Cancellations on streaming services regularly surprise viewers, including cases where a show clearly had a large audience. The decision is not made on audience size, and rarely on quality either.
The relevant figure is cost per viewing hour
A service compares what a season cost against how many hours of viewing it generated. Two shows with identical audiences can differ enormously on that measure.
A cheap show watched moderately can outperform an expensive one watched widely, because the denominator is what the service actually paid.
This is why budget growth is dangerous for a returning series. Audience must rise in step with cost, and audiences for continuing shows usually do the opposite.
Retention matters more than viewing
Services care most about whether a title stops someone cancelling their subscription. A show watched by people who would have stayed anyway generates viewing without generating value.
The valuable titles are those that bring subscribers in, or that keep a household paying through a month it would otherwise have skipped.
Because that contribution is measured against individual account behaviour, a service can see something no external observer can, which is why outside predictions are so often wrong.
Completion patterns shape the verdict
A season watched to the end by most of the people who started it signals a title that delivers on what it promised. Heavy drop-off after the first episodes signals the opposite.
Front-loaded viewing that collapses is common for shows with strong marketing and weak retention, and it looks like success in the first week's figures.
Services therefore wait some weeks before deciding, which is why renewal announcements arrive long after a season has finished airing.
Cost structure changes after the second season
Cast contracts commonly escalate, and production becomes more expensive as a show grows in ambition. The same audience becomes progressively less profitable each year.
Meanwhile the audience for a returning season is capped by the previous one, since almost nobody starts a series at its third year.
New shows have no such ceiling, which biases commissioning toward replacement rather than continuation whenever the marginal cost of a renewal rises.
Why the reasoning is never published
Viewing figures are competitive information, used in negotiations with producers, talent and advertisers. Publishing them would weaken the service's position in all three.
The consequence is that cancellation looks arbitrary from outside, and campaigns to reverse it are aimed at a decision made on data the campaigners cannot see.
Occasionally a show is rescued by another service, which is a straightforward transaction: the second buyer values the same title differently under its own cost structure.