Count how often, not how many

Two products each report forty thousand daily active users. From the outside they look like twins. Inside, one is a business and the other is a countdown.

The first has a dense core: eight thousand people who open it almost every day, plus a wide fringe of the curious who drift in and out. The second has no core at all — forty thousand people who each showed up once this week and mostly won’t be back, refilled every morning by ad spend. Same DAU. Opposite futures. And the single number that both teams put on the dashboard cannot tell them apart.

This is the trap of the headline metric. DAU, MAU, “active users” — they’re sums, and a sum throws away the shape of what it added up. You already know, in the abstract, that averages hide their distribution; this is the specific, operational version of that truth, and the instrument that fixes it is almost embarrassingly simple.

Count how many days each user showed up

Take a fixed window — twenty-eight days is the standard choice, because it holds four of every weekday and neutralizes the weekly rhythm. For each user, count how many of those 28 days they were active. Then plot the histogram: how many users came 1 day, 2 days, 3 … all the way to 28. That’s it. That’s the power-user curve, sometimes called the L28 (for “L”-days-active). It costs almost nothing to compute and it tells you the one thing DAU refuses to.

Because the shape is the diagnosis. A curve that decays smoothly to the right — lots of one-day users, fewer and fewer frequent ones, nothing much at the far end — is a product people try and abandon. A curve with a second hump on the right, a genuine spike of people active 25, 26, 27, 28 days out of 28, has something rarer: a habitual core. That right-side spike is the part of your user base that has wired you into their routine. The healthiest products show a “smile” — a bump on the left from newcomers, and a strong bump on the right from the devoted, with a valley of the merely-curious in between.

What you do once you can see the shape

The first payoff is that you stop celebrating the wrong wins. A marketing push that lifts DAU by flooding the left bars is not the same as a product change that thickens the right spike, even when the headline number moves identically. One rents attention; the other builds a floor. If you only watch the sum, you’ll cheer for both and can’t tell which one compounds.

The second payoff is that the curve tells you who to go learn from. Your power users — the far-right bars — have discovered a job your product does that keeps pulling them back, and it is often not the job you designed for. Interviewing the person who came 27 out of 28 days is worth ten interviews with people who came once, because they’ve already found the value you’re trying to manufacture for everyone else. The growth question stops being “how do we get more users” and becomes “what is the right-side spike doing, and can we move the valley toward it?”

The third is a caution. A power-user curve is a snapshot of frequency within a window; it is not the same as a retention curve, which tracks whether a cohort survives over time. They answer different questions — “how intensely do current users engage” versus “do users stick around” — and a healthy business generally wants both a right-side spike now and a curve that flattens across months. Watch them together, and be suspicious of any dashboard that offers you a single active-user number and calls it health. That number is an average wearing a very good disguise.