A free trial gives away a service that costs money to deliver. Whether that makes sense is decided by a model of what happens after the trial ends.
The trial is priced as acquisition spend
Delivering a trial has a real cost in servers, licensing, support and payment processing, and that cost is counted alongside marketing as the price of acquiring a customer.
The comparison is against expected lifetime value, meaning the margin a converted subscriber generates across the time they remain subscribed.
If lifetime value exceeds acquisition cost with an acceptable payback period, the trial is expanded, and if it does not, the offer is shortened or withdrawn.
Trial length is tuned rather than chosen
Longer trials increase the chance a user reaches the moment where the service becomes habitual, but they also increase cost and delay revenue.
Companies test lengths against conversion and against retention after conversion, since a trial that converts poorly-fitting users produces early cancellations.
Requiring payment details at sign-up raises conversion sharply while lowering trial starts, which is a trade-off each company resolves differently.
Cancellation timing follows the reminder
Most cancellations cluster immediately after a renewal reminder or immediately after the first charge appears on a statement.
Where reminders are required by law or by platform rules, that clustering is more pronounced, which is exactly the effect the requirement intends.
Companies respond by concentrating onboarding effort in the early days of a trial, since a user who has not engaged by then rarely converts and rarely stays.
Cohort analysis drives the decisions
Subscribers are grouped by the month they joined and tracked over time, which shows whether retention is improving independently of how many joined.
Aggregate subscriber counts hide this, because rapid acquisition can mask a worsening retention curve for several quarters.
Cohort curves also reveal which acquisition channels produce durable subscribers, and budget shifts toward those channels regardless of their headline cost per signup.
Winback offers are a separate calculation
Lapsed subscribers are cheaper to reacquire than new ones, since they have already used the product and need no introduction to it.
Discounted returns are therefore offered selectively, based on how long a former subscriber stayed and how they used the service before leaving.
That selectivity is why two people cancelling the same service receive noticeably different offers, and why some receive none at all.