Channel experiment
A channel experiment is a dated record of one change you make to a connected Telegram channel — a new posting schedule, a different content mix, a new call-to-action — logged so you can judge its effect later instead of trusting a gut feeling. You write down what you expected to happen and exactly what you changed; Autogram does not ask you to define a metric up front. Instead, it computes the subscriber and engagement movement on demand from the daily channel snapshots that already accumulate for every connected channel, comparing the days before the change against the days after. When you have seen enough of the after-window, you close the experiment with a verdict: kept, reverted, or inconclusive.
What you write down
Two free-text fields carry the whole record: a hypothesis, what you expected the change to do, and the change itself, in your own words. Autogram places no structure on either field and does not require you to name a specific metric before you start — the record is your log, and the comparison happens afterward against whatever movement the channel's snapshots actually show.
Where the before/after comes from
Every connected channel already accumulates a daily channel snapshot of its subscriber count and engagement. An experiment does not duplicate that data or freeze a number at creation time: the comparison is computed fresh each time you open the experiment, using the window before the day you logged the change against the window after. Snapshots that would make an old stored number stale keep arriving regardless, so nothing here ever needs a recalculation step — reading it again just re-derives the current answer.
Why the verdict is not a causal claim
An experiment records one thing you changed on one day. It does not — and cannot — isolate that change from everything else that also happened to the channel in the same window: a post that went unusually viral, a seasonal dip in engagement, an unrelated schedule change you made the same week. The kept/reverted/inconclusive verdict is your judgment call after looking at the correlated movement, not a statistical proof that the change caused it.
Still running versus decided
An experiment can sit in three states. While the end date is empty, it is still running and nothing has been judged. Once you set an end date, it waits for a verdict — Autogram deliberately keeps ending and deciding as separate steps, because ending an experiment on a Friday and deciding what it proved the following Monday are two different moments, and treating them as one would force a premature verdict.
Related terms
Related features
Put these concepts to work
Automate your Telegram publishing with AI content, quality scoring, and smart scheduling.