Magic Number

What a newly signed-up user should be asked to do first was settled by taste for a long time. One team made people upload a profile photo. Another walked them through five feature slides. Ask for the reasoning and the answer came back as "it just feels natural," and that feeling was usually the instinct of whoever built it.

Pagedeck's first onboarding meeting ran the same way. Noa wanted new users to pick a template on the opening screen. Theo wanted them to invite a teammate first. Both sounded reasonable, and nobody knew which one produced better retention — whether users who started that way kept coming back. When opinions split, the loudest or most senior voice decided.

Finding a magic number replaces that moment with a record. It opens the behavioral logs of users already on the product and compares what people who stayed did early on against what people who left did. Once a specific action and a specific count separate those two groups, the question of what onboarding should push narrows to one line.

The search runs in four steps.

1. Pick candidate actions. They have to sit directly against the value of the product — for Pagedeck, creating a document, inviting a teammate, leaving a comment. Login counts and time on screen get dropped, because a higher number there doesn't mean a user got more out of the product.
2. Set a window. The first 7 or first 14 days after signup are the common choices. Stretch the window too far and behavior from people who had already decided to stay leaks in and inflates the threshold.
3. Compare retention across counts. This uses cohort analysis, which groups users by when they signed up before comparing them. Skip the grouping and last month's users from a large ad campaign sit in the same table as word-of-mouth users from two years ago, and the number wobbles.
4. Read where the share bends. If users who created zero documents in week one are at 5 percent after two months, one document at 12 percent, two at 18 percent, three at 60 percent, and four at 63 percent, the threshold is three. The whole 40-point gap opens between two and three, and everything past it is nearly flat.

No bend means this action isn't the answer. If the share climbs gently from zero through five, go back to step one with a different candidate. Forcing a threshold onto a smooth curve puts every later decision on a line that isn't there.

None of this works unless the behavior is being recorded. On a product where creating a document leaves no trace anywhere, no amount of users produces this number, and there is no way to manufacture last quarter's logs after the fact. Event tracking has to be built months ahead of the analysis that needs it.

Once the number exists, work across several teams lines up behind it.

1. Onboarding screens — the goal shifts from "let them look around" to "get three documents created in week one." Onboarding is the process of helping a new user settle in, and writing its goal as a number settles which prompts get cut and which buttons move forward without argument.
2. Lifecycle email — a message goes out on day five only to users sitting at one document, and not to anyone already past three. Fewer messages go out than under a single blast to everyone, and response rates rise.
3. Intervention thresholds — Sofia, who handles enterprise accounts, reaches out when fewer than half the seats in a customer's team have crossed the threshold two weeks after signing. That is months earlier than reacting to a cancellation notice, and the conversation has something concrete in it rather than a general check-in.
4. Metric design — the share of new users crossing the threshold becomes a term inside north star decomposition. When the headline metric gets broken into the pieces that move it, this value serves as one of the handles.

"Our magic number is three documents in week one. Let's rebuild onboarding around it."

The threshold is settled, so the early screens should aim at one goal.

"Is that actually the magic number? Couldn't it just be that people who were going to stick around create more?"

The number hasn't been shown to be a cause rather than a symptom.

"I rechecked the threshold after the pricing change and it moved from three to five."

The product shifted, so the old value no longer describes it.

Ravi found the number at Pagedeck. What prompted it was doubling ad spend and seeing almost no change in how many users remained two months later. Splitting signups by documents created in week one showed retention breaking at three — 18 percent below it, 60 percent above — and almost everyone arriving through ads sat at one or zero.

Mia heard the presentation and argued against acting on it directly, because nothing separated "creating documents makes people stay" from "people who were going to stay create documents." So half of new signups got a prompt encouraging document creation in week one, the test ran four weeks, and only after that group's two-month retention actually came in higher did onboarding get rebuilt around the number.

The most common error is reading the threshold as a lever: push users past it and they stay. The gap between the two groups is so wide that it reads as causation. But the number shows two things moving together and says nothing about which produced which. People who arrived already intending to use the product may simply have created more documents and also stayed. Act without that distinction and onboarding starts forcing three documents, which produces users with three empty documents and the same two-month retention as before. That is why finding the value is followed by applying it to part of the audience and checking the result. The threshold earns its place in onboarding only after a group that was nudged toward it actually retains better than a group that wasn't.

Treating it as a permanent constant is the other frequent mistake. It resolves to a single digit, so it looks like a fixed fact about the product. But the value is attached to the product and the acquisition channels. Change the pricing or promote a new feature to the center of the experience and the action that carries the value changes with it; bring users in through a different channel and the same action means something different. Recompute it quarterly, and when the threshold moves, move the prompts and the emails with it. A team still optimizing toward a number found eighteen months ago is usually pushing users toward an action the product no longer depends on.