Retention Stage

When signups were only ever counted as a running total, there was no way to tell whether a product was working. A cumulative signup count never subtracts anyone who left. Someone who registered once and never opened the product again still sits inside that number, so it climbs every month even while the product is falling apart.

Pagedeck hit that wall in an investor meeting. Grace put up 50,000 total signups and the investor asked one question: how many of those came in last month. The answer was 3,100. The 50,000 was four years of arithmetic, and the people who actually opened the product last month were six percent of it.

So the funnel gained a stage that groups users by when they arrived and counts how many of each group are still around weeks later. It measures survival rather than accumulation, and when this stage is empty, nothing accumulates inside the product no matter how many people the earlier stages let in.

The stage is measured by taking everyone who arrived in the same period, then counting how many remain after one week, four weeks, and twelve. Grouping users by arrival date and following each group through time is called cohort analysis. If 1,000 people signed up in September and 320 are left after a week, 210 after four weeks, and 190 after twelve, that sequence is the September cohort's curve.

What you read is the shape of the curve, not the last number on it. A curve that falls steeply and then goes nearly flat means the product has a floor. Those 190 people stayed put for twelve straight weeks, which says the product found a place in their work. A curve still sloping at 40 people by week twelve and 12 by week twenty-four has no floor, and a product like that empties out the moment you stop buying traffic.

Where that floor belongs differs by product, and the reference point is how often people do the underlying job at all. A collaborative document tool like Pagedeck is in trouble if people don't come back weekly, while a product that sells plane tickets is fine with two visits a year. Borrowing another company's percentage as a target doesn't survive that difference. The practical move is to pick the interval that matches the natural rhythm of the job — weekly for Pagedeck, yearly for the ticket seller — and read every cohort at that interval from then on.

This is also the slowest stage to read and the hardest to fix. Acquisition responds to more ad spend by next week. This stage only moves when the product genuinely becomes more useful, and confirming that it moved takes another four to twelve weeks of waiting for a cohort to age.

Whether you're ready for referral, the next stage, comes down to whether the curve has flattened. Three conditions.

1. The curve settles at some value. Four-week and twelve-week counts of 210 and 190 have settled; 210 and 90 have not.
2. Enough people remain at that floor to count. If each cohort leaves under ten survivors, next month's different-looking number can't be separated from noise.
3. You can say in one sentence why they stay. At Pagedeck, nearly everyone who stayed was sharing a document with two or more colleagues. Knowing the reason is what tells you what to charge for in the revenue stage.

None of the three is a judgment call about whether the product is good enough. They are checks on whether the numbers in the later stages will mean anything once you start moving them. Run the stages out of order and the later ones inherit the earlier one's problem. Offer an invite reward while nobody is sticking around, and you add people who collect the reward and leave, which only raises acquisition cost. Put up a payment screen before the curve reaches its floor and the people who pay ask for a refund the following month. So no target gets attached to the later stages until this one holds.

"This quarter we're holding acquisition spend flat and only watching four-week survival."

Instead of letting more people in, fix whether the people already arriving stay.

"The August cohort still hasn't bent at week twelve. What changed that month?"

One period's users held unusually well, so find the cause and apply it to the other months.

"That's a retention metric. What we're looking at is conversion across this whole stage."

Drawing a line between a single number and a segment of the funnel.

Pagedeck first looked at this stage properly right after its Series A. Mia laid out every monthly cohort side by side and one thing stood out. One-week survival had been 32 percent for four months running, while four-week survival had slid from 24 percent to 18. The product was holding people through the first week exactly as well as before, and losing more of them before the month was out.

The cause turned out to sit outside the product. The ads Ravi had been running in that window read like a pitch for a personal note-taking app, and people arriving through that route used it alone and left. Rewriting the copy around team documents cut incoming traffic by 20 percent and brought four-week survival back up to 23. It's actually rare for this stage's problem to live inside this stage.

The common misreading is that the stage is about getting people to come back, full stop. It's measured in return visits, so the assumption follows naturally. But return visits go up on notifications alone. Send a push every morning and more people open the app, and that week's number improves. The trouble is that traffic dragged in this way doesn't turn into anyone using the product. Two months on, people mute the notifications or delete the app, and the curve flattens lower than it did before. The number that looked better for six weeks bought a worse floor. What the stage is trying to see isn't visit count — it's whether someone is still getting work done in the product.

Hunting for an industry benchmark to aim at is the other well-worn path. Someone reads that 40 percent is a good figure and adopts it. But that figure means nothing until you fix which week it's measured at and what action counts as having stayed. Forty percent at a product that counts a login and 40 percent at one that requires editing a document are different numbers. What comes first is the definition, not the target, and once the definition is set the thing worth comparing against is last month's cohort rather than another company.