Experiments and the optimiser
An experiment tries variations of one setting across a pipeline's posts, such as short against medium, to see which does better. The optimiser reads the results and suggests making the winner the setting.
Neither adds a post. An experiment only changes a setting of a post that was going to be made anyway, or a time inside what the outbox already allows.
What it takes
- An outbox set to Posts proposed at a time, for approval. A pipeline that only drafts runs as set.
- A step with an experiment on it.
- The optimiser set to anything but Off, which takes somebody who may publish the brand's pipelines.
- Posts with numbers. Only measured posts count, so connect accounts that report them. See Numbers.
Setting one up
- In the editor, find the step you want to vary. Steps that can vary show Experiment.
- Under Try variations of, choose the setting. Nothing: it runs as set turns it off.
- Tick at least two variations, today's setting among them, and at most six. Fewer variations reach an answer sooner, as posts are shared out among them.
- Save. Then on the pipeline's page, in Experiments and the optimiser, set What it does.
A step tries one setting at a time.
| Step | Try variations of | Variations |
|---|---|---|
| Write the post | Length | Short, Medium, Long |
| Write the post | How it opens | Whatever suits it, A question, A number, A bold claim, A quote, Plainly |
| Write the post | Emoji | With emoji, Without emoji |
| Add hashtags | At most | One to five hashtags at most |
| Cut a clip | Captions | With captions, Without captions |
| Cut a clip | Vertical version | The scene the stream marked, else the middle; The middle of the picture |
| Outbox | Time of day | Three-hour blocks, such as 18:00 to 21:00, inside the outbox's own timing |
What a post says stays yours: the brand's voice, the language and your Instructions are never tried. GreenMind can narrow what experiments may try, and the save says so if one is not allowed.
The optimiser
What it does has three settings:
| Setting | What happens |
|---|---|
| Off | Every step runs as set. |
| Try variations and suggest changes | Each post gets one of the variations its steps try, more often the ones doing better. When one does clearly better than the setting, it is suggested for somebody to accept. |
| Try variations and change settings by itself | As with suggestions, but a variation that does clearly better is made the setting by itself, within the bounds set for experiments. Each change is listed with one button to undo it, and is undone by itself if posts do worse after it. |
The third can only be chosen once somebody has accepted one of its suggestions on this pipeline. If the person who set it may no longer publish, the optimiser goes back to suggesting.
It only ever changes the one setting being tried, and never what the outbox does with a post, its budget or its channels.
Reading a suggestion
Results are read once a day. When a variation has done clearly better, the people looking after the pipeline are told, and the change appears in Experiments and the optimiser with its evidence: about how many times the old setting it did, the chance it really is better, and how many measured posts that comes from.
Somebody who may publish answers it:
- Accept makes it the setting, as a new version of the pipeline, which then runs as you.
- Let it go leaves the setting as it is.
- Undo puts back a change that was made. If the step was edited since, it says so, and you change it in the editor instead.
| Status | Meaning |
|---|---|
| Suggested | Waiting for somebody to accept it |
| In use | It is the setting now |
| Let go | A person did not accept it |
| Undone | It was put back |
| Set aside | The step was edited in the meantime, or the bounds moved, so the suggestion no longer applies |
After a change, it leaves that setting alone for a while.
The watch after a change
After a change is made, the next posts are compared with how the old setting did. The change says where this stands:
- "Watching the next posts against the old setting; it is undone by itself if they do worse."
- "The posts after it did not do worse than the old setting."
- "Undone by itself: the posts after it did worse than the old setting."
- "No longer watched: the step was changed by hand since."
If a change is undone by itself, the people looking after the pipeline are told.
See also
- What's working shows which times, lengths and first lines go with better results across the brand.
- Three lengths is a different thing: it makes three posts from one item, rather than trying one length on each.