Learnings

A learning is one accepted sentence about what works for your audience: “concrete demos outperform feature lists”, “Tuesday mornings get the highest engagement”. Individually they are small. Accumulated, they are the reason the twentieth campaign starts smarter than the first, because every new draft is written with them in hand.

Where learnings come from

Two producers, one bar for entry:

Concluded experiments. The verdict is drafted into a sentence that pairs what was predicted with what was measured, citing the arm numbers as evidence.
The weekly review. A pattern in the evidence that never went through an experiment can still be proposed as a learning, the same way it proposes any other marketing change.

Both arrive as proposals. Nothing becomes a learning until a person accepts it. An experiment-backed learning carries its verdict and a link to the experiment; a review-derived one carries the evidence it cited. The list shows which is which, so you always know how much weight a sentence has behind it.

Where to find them

Per project: the learnings list on the project’s marketing context page, and the learnings view on the marketing home.
Across the workspace: the Learnings tab on your company profile, which lists every project’s learnings in one place. Filter by text, by verdict, by project, and toggle retired learnings into view when you need the history.

A learning is recorded either against the project that earned it or as tenant-wide. Tenant-wide is the one that travels: every project’s drafts receive it alongside that project’s own. A learning recorded against a project belongs to that project, and the tab above is where you read across them.

Each learning links to its cited evidence and, when it has one, its source experiment. A learning that cites nothing displays exactly that, which tells you what it is worth.

How agents use them

This is the point of the whole exercise, and it happens in two ways, automatically:

Every marketing draft carries them. The grounding that agents draft against includes the project’s own active learnings and every tenant-wide one, alongside product truth and strategy. Accept “fear-based hooks underperform for this audience” today, and tomorrow’s draft is written by an agent that was told that.
Agents can ask. Knowledge search covers learnings, phrased however the question comes: an agent asking “what have we learned about marketing to solo founders?” before a campaign gets back the accepted learnings that bear on it, not a keyword match. It searches the project it is working in, and reaches another project’s learnings only when it is asked to, by name or across all of them. What it finds outside the current project is labelled with that project’s name, so a lesson from one product is never quoted as this one’s.

When a learning turns out to be wrong

Beliefs age. When a new learning contradicts an active one, the draft names the learning it supersedes, and accepting it does three things: the old learning is marked retired, it drops out of what agents recall and draft against, and it stays visible in the browser with a link forward to what replaced it.

Retirement is the honest alternative to deletion. The record that you once believed something, and what changed your mind, is itself organizational knowledge.

Next steps

Run marketing experiments: the strongest way to earn one.
Review marketing proposals: the gate every learning passes through.
The marketing loop: where learning sits in the larger cycle.