A modern instructor dashboard can show you forty metrics before breakfast, and the effect is usually paralysis: everything is measured, nothing is decisive. The discipline that separates operators who improve from operators who admire charts is ruthless metric selection. Five numbers carry nearly all the actionable signal in an online course — per-student watch time, engagement score, completion rate, assessment trend, and cohort velocity — and each one points at a different lever: content quality, interaction design, curriculum length, teaching clarity, and pacing respectively. Filter them by course, date range, or student group, and you have the entire operating system of a data-informed academy in one weekly ritual.
Number one: median watch time, not average #
Average watch time is a vanity statistic — one completionist student drags it up, one abandoned session drags it down, and the number drifts meaninglessly between cohorts. The median tells you what the typical student actually endures, and its per-lecture distribution is your content quality map: a lecture whose median watch time sits at 45% of its duration is a lecture students are skipping through, and no metric in your dashboard is more honest about which video to re-record. Watch the median weekly; when it moves two points, something in the course changed — find out what before the reviews tell you.
The other four, and the lever each one owns #
- Engagement score (quizzes answered, cards unlocked, notes taken): owns interaction design — falling scores mean your checkpoints are mis-placed, not that students got lazy.
- Completion rate by module: owns curriculum length — the module where completion craters is either too long or prerequisite-starved.
- Assessment trend across attempts: owns teaching clarity — flat scores after retries mean the explanation, not the effort, is the bottleneck.
- Cohort velocity (time from enrollment to module completion): owns pacing — wide variance means your course serves no default speed, and self-paced structure needs clearer milestones.
The reporting ritual that works is fifteen minutes, same day every week: export the five numbers per course, compare to last week, and pick exactly one intervention. One lecture to re-record, one quiz question to fix, one module to split. Analytics programs fail not from missing data but from undisciplined responses — teams that act on five findings ship none of them, while instructors who act on one finding weekly compound forty improvements a year. The dashboard's job is to make the signal unavoidable; your job is to keep the response singular.