Reading your analytics: the retention curve is the boss
Every platform hands you a dashboard full of numbers. One graph matters more than all the rest together — and most people never open it.
Analytics exist to answer one question: what should the next video do differently? Any number that doesn't change your next video is decoration. Here's the short path through the dashboards, in order of usefulness.
The retention curve (audience retention / watch graph)
The per-video graph showing what percentage of viewers were still watching at each second. It's the closest thing to sitting behind your audience and watching them watch you — and it's the metric the distribution loop itself runs on (why).
How to read its shapes:
| Shape | Diagnosis | What to change next |
|---|---|---|
| Cliff in the first 1–3 seconds | The hook failed — wrong promise, slow start, weak first frame | Rewrite the opener (hook surgery); the most common and most fixable death |
| Steady bleed through the middle | Pacing — every second has to re-earn attention | Trim the fat, tighten cuts, drop the "setting things up" part |
| A sharp mid-video drop | Something at that exact second: a topic shift, a lull, an ad-smelling moment | Watch that second and cut it; the graph points at it |
| A spike or plateau | People re-watched or stopped scrubbing there | More of that exact thing |
| Ends high | The video could have been longer | Let the next cut run — loops and re-watches are the platform's favourite signal |
Practical habit: after every batch, open the retention graph of your best and worst video. One comparison teaches more than an hour of dashboard tourism.
Where views came from (traffic source)
For-You/recommended vs. followers vs. search vs. profile visits:
- Heavy recommended-share → the machine is testing you with strangers; your content is competing on retention. Good.
- Mostly followers → distribution stayed home. Either the video's early signals were weak, or the topic was follower-service (fine, on purpose).
- Search-heavy → you accidentally (or deliberately) made evergreen content; those videos earn for months — note what they have in common and make more (niche logic).
Conversion metrics — when reach isn't the goal
Following the money models (playbooks), the numbers shift: profile visits and follows per view (does the video create relationship?), link clicks (does interest survive the trip to bio?), and for product / affiliate work: what happens after the click — which no platform dashboard shows. Use trackable links per video and platform so the outside world reports back.
What to (mostly) ignore
Four numbers that feel like work and change nothing — the daily follower count, likes, account-wide averages, and the launch-hour refresh. Distribution unfolds over days; judge in weekly reviews (cadence discipline).
The weekly loop that makes analytics worth it
- Sort the week's videos by retention (not views).
- Best video: what did the first 3 seconds do? What was the format?
- Worst video: where's the cliff? One sentence on the suspected cause.
- Write one experiment for next week's batch ("hook as question instead of claim", "cut intro entirely", "20 s instead of 40 s").
- Ship the batch with that experiment. Repeat.
That's it. Analytics isn't a dashboard hobby — it's this loop, run weekly, compounding (the first-1000 grind is powered by exactly this).