digest
Short Summary
Main recommendation: Stop relying on aggregate metrics like DAUs and MAUs, and instead build a dot plot, a grid with one row per user and one column per day, with a dot in each cell where that user performed a value-creating event. The speaker argues this reveals individual usage patterns, retention problems, and feature drivers that aggregate charts hide, and that it works alongside cohort retention curves.
Key Points
- Aggregate graphs like DAUs or MAUs can keep rising even when users aren't enjoying the product, because growth masks individual behavior.
- Dot plots surfaced patterns such as weekday-only users versus weekend-only users, which can inform product redesign or targeting.
- Dot plots can reveal one-day users who never return, pointing to possible onboarding problems.
- Overlaying feature symbols, such as search or joining a public playlist, can suggest which features drive sustained engagement.
- The method scales by sampling users; Google Photos teams printed dot plots for segments like iOS users in France or high-income US web users.
- In a B2B example from a recent YC batch company, a roughly $80,000-a-year contract churned after only three of ten purchased seats ever activated, a risk visible in the dot plot.
Details Worth Noting
- Concrete instructions:
- Build a two-dimensional grid: each row is one individual user, and each column is one time period (the speaker suggests days as the default unit).
- Choose one event that represents real value in your product, such as listening to a song, sharing a photo, or processing an invoice, rather than opening the app or signing in.
- Place a dot in a user's cell for each day they perform that event.
- Optionally mark each user's first day with a ring around the dot to show onboarding.
- Encode extra state with different symbols or cell shading, such as platform (iOS, Android, web), country, or spending level.
- Sort rows by attributes, such as iOS users only or users whose first use was a given Monday, to spot patterns.
- Use different symbols for different features (for example, S for search, P for joining a public playlist) to test which features correlate with consecutive days of use.
- For large user bases, sample users by segment and review printed plots with team members.
- Use dot plots alongside cohort retention curves for context on how retained cohorts actually behave.
- The speaker says a modern AI coding tool can build a basic dot plot in about 10 minutes, since it only requires parsing logs into a 2D grid with no complex computation.
- Common mistakes: charting easy events like "opened the app," and using weekly rather than daily time periods, which hides what is happening.
- The speaker suggests that until a product has hundreds of users, the dot plot could be the only dashboard needed.
- GitHub's contribution graph is cited as a familiar example of the same visual format, with days wrapped by week.
Bottom Line
Dot plots turn raw usage logs into a visual that exposes individual behavior, feature impact, and churn risk that aggregate metrics conceal, so they should be used with cohort curves to guide what to fix and build.
