Power User Curve

For years the headline number for product usage was a single line: how many people used the product in a month, or monthly active users (MAU). That number reports size. It says nothing about how tightly anyone is attached.

Put two products with 30,000 monthly users side by side. In the first, all 30,000 come in once and leave. In the second, 25,000 drop by once or twice and 5,000 show up on more than twenty days. The slide says 30,000 either way, but the first product empties out within a month of a decent competitor appearing, while in the second those 5,000 people are the ones who pull the team next door into the product.

The gap showed up as a question nobody could answer. In an investor meeting Grace gave the 30,000 figure and was asked how many of those people used the product every day. There was no prepared answer. The average number of active days, pulled together in a hurry, came to 3.4 — and one average can't separate 30,000 people who each show up three or four times from a crowd that visits once while a small group shows up daily. The curve exists to put back the distribution the average erased.

Count active days per user over the last 30 days, group the users who share a day count, and stack them. Five steps.

1. Fix the window and the action. The window is usually the trailing 30 days. Then pick one action that counts as "used it." For Pagedeck the fork is whether opening a document counts or only editing one does. The first choice produces a generous curve and the second a strict one, and the two are not comparable.
2. Count days per user, not sessions. Ten visits in one day is still one day. Counting days separates one long sitting from a habit of dropping in briefly every morning.
3. Bucket users by day count. You end up with thirty buckets: 8,200 people at one day, 4,100 at two days, 2,700 at three, and so on out to thirty.
4. Convert each bucket to a share of all active users. Plotting raw headcounts makes a month where the user base grew look identical to a month where the shape changed.
5. Plot day count from 1 to 30 on the horizontal axis and that share on the vertical axis as bars.

The reading is on the right-hand side. Most products slope down from left to right like a slide, with the tallest bar at one day. When the last few bars rise again into something like a smile, a group of near-daily users clearly exists — hence the name smile curve that the industry uses for the shape.

One chart is rarely enough. Draw this month's signups and users of three months or more as separate curves and overlay them, and you can see how thick the right tail gets as people stay. That is the same logic as cohort analysis, which compares groups by when they arrived. None of these numbers exist unless event tracking — the work of recording what users actually clicked — was designed months earlier.

The curve only makes sense when the product contains a reason to come back daily. Collaboration documents, chat tools, and calendars get opened and left open during the workday, so users with twenty active days out of thirty genuinely exist and the right tail carries information.

The other case is just as common. A tax filing service or a job posting tool is fully used at twice a month. A right tail near zero there is normal, not a symptom, and treating the curve as something to improve spends a quarter manufacturing visits nobody wanted. Decide first whether usage frequency tracks the value the product delivers, and only then reach for this chart.

Three limits are worth knowing.

First, mixing new and established users distorts the shape. Somebody who signed up fifteen days ago tops out at fifteen days no matter how devoted they are. Run a campaign that triples signups and the curve tilts left on its own, which looks like a product getting worse.

Second, the curve shows the shape and not the cause. That the right tail fell from 3 percent to 1.5 percent is all this chart owes you. Where it happened is a question for growth accounting, which splits the change in active users into new, resurrected, and churned, or for the magic number, the action count that separates people who stay from people who leave.

Third, products sold to teams need a second curve drawn per account. A team where one person is in every day and nine others log in monthly looks much like a team of ten spread evenly across three days a week, and the contract sits with the team.

"The power user curve turned into a smile. The last five buckets went from 3 percent to 7 percent since last quarter."

The share of near-daily users more than doubled.

"MAU is flat but the left side of the curve keeps getting thicker."

New arrivals are using the product once or twice and stopping — acquisition works, attachment doesn't.

"This product isn't one to judge by the curve. Twice a month is a fully used month here."

A judgment that usage frequency doesn't track value for this product, so the tool doesn't apply.

At Pagedeck, Ravi puts this chart on the second page of the monthly growth report. Last month the right five buckets came to 4.1 percent of active users, roughly 1,300 people, and more than half sat inside forty paying teams. Sofia moved those forty to the top of her renewal list, while Mia interviewed twenty people in the one-day bucket to find out what blocked them in their first week.

The most common mistake is reading a low right tail as a broken product. Once you learn the smile curve as the success story, every other shape looks like failure. What actually sets the shape first is how often the problem the product solves comes up. A service people use when they move house will never lift on the right, however well it is built. Left uncorrected, this reading costs a quarter spent inventing return visits instead of improving the one visit that matters.

Setting the curve itself as a target has the same root. Active days climb quickly if you send three notifications a day and attach a reward to logging in every morning. Those visits give the user nothing, so retention — how far people keep coming back on their own — stays flat or slips while the curve improves. The curve is a mirror held up to a result, and polishing a mirror doesn't change the face in it.