Viewer Journey: see which episode loses your audience

A short drama creator posting to Instagram or YouTube can find out how one clip did. They cannot find out whether the people who finished episode 3 went on to episode 4, because no social platform knows that episode 3 and episode 4 are the same story. That question is the whole craft of serialised drama, and until now nobody was answering it for the person writing the thing.

01Three charts, one series

• ›How far viewers get. Of the people who started episode 1, the share who reached each later episode. The paywall is drawn on the same axis, so a drop at the wall is never mistaken for a drop in the writing.
• ›Did they finish it, and did they go on. Two lines per episode: the share of starters who finished, and the share of finishers who started the next one. High finishing with low continuing is an ending problem, not an episode problem.
• ›After finishing an episode. Everyone who finished, split by what they did next — the next episode within ten minutes, came back later, stopped at the paywall, or did not continue.

02The definitions are fixed, and they are conservative

An episode counts as started when the picture actually began moving, not when the page opened. That distinction matters: a video that never started playing is not an audience who walked out, and it should never be counted as one against the writing. Finishing means the player reported at least 95% watched. People are counted once per episode, with robots left out.

03Your series, compared with nothing

There is no catalogue median on this page, no percentile, no rank, no platform average. A benchmark answers a question the creator did not ask, using other people's work, and the number it produces mostly describes how much reach a series had rather than whether the writing holds. Your series is measured against itself, episode by episode.

04A worked example, because most series start small

A page of charts is useless on the day you publish, when nobody has watched yet. So every creator can open an annotated example: a made-up sixty-episode series, free to episode 10 and paid after, with the shape a paid short drama usually has — a hard fall from episode 1 to 2, a flat run through the free episodes, a cliff at the paywall, and a slow leak through the paid half. Each chart carries notes on what a shape means and what to try. The numbers in it are invented, labelled as invented, and are not a target to match.

The example ends with the arithmetic nobody does out loud: every unlock on that curve at the episode price a creator sets, less the card fee, times the creator's 80% share. It is the same calculation as the estimator on our creator page, so the two never disagree. It is arithmetic on an invented curve, not a forecast.

05What it is built on

Playback events, and nothing else. We considered having a language model read each episode on upload and score its cliffhangers, so the page would say something on day one. We cut it. An unvalidated score expressed with confidence is worse than an empty chart, and viewer events are ground truth. The model work can come back once there is enough measured behaviour to check it against.