06 / 07 · Organic social
A posting cadence that learns.
The social agent runs your company page as a series of experiments rather than a content calendar. It drafts posts from a playbook, reads the settled analytics a week later, verdicts each experiment with rules built for small numbers, and rewrites the playbook so the next cycle starts from what actually worked. Most of the time it is exploiting what it knows; a set share of posts are deliberate explorations.
What it reads
What it analyses
- Which formats, hooks and topics earn reach and which earn the engagement that matters, judged only after posts have settled.
- Whether a result is a real difference or the channel's own trend that week.
- Where the first-post boost and one-off spikes are flattering the numbers, and excluding them.
- What the playbook currently believes, and which belief the next experiment should test.
How it decides
One open experiment at a time. Verdicts use paired comparison and medians rather than averages, need a minimum number of settled posts per arm, are stratified so like is compared with like, control for the channel's trend, and are called in directional tiers rather than with p-values that a small page cannot support. Comments and reposts never decide a verdict. Confirmed learnings become the playbook's defaults; a fixed share of posts keeps exploring.
Example experiments
The shape of a test, written before it ships.
Illustrative, not a promise. Every experiment on your account gets this record before anything changes: the hypothesis, what the verdict is read from, what stays fixed, how long until a verdict, and where a person decides.
Hypothesis 01
Posts that lead with a specific observation from your own work outperform posts that lead with sector news on engagement from the audience you sell to.
- Measured on
- Median engagement per post per arm, from settled analytics, stratified by topic.
- Held constant
- Posting days, times and volume.
- Verdict after
- Until both arms have the minimum settled posts, typically four to six weeks.
- Human gate
- You approve the voice brief; every post is pasted by a person until posting is automated.
Hypothesis 02
Shorter posts with one point earn more follows than longer posts covering several, even if they earn fewer reactions.
- Measured on
- New follows attributed to each arm, alongside engagement so the trade-off is visible.
- Held constant
- Topics and cadence.
- Verdict after
- Six to eight weeks.
- Human gate
- Verdict reviewed before the playbook default changes.
Hypothesis 03
Commenting from the company page on posts from named accounts your buyers follow earns profile visits and follows that posting alone does not.
- Measured on
- Profile visits and follows in weeks with the commenting routine against weeks without.
- Held constant
- Posting cadence and content mix.
- Verdict after
- Six weeks.
- Human gate
- Watchlist of accounts and comment voice approved by you.
Where people decide
The voice brief and the topic boundaries are yours.
Drafts are handed over ready to post; the agent does not post on its own until you switch that on.
Playbook changes are reported, so you can see what the page now believes.
What you get
- 01Ready-to-post drafts with who-to-tag, on a cadence.
- 02A playbook that rewrites itself from verdicts, not opinions.
- 03Verdicts that hold at small numbers, with the reasoning shown.
- 04A monthly read on reach, engagement and follows, and what changed the playbook.