Configured in an afternoon.
Yours from then on.
If you can write a brief, you can drive AgentMark — no agent-building, no workflow config, no dev tickets. Here's exactly what setup means.
Connect, read-only
Meta, Google, TikTok, Microsoft — OAuth in under an hour. AgentMark can see your accounts, never touch them: no spend changes, no campaign edits, ever. The first read runs overnight.
- OAuth-2, revocable anytime
- No spend access by design
- First read overnight
Set the bar — your KPI, per brand
Every brand grades against its own primary KPI and target: ROAS 3.0× for one, cost per new customer for another. Flip the KPI and the whole library re-scores. Grades are creative-level, over 30 days, behind a 10% spend floor, with low samples flagged.
- Per-brand KPI + target
- Re-score on change
- 10% spend floor, always
Teach your tags and vocabulary
Your creative language becomes the system's: define what 'founder tone' or 'problem-first hook' means to your team, and AgentMark tags every creative with it — auto-confirming only above a confidence bar, asking otherwise. Taught once, applied everywhere.
- Your taxonomy, not ours
- Confident tags auto-apply; uncertain ones ask
- Corrections stick
Applied 3 tags — drafted ad-name rules + read 12 creatives.
Bring your brief and report templates
Rounds and reports land in YOUR formats. Set the sections, the length dial, the fields your team expects — output arrives ready to edit, not ready to reformat.
- Your sections and order
- Length dial per template
- Copy templates across brands
Rounds write grounded — with receipts
Pick an approach (what's working, a rival gap, a buyer insight, a fresh angle) and the round writes: up to 100 grounded starts, each stamped with the receipt showing the winners, rival ads, and research it stands on. Your team cuts, punches up, ships.
- Grounding you can inspect
- 100 starts, you cut to the 6
- Draft visuals — inspiration only
| ▎Concept | Stage | Approach | ▎Opening hook | Creative | Portal |
|---|---|---|---|---|---|
The Question Every 40-Something Asks Her Mirror your top winner · 67 of 73 graded, 30d | ● In review ▾ | What's working | “The question every 40-something asks her mirror” | | Ship → |
What the Rival's Sale Ad Leaves Out rival sale ad · alive 6 weeks | ● In review ▾ | Competitors | “Their sale ends. Your skin doesn't.” |  | Ship → |
Clean Enough for the Most Sensitive Skin buyer question · shifted ↑ this week | ● In review ▾ | Persona insights | “Read the label out loud. We'll wait.” |  | Sent |
Six Steps, One Balm — the Switch Story top pattern · founder-story | ● In review ▾ | What's working | “Her shelf: before and after.” |  | Ship → |
Switch on the client portal
Per client, optional, white-labeled: shipped rounds and published reports land with their rationale attached. Nothing reaches a client without your approval — AI never publishes. At renewal, the year of cited decisions is already on file.
- Per-client on/off
- Full white-label
- You approve everything that ships
