campaign studio
brief to voice-locked copy variants to batch plan to calendar to the numbers pass. the whole campaign as data your scripts can check — and an approval gate nothing skips. built for the person who owns the send button.
The campaign is the unit of work
You don't ship "content." You ship campaigns: a brief, a wave of copy variants, a batch of creative, a dated calendar, and — two weeks later — a results report someone upstairs actually reads. AI can now draft every one of those artifacts in minutes. Which means the job quietly changed: you're no longer the person who writes the first draft. You're the person who decides what's allowed to leave the building.
This chapter runs one campaign end to end — the fall relaunch for a coffee-subscription brand called Fieldnote — and turns every stage into data your own scripts can check. Voice pack as a dict, copy variants as records, calendar as rows with an approval column, report numbers traced to the export cell by cell. The spine of the whole thing is one rule: nothing publishes without a named approver, and nothing with an invented number gets an approver.
Why it matters
- The volume already moved. The CMO Survey's Spring 2026 wave has AI powering 24.2% of marketing activities, up from 13.1% in 2024, with respondents projecting 55.9% within three years. The drafting is happening. The checking mostly isn't — Gartner's 2026 CMO Spend Survey found only 30% of CMOs say they're ready to scale AI capabilities, even while allocating 15.3% of budget to it.
- The workflow is the product, not the model. The consistent practitioner finding across the 2025-26 case literature: teams that win structure the workflow around the AI — briefs, voice packs, review gates — and teams that fail bought the same tools without the structure.
- Some failure modes now carry a price tag. The FTC's Consumer Reviews and Testimonials Rule (16 CFR Part 465, effective October 2024) bans reviews "by someone who does not exist" — AI-invented testimonials included — at up to $51,744 per knowing violation (the 2024 inflation-adjusted figure; it ratchets annually). That's not an ethics-module hypothetical. That's a line item.
What AI specifically gets wrong
- It writes off-voice by default. Paste your brand guidelines into a prompt and you get compliance theater: the model nods at "warm but confident" and then ships "game-changing" in the first line. Mood words aren't checkable. A voice pack made of testable rules and a banned-phrase list is — by a twelve-line script.
- It invents numbers under social pressure. Ask for a results summary and the model will round 3.4% up to 3.6% because 3.6% makes a better sentence. Every figure in an AI-drafted report is either traced to a cell in the export or it's fiction wearing a percent sign.
- It treats generation and publication as one motion. The model drafts a post and the scheduling tool will happily take it. Nothing in that pipeline asks "who approved this?" unless you build the gate yourself — which is exactly what lesson 2 does.
- It confuses denominators with a straight face. "75% of marketers use AI" and "AI is used in 24.2% of marketing activities" are both real 2026 figures measuring completely different things. A report that swaps them isn't lying, exactly. It's worse: it's confidently meaningless.
What you'll be able to do at the end
By the end of this chapter you'll have:
- Encoded a voice pack as data — testable rules, banned phrases, owner, version — and linted a batch of AI copy variants against it automatically.
- Turned a creative brief into a batch plan with slots, counts, and cost, so "make some assets" becomes a concrete variant count — which is really a review-time budget, because generation itself costs pocket change.
- Built the calendar gate: rows only reach the scheduler with
status: approvedAND a named approver, and rejected rows stay in the file as the record of what you said no to. - Run the numbers pass: every claimed metric in the campaign report traced to the export, mismatches flagged, and the whole campaign held from publish until both checks come back clean.
The difference between "I use AI for marketing" and "I run an AI-drafted campaign my name can safely sit under" is the set of checks in this chapter. Drafting is cheap now. Judgment, encoded so it runs every time — that's the part they can't download.