How does feature frequency work?
In short, Bucket categorizes feature usage per company into these four categories:
The retention result is the average.
How does the categorization work?
Bucket will look for how often each company does the chosen event. The key metric for this is 57%. If a company has done the event more than 57% of all days, in the chosen time period, the company's frequency will be categorized as daily. If not, Bucket will try the same calculation for the following categories; weeks, then months, then years. (Years is a bit different, more on that later)
If a company has triggered an event more than 57% of all of the days in the chosen time period, the retention of that company is "Daily".
If the company has triggered the event less than 57% of all days in the chose time period, but more than 57% of all of the weeks, the company's retention is marked as "Weekly.
And so on.
We think it makes sense to mark a company as using a feature "Daily", if the company interacts with the feature 57% of all days. There's 7 days in a week, 5/7 equals 71.43%. 4/7 equals 57.14%. So, for a company to be marked as using a feature daily, they have to use it at least 4/7 days a week.
For weeks, 3/4 weeks in a given month equals 75%, so to be marked as "Weekly" the company needs to use the feature at least 3/4 weeks.
For months, 7/12 months is 58%, so to be marked as "Monthly" the company needs to use the feature at least 7 out of 12 months.
For years, it's the threadshold is lower. If a company has used a feature in a given year, but doesn't qualify as "Monthly", the company will be marked as "Yearly".
If a company has never interacted with the feature in a give period, the company will be marked as "Never".
What happens if a company signed up in the middle of the time period?
If a company was created after the start of the time period, the days between the time period start date and the company creation date will be disregarded when measuring retention.
What time period does a point on the matrix represent?
Each point represents a full calendar month of data (or how much data there is available of the current month).