Founding as a Developer — Validate Before You Build
How You Define an Active User Changes the Number
In one line
DAU, WAU, and MAU are not numbers until you decide three things: "what counts a person as active", "which time zone's day", and "whom to exclude". And the number a product team watches every week (the North Star) must count not logins but the behavior through which customers get value.
Why this was needed
A small team captures its dashboard the day before an investor meeting. MAU is up 20% from last month. But a good part of that 20% was internal accounts — developers, designers, and salespeople were logging in every day for demos. Another team saw its metric jump the week it changed the definition of "active" from "login" to "any event". That was because merely opening a push notification logged an event. In both cases the calculation was not wrong. The definition simply had not been written down.
In the early days of a startup, you decide hiring, funding, and feature priorities with these numbers. When the definition changes, the trend lies, and decisions made by trusting that trend are hard to undo. So a metric is something you treat like code: write the definition first, then recompute with that definition and check that the same number comes out.
How it works
To define one metric, fill in at least the four slots below.
| Slot | This lab's definition | Where it usually leaks |
|---|---|---|
| What | session_start at least once |
If you leave it as "any event", notification opens and automatic syncs get mixed in |
| Whom | Exclude is_internal = 1 |
Internal and test accounts log in every day and raise the floor |
| When | Asia/Seoul date, weeks start Monday 00:00 | Grouping by UTC date sends activity before 9 a.m. to the previous day |
| How many | Number of people (deduplicated) | If you count events, one person who logs in ten times becomes ten people |
DAU, WAU, and MAU. DAU is the number of people active that day, and WAU and MAU are the number of people active at least once within the past 7 and 28 days, including the reference day. The reason to use 28 days is that a calendar month is 28–31 days long, so month-by-month comparisons are swayed by the weekday mix — a 28-day window always contains each weekday four times, whenever you measure. This lab defines it that way, and if your organization uses 30 days, you only need to write that definition down. What matters is not which one, but that it does not change.
Stickiness. It is a ratio that looks at "of the people who came at least once a month, how many come on a typical weekday". This lab divides the average DAU over the 28 days up to the reference day by that day's MAU. If you divide by a single day's DAU, the number swings depending on whether that day is a Sunday or a Monday.
North Star. Amplitude's North Star Metric article lists as the conditions of a good North Star that it connects with the value customers actually get, that it reveals the product strategy, and that it is a leading indicator that moves ahead of revenue, and it advises avoiding vanity metrics such as DAU or the number of signups, and lagging indicators such as monthly revenue. For a team document tool, "people who shared a document with someone else" is closer to the value than "people who created a document" — the value of a collaboration tool arises when a second person comes in. So this lab calculates "weekly sharing users (people who did share_doc at least once that week)" as a North Star candidate. This does not mean it is the right answer. The purpose is to pick a candidate and look at what moves differently from other definitions of active on the same data.
Activation. Decide the moment when a person who signed up first tastes the value, and measure the share who reach that moment within a set time after signing up. In this lab it is "create_doc within 7×24 hours of the signup time". If you count 7 days by calendar date, someone who signs up at 11 p.m. gets one day less — count by time of day.
What it looks like in the field
- The MAU sent to investors differs from the MAU on the internal dashboard. One excluded internal accounts and the other did not. With no definition document, nobody knows which is right.
- It is a service in Korea, but the analytics tool's default time zone is UTC, so activity from midnight to 9 a.m. goes into the previous day. The weekday patterns look shifted by one slot.
- DAU jumps in the week a push notification is sent in bulk. That is because "any event" was counted as active. A notification is a cost, not value.
The example numbers below were computed from this lab's materials and are not general reference values. Even the DAU of the same day varies by tens of percent depending on whether you include internal accounts and which events you count — you compare them directly in the lab.
What you will do in the next lab
From 16 weeks of "Moanote" records, you build DAU, WAU, MAU, and stickiness as functions, and measure how much the numbers swing when internal accounts are included and when the definition of active changes. Finally, you calculate weekly sharing users and the 7-day activation rate and bundle them into a one-page report. The grader also runs your functions against records made with a different seed, to check that you did not just write down memorized numbers.