What moved is that the moment where the customer says “stop” went from twelve a year to one.

The ceiling does genuinely rise anyway. What enters the steady-state calculation is whether the contract continues, not whether the product gets used, and annual billing is paid up front, so the cash arrives earlier too. As a number, putting a lid over the hole makes the hole smaller.

The price shows up in the same place. Allowing one decision a year also means detecting dissatisfaction once a year. The Lab capping membership at 200 and building for a long relationship with the same people reads as paying that price in headcount. Copy the annual billing alone and grow the member count, and a year of undetected complaints piles up.

From here we are speculating, but annual billing works only where the job outlasts the contract. Put a yearly plan on a product whose use is finished in three months and it comes back as refund requests and bad reviews. IsTalk pricing its annual plan at ¥2,500 keeps that order of operations intact.

The 3% end sat where the customer’s work never ends

The lowest figure, FeedbackPanda at under 3% a month, was not produced by a billing cycle. It sits inside a task that recurs daily, the feedback an online English teacher writes after every lesson. Around 5,000 customers, MRR $55,000, sold to SureSwift Capital for seven figures two years after founding.

That 3% carried a second meaning. Most of the customers taught for one particular online English company, and a change of policy there would have taken them all at once. Arvid Kahl sold before that risk showed up. A low churn rate is also a sign that the window for selling at the top is open.

Ghost sits at the low end too with 2.92% net churn, though that figure nets off expansion and cannot go in the same column as the five above. This is a company that built ARR of $11,055,673 over thirteen years and publishes the number live on its own site. The more telling fact is probably that publishers do not switch CMS.

What the low end has in common is not product quality. It is that the customer’s work never finishes. Teachers teach every day, publishers publish every day.

What we could count and what we could not

This aggregation has more limits than findings.

Only five cases were comparable, and not one of the five states whether the rate is customer churn or revenue churn. Two are given as “under” a value and were treated as upper bounds, which drags the 7% median upward.

The ceiling could be verified against reality exactly once. No other case publishes the new MRR that forms the numerator, so for the remaining nine we can only produce multiples. A multiple approximates a steady state, and it does not describe where a growing business stands today.

The population is skewed as well. Ten cases out of 431, and the ones that exist cluster around people who were about to sell or who wrote up something that had stopped. There is rarely a reason to publish your churn rate alone while things are going well.

For distributions there is how far one person can go, which aggregates monthly revenue for solo and side businesses, and the map by business type. The SaaS cases are listed on the SaaS topic page, and the definitions sit in how to read the numbers.

There is no term for effort in this formula

The four years at ScreenshotOne are not a record of insufficient effort. The founder added $3,000 of new MRR every month, collected more than a thousand customers, and still went flat at $32,000. Leave the denominator alone and the numerator moves the ceiling only in proportion. Search the formula and there is nowhere for trying harder to enter.

What we wrote in July, that hours worked do not set the ceiling, stays where it is. What we would add is that beyond decoupling revenue from labour there is a second ceiling made of churn, and that one is set by when the customer’s job ends rather than by how good the product is. Decoupling was not the end of it.

The part that sits least comfortably is that this ceiling works perfectly well on people who have never heard of it. Nine of the ten publish no numerator, so from outside there is no way to tell whether they know where their own ceiling is. Starting a subscription business, the calculation needs two numbers, new MRR and churn, and both are in your hands from the first month. Nobody runs it anyway.