Leaky Bucket

For a long time the only number a team carried into a growth review was new signups. Three thousand this month, two hundred more than last month. Count only the people coming in and the chart always points up and to the right, so the meeting ends on a good note.

The mismatch surfaced in a different meeting. At Pagedeck, three thousand people signed up every month for half a year while monthly active users (MAU) — the count of people who opened the product at least once in a month — sat around ten thousand and refused to move. Marketing reported that acquisition was growing, product reported that the user base was not, and both reports were accurate. Weeks went by with people questioning each other's numbers.

What was missing was the number on the other side. Count arrivals without counting departures and those two reports never meet. Leaky bucket is the phrase that forced both onto one screen. Water pours in at the top, a hole in the bottom keeps the level from rising, and that single picture moves attention from inflow alone to the gap between inflow and outflow.

The water level is set by what you pour in minus what leaks out, and the active user count works the same way: this month's number is last month's number plus the people who arrived minus the people who left.

Two quantities move it over a month.

1. New arrivals — people using the product for the first time this month, whether they came from ads, search, or an invitation from a colleague.
2. Departures — people who used it last month and did not show up this month. Deleting an account is the rare version; quietly not coming back is the common one. That pattern of users leaving is called churn.

Numbers make the leak visible. Take a product with ten thousand active users and a monthly churn rate of 30 percent: three thousand people leave every month. If exactly three thousand new users arrive, next month is ten thousand and the month after that is ten thousand. Six months and eighteen thousand new users later, the number on the dashboard is what it was half a year ago.

Now double the ad budget and bring in six thousand a month. The number rises, but not forever, because a larger base means a larger 30 percent leaving alongside it. At thirteen thousand users, thirty-nine hundred leave. Growth stops at twenty thousand, where inflow and outflow match. The level a product settles at is monthly arrivals divided by the churn rate.

Work on the other side and you arrive at the same place for a different price. Keep arrivals at three thousand and cut churn from 30 percent to 15 and the ceiling is still twenty thousand. The difference is that the ad route costs double every month to hold that level, while the churn route holds on its own once it is fixed. Raising retention — how reliably existing users keep coming back — buys the same water level far more cheaply.

The phrase comes up most often in the meeting about whether to spend more on ads. Channel numbers look fine, the total refuses to move, and the room splits between spending more and calling the current spend wasted. If customer acquisition cost (CAC) — the money spent to bring in one customer — is $25 and that customer leaves after two months, doubling the budget doubles the loss.

The second place is the work of locating the hole. Cohort analysis, which groups users by when they arrived and tracks how many of them remain over time, usually shows more than half disappearing in the first week. That puts the hole in onboarding, the stretch where a new user is still finding their footing, rather than among long-time users. People are leaving before they ever reach activation, the moment the product's value lands for the first time.

The third is deciding when to build new growth machinery. A growth loop, where what users produce brings in more users, normally gets attached after someone confirms the bucket is not leaking. Users who do not stay do not bring anyone else in either.

"The bucket is leaking right now, so pouring more money into ads doesn't buy us anything."

More spend means more arrivals and an equal number of departures, so hold off on the increase.

"Three thousand signups a month and active users stuck at ten thousand for six months. That's a leaky bucket."

Acquisition numbers on their own don't show that the product is growing.

"Let's patch the hole before we pour. This quarter goes to retention."

Acquisition work is being deferred so the team can reduce departures first.

At Pagedeck the phrase came out when someone proposed doubling the paid channel budget. Ravi showed that cost per signup had held steady for three months. In the same meeting Mia opened a cohort table: of everyone who had signed up over the previous six months, 18 percent were still around in week four, and that figure barely moved no matter how the budget was spent.

Grace pushed the increase to the next quarter and gave a quarter to onboarding work instead. Leaky bucket was the phrase she used to explain the call. It didn't mean the ads had failed; it meant that spending more in the current state would lose the extra at exactly the same rate. Four months later, with week-four retention at 31 percent, the budget conversation came back.

The common mistake is treating a sealed bucket as the goal. The image is vivid enough that zero leakage looks like the normal state. But no product has zero churn. Someone who tries the product, finds it doesn't fit their work, and leaves is a healthy outcome, and chasing that person too tends to swell the product into something that fits nobody. The target isn't zero; it's arrivals comfortably exceeding departures, and where that line sits differs by product.

The opposite reading causes as much trouble: that a leaky-bucket diagnosis means acquisition has to stop. Pouring water into a leaking bucket feels wasteful, so the conclusion follows naturally. But cut off arrivals entirely and the material for checking whether the hole is fixed disappears with them, because retention is measured by watching new arrivals stay over time. What actually happens is not a stop but a postponement of the planned increase.