Engagement is how often somebody comes back to a product and how much they do once they arrive, and frequency is the part of that question a team can count without arguing about definitions.
Activation explains why a share of each intake keeps coming back at all. Frequency asks the next question about those people, which is whether coming back means every morning, every Monday or twice a quarter. Two products can hold the same 22 per cent of a cohort at day ninety and be completely different businesses.
The vocabulary came from mobile and social products in the early 2010s, where daily use was the goal and the daily active user count was the headline number in every funding round. Business software borrowed the measure afterwards and inherited an assumption with it, which is that more often is better. That assumption holds for a messaging app and fails for a tax filing tool.
A product manager needs frequency because pricing, support cost, sales forecasting and the roadmap all depend on it. A product somebody opens daily can carry an in product announcement, can recover from a bad release in a week and can be sold on a seat licence. A product somebody opens twice a year can do none of those things. A team that has never counted the interval will price, staff and plan the product as though it were the first kind.
The sections below define the ratio everybody quotes and work through a case where it hides the difference it was meant to show. Counting active days follows, with the distribution the ratio leaves out. Finally they cover the natural frequency of a product and what it does to every target above it.
Daily active users over monthly active users as a ratio
The ratio divides the average number of people active on a day by the number active across the month, and it is usually read as how much of the monthly audience shows up on a typical day. A figure of 20 per cent means one monthly user in five is there on an average day.
Facebook is the benchmark everybody reaches for. Its fourth quarter results for 2019 reported an average of 1.66 billion daily active people during December and 2.50 billion monthly active people as at 31 December, which is a ratio of 66 per cent. Almost no product outside social messaging comes close.
That headline figure carries a technical wrinkle worth knowing about before anybody builds a target on it. The numerator is an average across a month and the denominator is a count at one instant, so the two halves of the fraction are measured differently. The ratio is a convention with a long history and useful properties. Nobody should read it as a clean share of one population.
The definition of active matters more than the arithmetic. A product counting an app opening will report a far higher ratio than the same product counting a completed task, and neither team is being dishonest. The number belongs in the tracking plan with the qualifying event written next to it.
Two products with the same ratio and different distributions
An agreed definition makes the ratio comparable and leaves its largest weakness untouched. Take a product with fifty thousand monthly active people and an average of ten thousand active on a day across a thirty day month. The ratio is 20 per cent. Total active days across the month are ten thousand times thirty, or three hundred thousand, which divided across fifty thousand people is a mean of six active days each.
Two very different populations produce that same mean.
| Population | People | Active days each | Active days contributed |
|---|---|---|---|
| Product A, the committed minority | 5,000 | 24 | 120,000 |
| Product A, everybody else | 45,000 | 4 | 180,000 |
| Product B, one even population | 50,000 | 6 | 300,000 |
Product A holds five thousand people who open it almost every working day and forty five thousand who open it about once a week. Product B holds fifty thousand people who all open it six times a month. Both report a mean of six days and both report a ratio of 20 per cent. The two businesses behind those identical figures have almost nothing in common.
The difference shows up the moment anybody acts on the number. Product A has a core of five thousand people producing 40 per cent of all activity, so an interface change that annoys them costs a great deal and a change aimed at casual users touches nine tenths of the audience. Product B has no core at all, so there is nobody to protect and nobody to build a power feature for.
Counting active days in the period
Counting how many days each person was active in a fixed window is what recovers the distribution the ratio threw away. A twenty eight day window is the usual choice, because it holds exactly four of every weekday and removes the month length problem from the comparison.
The output is a histogram with twenty nine columns, running from nobody active on any day to somebody active on all twenty eight. Reading it takes a minute and answers questions the single ratio cannot. Product A above shows two clusters, one at four days and one at twenty four, with almost nothing in between. Product B shows one column at six days holding everybody.
Two summaries of that histogram are worth putting on a dashboard. The first is the share of monthly users active on eleven or more days, which is a defensible definition of a regular user. The second is the share active on three days or fewer, which is the population most likely to be gone next month.
The average user and the people who move the aggregate
The mean of six active days in Product A describes nobody at all. Forty five thousand of the fifty thousand people were active on four days, so the median is four, and the five thousand at twenty four days pull the mean two days above the experience of nine tenths of the population.
Usage distributions in software are skewed this way almost without exception. A small group uses a product far more than everybody else, and because every aggregate weights people by how much they do, that small group moves every aggregate on the dashboard. Total sessions, total events, average time in the product and support contacts are all dominated by the people who were there most.
This has a practical consequence for research as much as for reporting. A team recruiting interview participants at random from active users will mostly reach the casual majority, and a team recruiting from the top of a usage list will mostly reach the committed minority. Both groups are real customers and they will describe different products.
The natural frequency of a product
Every product has a rhythm set by the task it performs, and no amount of design moves it very far. A payroll tool is opened when payroll runs. If every customer runs payroll once in a thirty day month and opens the product on exactly one day, the ratio of daily to monthly active users is one thirtieth, or about 3.3 per cent, and that is the ceiling.
Judging that product against a benchmark of 20 per cent produces bad work. The team will add a daily digest, a notification and a dashboard nobody asked for, and the ratio will move a little because people opened a page and left. The measure improved and the product got worse.
The repair is to ask what the natural interval is before choosing the denominator. A weekly product is measured on weekly active users over monthly. A quarterly product is measured on whether the customer came back for the next quarter at all. Sean Ellis and Morgan Brown put this plainly in 2017, arguing that the interval has to come from the product's own use case before any target is set on it.
Every figure on this page has been an average over a population that was never uniform. The mean of six days hid two clusters, and the same problem sits under every rate in the course so far, which is the subject the next page takes apart.