Growth Model

For a long time, growth was argued about rather than calculated. Everyone agreed the user count should go up. Nobody agreed on what to do first. Marketing wanted more ad spend, product wanted to fix the first day of the experience, sales wanted one large account. None of those answers was wrong, which is exactly what made the argument unresolvable.

Quarterly planning at Pagedeck had the same shape every time. Grace would say the goal is to double active users. Ravi would answer that this means tripling the ad budget. Mia would answer that half the people who sign up disappear within a week, so the leak comes first. Both arguments were reasonable and there was no way to weigh one against the other, because the room had no shared unit for "how much does this help." The proposal that won was usually the one from whoever spoke loudest or had shipped something that worked last quarter.

The reason words couldn't settle it is that growth never has one cause. This month's active users are last month's users who stayed, plus the new ones who arrived. The share who stay and the number who arrive are moved by different teams using completely different methods, and yet they land in the same cell as a single figure. Looking at the combined number tells you nothing about which half did the work, and if you can't tell, you can't decide what to do next.

Writing it as a formula makes the comparison possible. Take three numbers: 10,000 active users, 70 percent of them still around a month later, and 2,000 new signups. Next month comes to 7,000 survivors plus 2,000 arrivals, or 9,000. Far from doubling, the product is shrinking. Raise new signups alone to 3,000 and the total gets back to 10,000, which only holds the line. Leave signups untouched and lift the survival rate to 80 percent instead, and the total is also 10,000. Once both paths sit on the page with the same result, the question in the room changes from "who is right" to "which one is cheaper and faster." A hundred million won of ad spend and a three-week onboarding rebuild become things you can put side by side and choose between.

The growth model is the name for that formula. Within the wider growth practice, this branch owns the step of deciding where to push. The funnel breaks the journey into stages, experiments run one change at a time, and analytics read the result — the model comes before all of them and works out which stage is worth touching at all.

The terms under the growth model split into two branches: the shapes growth takes when it works (structures), and the single picture that shows what it looks like when it doesn't (the metaphor).

There are five structures, and what separates them is the question of what fuels the next turn of the wheel.

1. Growth loop — a structure where the output comes back as input and keeps itself turning. It's the skeleton the other four hang on, so start here. A new user arrives, makes something, that something brings in another user, and that user makes something too, so the end attaches to the beginning. This is where a loop parts ways with a funnel. A funnel flows top to bottom once and ends, which means pouring new people in at the top every month. A loop feeds each turn's result into the next turn, so the second month is bigger than the first on the same amount of spend.
2. Viral loop — the cycle where existing users bring in new ones. Someone invites a colleague so they can read a document together, that colleague creates an account, and that colleague pulls in their own team. Whether this branch turns comes down to how many people each user brings on average. At 0.4, a hundred users bring forty, those forty bring sixteen, and the thing dies out. At 1.2, the same hundred bring a hundred and twenty, who bring a hundred and forty-four, and it grows without anyone touching it. Above or below one is the whole fork in the road.
3. Content loop — the cycle where things users make turn into search traffic. Somebody publishes a meeting-notes template, it shows up in search results, and whoever lands on it signs up and publishes a template of their own. Where a viral loop hops directly from person to person, this one routes through a search engine first, which makes it slow. A page takes three or four months to climb into the top results, and in exchange it brings people in for years afterward with no ad budget behind it.
4. Paid loop — the cycle where money earned goes back into advertising. The first two branches burn user behavior as fuel; this one burns cash. Whether it turns depends on when what a customer pays overtakes what it cost to acquire them, a figure known as customer acquisition cost (CAC). Spend 30,000 won on someone who pays that back over four months and the fifth month funds the next customer. Stretch payback to twenty months and you need twenty months of cash sitting somewhere to survive the gap.
5. Compounding growth — the pattern where accumulated results make the next round bigger. It's closer to an outcome than a mechanism, since it's what shows up when the other three loops are actually turning. Five percent a month takes 10,000 users to 10,500 in the first month, which nobody notices. Hold that rate for twelve months and it's 18,000; hold it for twenty-four and it's 32,000. The flat-looking early stretch is why plenty of teams change direction right before the curve would have bent.

The metaphor branch holds one term. A leaky bucket is the state where as much drains out as pours in, so growth stalls. Picture water going in at the top and a hole in the bottom keeping the level flat. A picture earns its place here because the formula alone doesn't land this situation. Three thousand signups a month sounds like a real number, but if active users have been stuck at 10,000 for six months, three thousand people are quietly leaving over the same period. Double the ad budget in that state and the outflow doubles with it while the level stays put. Patching the hole — raising retention, the share of people who stay — comes first, and that call falls out of this single image.

Which of the five structures to reach for depends on what the product produces while people use it. If it only becomes useful with several people in it, invitations happen on their own and the viral shape turns first. If it's a solo tool whose output ends up public, the content shape turns first. If neither is true, paid is what's left, and the ad budget becomes the growth rate. Pagedeck sits in the first category, which is why fixing one invitation screen gets treated as seriously as doubling the ad budget.

The two branches get used at different moments. Structures tell you what to build; the metaphor tells you when not to build. Adding a new loop while the bucket is leaking just increases the inflow, so the usual order is patch the hole, attach the loop after.

"The most sensitive variable in this model is week-four retention. One point there beats 500 extra signups."

Running numbers through the formula exposed the gap between two levers, and the smaller one gets parked for the quarter.

"Even a 30 percent lift in signups only moves next quarter's active users by five percent."

The ceiling on the acquisition path is already visible in the current structure, which is usually the cue to go looking for a leaky bucket.

"I don't have a number to put in that cell. We aren't counting people who arrive through invite links separately."

The formula exists but the product isn't recording what would fill it, so the next task is instrumentation rather than analysis.

At Pagedeck the formula lives as a single spreadsheet. Ravi's sheet has five rows: last month's active users, the share still around a month later, new signups, the share arriving via invitation, and the paid conversion rate. Twelve monthly columns run to the right, and changing any figure on the left recalculates all twelve.

Quarterly planning starts by putting that sheet on screen. When Grace names 30,000 active users by year-end, Ravi works backward and fills in what each cell would have to be. Last quarter the answer came out like this: hold the survival rate at 68 percent and rely on acquisition, and monthly signups need to reach 9,000. The current number is 3,200, so that's a tripling. In the same sitting, lifting the survival rate to 78 percent hit the identical target with 4,500 signups.

Those two lines changed the conversation. Mia brought the month-later retention gap between accounts that invited a teammate in week one and accounts that stayed solo. Theo put the invitation-flow rework at three weeks. The decision was to hold ad spend flat and rebuild the invite flow. Without the sheet, the same meeting would have spent an hour on whose argument sounded better.

The most common mistake is reading the formula as a forecast. Twelve months of figures come out to the decimal, so it looks like one. But the inputs are averages from the last few months, and those averages can move next month. What the model actually does is hold two options against the same set of assumptions and show which one is larger. Read it as a forecast and the first time reality lands somewhere else, three months later, the whole thing gets written off as wrong — and what gets thrown away isn't a prediction, it's the habit of comparing. The quarter after that, the loudest person wins again.

The belief that a better model is a more detailed one goes wrong just as often. More variables feel closer to reality, but a sheet with forty cells brings three problems at once. First, every cell needs a number, and half of them aren't recorded anywhere in the product, so estimates go in. Second, when the output moves, nobody can point to which cell moved it. Third, nobody but the author ever opens it. Models that get used tend to have three to five variables, because reducing the number of levers is the point rather than reproducing reality. The team that builds the forty-cell version ends up six months later with a file nobody updates, and meetings run without opening it.