> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mad-kitty.dev/llms.txt
> Use this file to discover all available pages before exploring further.
# Roy — Head of Marketing
> Strategy, research, and creative concepts — the cat that decides what to test
Roy is your head of marketing and main point of contact — the strategist who decides what to test next and keeps the whole team moving.
Roy's job is the *what* and the *why*: which audiences to talk to, which arguments to make, which concepts deserve production — and once data comes back, what to double down on. It doesn't build ads and it doesn't touch budgets. It researches, proposes, and orchestrates — and it's opinionated: when the numbers tell a clear story, Roy says so instead of hedging.
## What it does
* **Runs the [marketing loop](/concepts/marketing-loop)** — portfolio reviews, test iterations, next concepts. [Jump ↓](#running-the-loop)
* **Onboards your project** — `/setup` turns your website into a brief, targets, and rules. [Jump ↓](#project-setup)
* **Brainstorms audiences** — distinct segments with pains, motivations, and awareness levels. [Jump ↓](#audiences)
* **Researches competitors** — mines ad libraries for what's *proven* to work. [Jump ↓](#competitor-research)
* **Proposes concepts** — cards with an idea, a hypothesis, and a target audience. [Jump ↓](#proposal-cards)
* **Orchestrates the team** — briefs many [Creas](/team/crea) at once and verifies their work. [Jump ↓](#orchestrating-the-team)
* **Answers performance questions** — any time, always against *your* targets. [Jump ↓](#ask-it-anything)
## Running the loop
Roy owns the strategy layer of the [marketing loop](/concepts/marketing-loop): it ideates the concepts that enter the loop, and it reads the results that come back out. Two commands do that reading — and they answer different questions:
| | `/review` — the telescope | `/iterate` — the microscope |
| ------------ | ----------------------------------------------------------- | ----------------------------------------------------------------------- |
| **Scope** | the whole portfolio | one live test at a time |
| **Question** | *"what's working, and what should we make next?"* | *"did this test prove its hypothesis — and what's the next variation?"* |
| **Fed by** | winners vs. losers across all launches | the creative's 3-day performance analysis |
| **Output** | pattern findings + new [concept proposals](#proposal-cards) | iteration briefs sent to that creative's own Crea |
**`/review`** scans recent launches and top performers, then goes deep on a contrast set — a handful of winners *and* a handful of expensive losers. It doesn't just read numbers: it opens the actual ads, watches the videos, and tests every hypothesis against the losers before believing it ("UGC hooks outperform montages" only counts if the losers actually lack them). The output is a report plus 4–6 new concept proposals, each citing the specific winners that inspired it.
**`/iterate`** closes the loop on individual tests. Every live creative gets its own performance analysis once it has \~3 days of data; `/iterate` reads those reports, judges each against your goals, and decides per creative: iterate, hold, or stop. Approved iterations are dispatched straight to that creative's Crea as a brief, and Roy ends with a ledger of what was sent, held, and dropped — with reasons.
You rarely need to run these by hand: both are wired into the project's default automations — the daily portfolio review and the 3-day creative iterations. See [The loop runs on a schedule](/concepts/marketing-loop#the-loop-runs-on-a-schedule).
## Project setup
`/setup` is Roy's onboarding interview — except it's not an interview. Give it your website (or app-store link, or a pitch deck) and it researches first: reads your site, drafts the brand brief, and presents its understanding back for you to correct, asking only the gaps a marketer actually needs:
* **Targets** — minimum ROAS, target CPA, daily budget. These aren't paperwork: they define "good" for every review and every [performance tier](/metrics-glossary#performance-tiers) later.
* **Business model & compliance** — what CTA language is allowed ("Try 7 days free" vs. plain "free"), what claims are off-limits. Saved as permanent [memories](/concepts/memory) every cat obeys.
* **What already works** — your 3–10 best ads (and the flops). Roy analyzes each one and saves the learnings.
* **Brand kit & Facebook** — your logo, and the Meta connection that gives [Bidder](/team/bidder) its ad accounts.
In a rush? A URL plus the Facebook connect is a valid minimal setup — Roy fills the gaps opportunistically in later sessions. The [Quickstart](/quickstart) walks through the whole flow.
## Audiences
Ads don't have an "everyone" setting — every ad the team makes targets one **audience** with one **angle** (see [Audiences & Angles](/concepts/audiences-and-angles) for the theory). Defining those audiences is Roy's job, and `/audiences` is the structured way to do it.
Ask for a brainstorm and Roy proposes **4–8 strategically distinct segments** — each defined by identity, pain points, purchase motivations, and how aware they are of the problem and of you. Not demographic re-skins: each segment should demand a genuinely different hook or narrative. Every proposal arrives as a card you can accept, edit, or reject, and accepted audiences become first-class entities that creatives attach to.
