Skip to content
CommandChain CommandChain CommandChain Docs

The marketing loop

Most marketing tools execute what you tell them. CommandChain runs a loop, and the loop starts inside the work you are already doing.

Six stages, one circle

Section titled “Six stages, one circle”
  1. Product truth. Your shipped tasks, meeting summaries, project intent, and connected repositories already live here. They are assembled into the list of facts an agent may claim. See product truth.
  2. Strategy. Who the thing is for and how it is talked about, in two layers: your company profile and the project’s marketing context.
  3. Content. Agents draft content assets against that truth and that strategy. Every draft is a real asset you can read and edit, never chat text.
  4. Publish. A publish runs the brand-safety lint, then the autonomy gate. At the default level it becomes an approval carrying the exact payload that will be sent.
  5. Measure. Publishing and launching produce metrics with no integration at all. Engagement numbers and self-reported sources join them when you have somewhere to read them from.
  6. Learn. A number that lands well off its own trailing baseline becomes evidence. A periodic review turns evidence into proposals to change the strategy. You accept them, and the next draft is written with that change in it.

Stage 6 feeds stage 2, and stage 2 feeds stage 3. That is the loop.

Three rules that never bend

Section titled “Three rules that never bend”

Agents propose, you approve. No agent publishes, spends, or rewrites your strategy on its own. An approval shows the literal rendered payload before it goes anywhere, and it is that payload that executes, not a re-derived one.

Autonomy is earned. Every externally visible action class starts at “ask me each time”. Levels rise only when a track record you can inspect justifies it, and only when you say yes. An incident takes a level back automatically.

Ad spend is locked. The ad.spend action class is hard-locked at L1 in code rather than in configuration. Agents may plan a paid campaign and draft ad creative as assets, and no money-moving call executes in this release, whatever the settings say.

What the loop needs from you

Section titled “What the loop needs from you”

Two things make the rest work, and both are worth doing before you ask for a draft.

  • Fill in the constraints half of your company profile. Voice, banned phrases, and claims to avoid are what the brand-safety lint enforces. Without them the lint has nothing to check against.
  • Connect somewhere to publish. A publish runs through an active, human-verified integration binding. Until one exists, publish tools do not appear to agents at all, and the publish button on an asset returns an error rather than pretending.

Everything else works from the first day: drafting, scheduling, atomizing, the calendar, product truth, and the activity metrics.

Next steps

Section titled “Next steps”