SUPERSEDED (Sep 2026): offer wording on this page is retired. The current pair at /onboarding/ wins over anything here.
team.stpierre.ai / internal

Marketing Pipeline — stpierre.ai. How it runs today by hand, and what the agent will run.

Marketing Pipeline

Primary KPI: Cost Per Qualified Booked Call
What this is

Turn ad ideas into tested winners. Script → brief → build → test → sort into winners and losers.

Master pipeline

Builder build Sales money-lever
Collins builds and runs the ad testing. The spend lever sits in Sales and is Akash's.

KPI

Cost Per Qualified Booked Call. What one good booked call costs us in ad spend.

What Collins does here

Read this two ways: how it runs by hand today, and what it becomes when the agent runs it. Then close the gap.

Read time

About 4 minutes.

Read it two ways

Left: how the pipeline runs today, by hand. Right: what it looks like when the agent owns it.

How it runs today LIVE

Human-run SOP — the steps Collins does now
  • Write the hook and script the idea. Every hook must name the niche — foundation and waterproofing. Add the caption and format into a short creative brief.
  • Register every hook in the hook registry (sp_creative_angles, our brain table of hooks) before building any ads.
  • Build a pack in Ads Manager, then stage it PAUSED. Read rolling 3-day results each day and sort ads into winners or losers.
  • Losers get auto-logged with a loss reason. Winners get scaled or copied, and the angle is written down.
The real structure today
  • Campaign = offer. Ad set = hook. Ad = creative × caption.
  • R2 packs are 3×3×3: 3 hooks × 3 formats × 3 captions = 27 ads per pack, in one CBO campaign (CBO = Meta sets the budget across the pack).
  • Pack 01 is live at $100/day.

Agent-run end state BUILDING

What Collins is building toward
  • A media-buying agent composes packs on its own. It pulls the best hook, format, and caption from the last 3 days, builds the next pack, and kills weak ads — all inside set caps.
  • The VSL runs as a feed ad, not a standalone page. (VSL = the video sales letter.)
  • Hooks still get registered in sp_creative_angles first.
The verdict engine that already exists
  • ads-autopilot is live (decide/decision_engine.py). It reads results and picks the per-axis winners each night.
  • So the reads are built. The part that acts on them is not.

The gap Collins builds BUILDING

Who owns what

Collins builds & runs
The production and testing agent. He composes packs, stages them PAUSED, reads results, and hands Akash a graduate/kill list. He never fires a spend change.
Akash owns the money
Ad spend and the budget click. Flipping an ad ACTIVE and setting CBO budgets is Akash-only. It is never automated and never handed off.
The budget lever is Akash's alone. The agent can propose. Akash makes the click. This never changes.
Frozen rule — do not touch. The winning ads live in the FROZEN Fresh-9 campaign (packs 001–008). It is byte-for-byte untouchable. New packs stage into the CBO test campaign only, and always PAUSED. "A new pack a day" means a new pack in the test campaign — never an edit to Fresh-9.

The stage map (visual reference)

The older visual board. Kept for reference. Read the two-way view above for how the pipeline works now.

Read the map with these current facts
Media buying is now campaign = offer, ad set = hook, ad = creative × caption. The R2 rebuild runs 3×3×3 factorial packs — 27 ads per pack in a CBO campaign, built from trailing-3-day per-axis winners. Reads are rolling 3-day. There are no automatic 3-day shutdowns.

The fixed dollar kill rules on the map below ($50 / $500 / $1,000) are LEGACY — EyeFly-era, not the current process. Current rule: a data floor before any kill (at least $10 spend, 500 views, and 48 hours live), plus hard stops at $300 with no action or $500 with no booked call. No automatic 3-day shutdowns.
Approved Creative from Creative Production Pipeline
  • Script the idea — hook, body, CTA outline
  • Register the hook, then build the creative brief (offer, hook, format, caption)
Ad Ideas Collins Brief within 48h
  • Compose one 27-ad factorial pack; stage it PAUSED in the CBO test campaign
  • Never touch the FROZEN Fresh-9 campaign
Ready To Test Collins stages Akash flips ACTIVE
  • Current: data floor before any kill — $10 spend, 500 views, 48h live
  • Current hard stops: $300 no action, or $500 no booked call
  • Legacy Kill at $50 / <5 clicks; kill at $500 / 0 calls; winner at $1,000 / 4+ calls
In Progress Collins reads Review daily
Losers
  • Auto-logged — loss reason: hook, offer, audience, or format
  • Graduate the winner into the scale campaign as a new pack
  • Write down the winning hook, format, and caption for the brief library
Winners Collins graduates $CPQBC ✓
Scale + New Split Test ↓
Active stage
Winner stage
Terminal branch
Trigger / input