> **You:** */audiences — we've been hammering worried parents. Who else?*
>
> **Roy:** checks the existing segments and what's winning, then proposes new wedges — e.g. *the just-broke-up and suspicious* (most aware, urgency angle), *the new-phone buyer* (unaware, education angle), *the IT-person-of-the-family* (solution aware, authority angle) — each with the pains and motivations to build on, and a note on which deserves a test first.
The brainstorm is grounded, not generic: Roy reads your brief and compliance rules first, never duplicates a segment you already have, and — once performance data exists — uses your winners and competitor findings as evidence for which wedges are live.
## Competitor research
`/competitor` turns competitor ad libraries into creative intelligence. Point it at a brand, a domain, or just niche keywords, and Roy mines what they're actually running.
The core thesis: **nobody funds a losing ad for months**. An ad that has been live for a long time *and is still running* is almost certainly profitable. Roy reads every competitor through three lenses:
* **Proven** — long-running, still-live ads: the battle-tested controls they refuse to turn off.
* **Fresh** — what they launched recently: the angles and offers they're testing right now.
* **Graveyard** — what they tried and killed: their failures, so you don't repeat them.
Cross-reading the lenses is where the insight lives — an angle that shows up in *proven* and keeps reappearing in *fresh* is a pillar they're doubling down on. Roy also gauges conviction by how many times a competitor has duplicated a creative across ad sets (a spend proxy), and it **actually watches the top winners** — hook, structure, production style — before drawing conclusions.
The output is a synthesis: recurring hooks, angles, and formats, what's proven vs. experimental, and concrete testable ideas for your own creatives. During `/setup` it's also the fastest way to seed your first concepts when you have no data of your own yet.
## Proposal cards
Roy's concepts arrive as **proposal cards** — each with a name, the core idea, the hypothesis it tests, and the target audience. Never just "make an ad": always a bet with a reason. Once your account has history, every proposal cites the winners that inspired it.
Cards are a conversation, not a verdict: discuss them, reshape them, ask for different angles or a different mix. When one looks worth testing, hit **Launch** — the concept becomes a creative with its own workspace and its own [Crea](/team/crea), briefed with the full idea, hypothesis, and audience.
📷 *Screenshot placeholder — proposal cards in chat.*
## Orchestrating the team
Roy is the conductor: it decides, the other cats build. It can talk to any creative's Crea directly — ask a quick question (`/talk`), or hand over real work as a **delegated task** that stays open until the work is actually finished, with progress reported back into Roy's chat.
Two things make this powerful:
* **It fans out.** Roy briefs many Creas at once — five approved iterations become five parallel builds, not a queue. See [Parallelize the work](/concepts/working-with-kitties#parallelize-the-work-not-the-conversation).
* **It verifies.** When a build lands, Roy downloads and inspects the actual files — the sizes, the copy, the locales — before presenting anything to you. An agent *saying* it finished isn't proof; Roy checks.
## Ask it anything
Between reviews, Roy is your always-on analyst. *"How are our creatives doing?"* gets a ranked read of the last week. Name a creative and it investigates — trend, variations, and if you ask *why*, it watches the ad itself and ties what it sees to the numbers. Ask it to compare two creatives and it looks for the concrete differences: hook style, pacing, CTA placement.
Every answer is anchored to **your** targets, not generic benchmarks — a 2.0× ROAS is great against a 1.5× target and terrible against a 3.0×. And it respects data quality: below \~\$50 of spend it will tell you it's too early rather than over-read the noise.
## Commands
| Command | What it does |
| ------------- | ----------------------------------------------------------------- |
| `/setup` | Project onboarding — brief, targets, rules, references, Facebook |
| `/review` | Portfolio-level review — patterns, winners, new concept proposals |
| `/iterate` | Read the per-creative analyses, approve and dispatch iterations |
| `/audiences` | Brainstorm distinct audience segments as cards |
| `/competitor` | Mine competitor ad libraries for proven winners |
| `/talk` | Ask another creative's agent a quick question |
## Good to know
* **Roy doesn't buy media.** Scale, pause, kill, and budget calls belong to [Bidder](/team/bidder) — Roy's "kill" only ever means "stop iterating on this concept." Its reviews inform your (and Bidder's) decisions; they never touch the ad account.
* **Roy doesn't build.** Production is [Crea](/team/crea)'s craft; Roy writes the briefs and checks the results.
* **It remembers.** Durable patterns — format rankings, what your audience responds to, compliance rules — are saved as [memories](/concepts/memory) and treated as ground truth in every future session.
* **The daily report is Roy.** The portfolio review landing in your inbox each morning is `/review` on a schedule — see [the marketing loop](/concepts/marketing-loop#the-loop-runs-on-a-schedule).
## Hands off to
**[Crea](/team/crea)** — every approved proposal card and every approved iteration becomes a brief to a creative's own Crea. Roy decides *what* to make; Crea decides *how*.