Campaign Creation Agent (All Channels)

What this agent does

Turns an approved campaign brief into built campaigns across channels: structure per the Campaign Mapping, ad copy variants drafted from the brief and asset library, audiences configured, UTMs and naming exactly to standard — all staged in paused/draft state for review. The tedious, error-prone hours between "the brief is approved" and "it's ready to launch" become agent work; humans review a finished build instead of assembling one.

Why this agent

• The map already exists — Campaign Mapping v1.6 defines the macros, the 43 micro-campaigns, and the SIG-01..45 signal map with a 480-asset library. That's a specification an agent can build against; today it's executed by hand.
• Setup is where velocity dies — multi-channel builds take days of clicking and are where naming and tracking errors breed; those errors then poison measurement downstream (the exact thing Phase 04's tracking standards exist to prevent).
• Signal-based orchestration needs supply — the orchestration model maps signals to plays to campaigns; when a signal fires, the campaign response should be a build away, not a sprint away.
• One brief → many channels is the H2 promise of the agent-augmented model in its most literal form.

Trigger & inputs

• Trigger — an approved brief (from the calendar, a play, or a mapped signal); never self-initiates a campaign concept.
• Reads — the Campaign Mapping structure, the 480-asset library, the tracking standards (UTM convention + naming taxonomy), audience definitions from the signals stack, and past campaign performance for copy patterns.
• Context — the awareness vs performance split and budget guardrails per channel, so builds land in the right bucket with the right settings.

What it produces

• Built campaigns in draft/paused state — structure, ads, audiences, budgets-as-specified, across the channels the brief calls for.
• A launch checklist per build — what was built, what needs human eyes (landing page live? tracking verified? budget confirmed?), diffs from the brief flagged.
• The build log — every campaign traceable to its brief and its approver, named to taxonomy.

How it works

• Parse the brief: objective, audience, channels, budget bucket, mapped SIG/micro-campaign slot.
• Assemble structure from the Campaign Mapping; pull and adapt copy/assets from the library per channel spec.
• Apply naming + UTMs generated from the taxonomy — never hand-typed.
• Stage everything paused; run the self-checklist; route to review with the diff-from-brief noted.

Guardrails & human-in-the-loop

• Nothing goes live by itself — the agent builds in paused/draft only; a human launches, always, at every planned autonomy level.
• Never-do list — never invents offers, claims, or pricing in ad copy; never creates audiences outside the defined stack; never deviates from the naming/UTM standard (a build that can't comply stops and asks).
• Brand + claims — copy variants ride the existing brand review path before launch; benchmark or customer claims only from approved sources.
• Escalation — missing assets, ambiguous briefs, or landing pages that don't exist yet → blocked with a note, not improvised.

Success metrics

• North star — brief-to-ready-for-review turnaround: days, not weeks. Baseline: current manual setup time (TBD from the first M2 builds).
• Quality bar — review pass rate: builds approved without structural rework.
• Guardrail metric — 100% naming/UTM compliance on every build; zero self-launches, zero unapproved live spend.
• Hours returned — setup hours per campaign × campaign volume, on the program ledger.

Build plan

• V0 · Shadow mode (Oct) — rebuilds recent real campaigns from their briefs; output compared to what humans shipped. Exit: structural parity + compliance on the sample.
• V1 · Assisted (Oct–Nov) — builds the real calendar in draft; humans review and launch. Exit: pass-rate target at calendar volume.
• V2 · Scale (Nov+) — signal-triggered builds queue automatically when SIG plays fire; launch stays human. Pairs with the optimization agent watching what it launched.

Dependencies & risks

• Campaign Mapping v1.6 handoff — the map is the spec and its handoff is an open ask on the dashboard; the agent inherits whatever state it's in.
• Asset library access + hygiene — 480 assets are leverage only if findable and current.
• Tracking standards locked — Phase 04's UTM/naming convention must be ratified before V1, or compliance means compliance with a moving target.
• Landing pages are the mate — campaigns built faster than pages ship just queue; the LP system (wm-website-build) sets the pace. [owner to complete]