Usage Metrics

For a long time the answer to "how is the product doing?" was the signup count. It is easy to collect, it never goes down, and it looks good on a slide. The trouble is that it adds up everyone who ever created an account. Someone who signed up two years ago and never returned is still sitting in that total. A product with 120,000 accounts and a product where 120,000 people show up every day are completely different things, and the dashboard shows them the same number.

The gap usually surfaces when money enters the conversation. Pagedeck was sizing a server upgrade and Theo asked whether the team should plan for 120,000 people. Nobody in the room could answer. Counting the logs gave a different picture: 8,400 accounts had opened a document at least once in the last 30 days, and 900 of those came back every single day. The capacity worth paying for sat somewhere between 900 and 8,400, not at 120,000. A second problem showed up in the same meeting. About half of those 8,400 accounts only read documents and left, while the load on the servers came from people editing together, and a headcount alone could not separate the two.

Usage metrics exist to fill that gap. They move what gets measured one step in two directions — from cumulative to windowed, and from people to actions. Instead of total signups they count who showed up in the last day, week, or month, and instead of stopping at how many people arrived they count what those people did once inside. The result is a number that can fall. A number that can fall is the only kind that shows a product getting worse while it is getting worse. Pagedeck moved the top line of its weekly report from cumulative signups to weekly active users after that meeting. The headline dropped from 120,000 to 3,200, and only after it dropped did that line start telling anyone what to do this week.

Within the product axis this family sits inside product metrics, and it differs from its three neighbors in what moment it measures. Activation covers the point where a new signup first feels what the product is for. Retention covers whether those people keep coming back afterward. Satisfaction metrics cover the part no log captures, asked directly through surveys. Usage metrics sit in between and measure how much the product is running right now. If retention asks whether people return, usage metrics ask what they do once they have returned.

Seven terms sit under this one, split into two branches by what gets counted: people, or the things those people did. The first branch reads scale and the second reads depth. Looking at only one of them makes the picture wrong, because a product where the crowd grew but nobody goes deep looks identical on screen to a product where the crowd held steady and the depth increased.

Active users is the branch that fixes a window of time and counts everyone who showed up inside it. The four terms differ by window length and by ratio.

1. Daily active users (DAU) — the number of people who used the product on a given day. The short window makes changes visible fast, but it swings hard with the day of the week. Work tools routinely drop to a third of their weekday figure on Saturday, so anyone reading a single day as up or down is usually reading the calendar.
2. Weekly active users (WAU) — the number of people who used the product over a week. The weekday effect cancels out inside the window, which makes this the easiest one to read as a trend. It is the figure Pagedeck keeps on the wall in its weekly meeting.
3. Monthly active users (MAU) — the number of people who used the product over a month. This is the figure that goes outside the company when someone describes scale, and it is blunt for internal decisions because showing up once in thirty days is enough to be counted.
4. DAU/MAU ratio — the share of the monthly crowd that shows up daily, found by dividing daily active users by monthly active users. Divide 900 by 8,400 and the answer is 11 percent, meaning about one in nine of the people who touch the product in a month touches it every day. Where the first three describe size, this one describes attachment.

Three windows are worth keeping side by side because each answers a different question. The daily figure answers what happened yesterday, the weekly one answers which way this month is leaning, and the monthly one answers how large the product currently is. Drop two of them and two of those questions lose their answer. Keeping all three also exposes something in the relationship between them: a product whose weekly figure is more than half its monthly figure has almost no once-a-month visitors, while one below a quarter is mostly made of people who drop in occasionally.

Behavior is the branch that counts what people did after they arrived. The three terms differ in what they group together.

1. Session — one visit, from arrival to departure, treated as a single unit. The usual cutoff is thirty minutes of no activity. How many visits a person makes and how long each one lasts both come from here.
2. Feature adoption rate — the share of people who have tried a specific feature at least once. If comments shipped a month ago and only 4 percent of weekly active users have touched it, the next question is whether the feature is weak or whether nobody knows it exists. The denominator has to be decided in advance too. Using every registered account keeps the number permanently low, while using the people who reached the screen the feature lives on separates a problem with the feature from a problem with the path to it.
3. Power user — someone who uses the product far more deeply and frequently than the average. Where to draw that line is a per-product decision. Pagedeck counts an account as one if it shows up five or more days a week and co-edits at least three documents.

None of the seven has a value that the name alone determines. Defining a single active-user number takes four decisions written down in advance.

1. What counts as using it — whether someone who logged in and left is in or out moves the figure by roughly 20 percent. Pagedeck counts only accounts that opened at least one document.
2. Where the window starts and stops — counting from the first of the month to the last day gives a different answer than counting backward thirty days from today. The calendar version makes the number look like it collapses at the start of every month.
3. How one person is counted — accounts and devices give different totals, and someone using a work login and a personal login shows up twice.
4. Who is excluded — leave internal staff accounts and automated accounts in and a company of 45 people inflates its own daily figure by 45.

Good levels do not come from outside. The most common mistake is reading that the industry average attachment ratio is 20 percent, seeing 11 percent, and calling it bad. A product people have a reason to open daily and a product people have a reason to open monthly play in different ranges from the start. Messaging apps clear 50 percent; a payroll tool at 5 percent is behaving normally. Three references work better: this product's figure last quarter, the figure before and after a specific change, and the figure the business plan requires, worked backward from the revenue it needs.

The movement of one number against another matters as much as either alone. If daily active users climb from 900 to 1,100 while monthly active users climb from 8,400 to 12,000, attachment has fallen, not risen. Advertising brought in a crowd that looked once and left. This is why the family belongs on one screen. Isolate any single figure and there is a way to narrate whichever direction it moved as good news.

"DAU is down 12 percent this week, but last week had a holiday in it."

The drop is a calendar artifact rather than a product problem.

"MAU looks great in the board deck, but only 11 percent of those people show up daily."

The crowd grew while attachment stayed flat, so this is not a moment to celebrate.

"Adoption on that feature is 4 percent, and people who try it keep using it. Sounds like a discovery problem."

The feature works; the route to it does not.

Pagedeck reviews these figures together every Monday morning. Mia puts weekly active users and the attachment ratio on screen first, and the team chases explanations only for the ones that moved more than 3 percent against the prior week. Last month weekly active users rose four weeks running while attachment slid from 12 percent to 9 percent. Ravi traced it back to an advertising channel he had just funded: teams arriving through it created one document and never returned at three times the rate of everyone else.

Theo watches the same screen for a different reason, because the timing of the next server upgrade hangs on actions rather than headcount. Capacity is now planned against the number of concurrent editing sessions at the busiest moment of the week, not against weekly active users. Looking at one screen, Mia reads the people first and Theo reads the actions first.

The most common mistake is reading more usage as a better product. A larger figure looks like a product being used well, which is a fair assumption on its face. But usage carries two things at once: the value the product delivers and the effort it takes to get that value. A product that makes people click through six screens to find one document scores high on screens per session, and that score falls once the annoyance is fixed. Time spent behaves the same way. On a product that sells something to watch, longer is better; on a work tool, cutting a three-minute task down to forty seconds is the win. Which direction is good depends on what the product promised its users. Without settling that first, a product that got harder to use can look improved on the dashboard.

The second mistake is reading monthly active users as the size of the business. It is the figure that travels outside, so a big one makes the company look big. Inside that count, though, a person who showed up once and a person who showed up daily are both a 1. So is a free account used by one person and a paid team account shared by forty. Planning a sales target against 8,400 monthly active users and discovering later that only 600 of them are team accounts capable of paying is exactly how this goes wrong. Usage metrics measure how much a product is used; a different set measures how much money comes in. Connecting the two takes a separate count of how someone using the product turns into someone paying for it, and tracking when free teams cross into a paid plan is what finally joins the two figures.