◆ COUNTER-PROPOSAL · APRIL 2026 ◆

JUPITER'S
AI CREATIVE
SYSTEM.

:: A PROPOSAL FOR JUPITER · ROSS + ROBBIE + LAURA ::
» MISSION BRIEF «

Five years in market and you've never had a clean answer to what about our ads worked. Meanwhile a performance creative agency keeps shipping briefs your team didn't write. The unlock isn't another vendor, it's mining Jupiter's own data and putting the system in every seat on your team. This is a 3-month engagement to do exactly that.

3-month engagement Whole-team training 5yrs of Meta data mined Slack-embedded throughout
▶ LET'S RUN IT
M1 mines the data. M2 trains the team. M3 runs without me.
CHAPTER I · WHERE YOU ARE

Five years of ad data. No system pulling learnings from it.

From our calls on April 21 and April 24. Jupiter is scaling 4–7x in the next 2–3 years, ships 60+ ads/month, spends $50–75K/mo on Meta. The blocker isn't the spend, it's that none of it teaches the next ad. Two pieces of those calls this proposal is built around:

▶ ROSS · ON FIVE YEARS LOST

"It frustrates, it kills me that for five years that we've been in business, we haven't learned much. You didn't even have to be in our organization to guess what's working, you could just look at our meta account... There's so much more that we could have understood, could have taken."

Five years of Meta data is sitting in your account untagged, unmined, and uncodified. The first thing this engagement does is fix that.

▶ ROSS · ON THE WHOLE TEAM

"I want the whole team to hear it, even the people that aren't using it. I want them to see how this is working so they don't try to block it. If it's just like 'here's a tool, you guys are over here, we're using this tool over here', they're not gonna own it. They're not gonna internalize it."

Month 2 is built around exactly this. Six team members, dedicated training time, the system installed in every seat, not just the growth chair.

TL;DR: You don't need another performance creative agency. You need someone to mine five years of Jupiter's own data, install the system that turns ad fatigue into learning, and train your whole team to run it, including the head of growth you haven't hired yet.
CHAPTER II · THE BET

Three reasons this works for a brand, not just an agency.

▶ THESIS 01

Five years of data is your unfair advantage

One perfect ad doesn't move the needle for a brand at your spend level. Five years of Meta data, mined and tagged, does. Every static, every angle, every persona, every winner, every flame-out, fed in as the floor under every new brief.

No agency you've worked with has ever done this. Most can't.

▶ THESIS 02

A system every seat can run, not a tool one person owns

You said it yourself: if it's just a tool the growth person uses while everyone else watches, the team blocks it. This installs in every seat, Ross, Robbie, Laura, the new growth lead, the designer, the community manager, so it's the way Jupiter ships ads, not a side project.

Month 2 is built around team-wide training. That's why it's three months, not one.

▶ THESIS 03

Replace the performance creative agency, not augment it

You're spending $50–75K/mo on Meta and outsourcing 60+ ads/month to a creative agency that hasn't taught your team anything about why your ads work. This system replaces that line item, keeps the volume, and leaves the institutional learning inside Jupiter, where it should have been the whole time.

The 3-month total is less than one month of agency fees. You keep what you build.

CHAPTER III · THE STACK

What gets installed inside Jupiter.

Three tranches. Historical mining lands first because every downstream decision is sharper once we know what worked. Production and the whole-team install layer on next. Optimization, iteration on winners, and handoff close out the engagement.

M1

Foundation

▸ INSTALLS WEEKS 1–2
  • Meta API ingest of all 5 years of Jupiter ads, every static auto-tagged by persona, angle, emotion, asset type, format
  • Historical winner/loser analysis, why the PCOS-Target ad worked, why before-and-afters die off, what patterns repeat across five years
  • Adlib gap analysis on Jupiter vs. the anti-dandruff and scalp-care category
  • Jupiter brand bible + spec cards, generated from your site, your winning ads, and your tone
  • The 50-universal-personas framework, calibrated to scalp-care buyers
M1

M2

Production + team install

▸ INSTALLS WEEKS 3–6
  • Brief-generation pipeline, 10 copy agents iterate each brief until it scores 90+
  • Brief-to-prompt converter, turns approved briefs into the structured prompt format GPT-Image and Nano Banana actually want
  • AI static generation, fal.ai integration, variant generation, folder organization, cost tracking, all from inside Claude Code
  • 15 asset-type markdown files (before/after, founder story, social proof, education, comparison, us-vs-them, et al.)
  • Whole-team training, Ross, Robbie, Laura, the new growth lead, the designer, and the community manager all walked through how the system runs
M2

M3

Optimization + handoff

▸ INSTALLS WEEKS 7–12
  • QC agent loop, reviews each generated static against a 50-point checklist, regenerates until it passes
  • Human-feedback markdown updater, when your designer or growth lead flags an issue, the relevant markdown file updates so the system stops repeating it
  • Iteration-on-winners pipeline, variants of any ad that hits, changing one variable at a time (persona / angle / emotion / format) so winners keep working
  • Brand-voice agent trained on Jupiter's winning ads, every future brief sounds like Jupiter
  • Handoff doc + "running it without me" checklist + onboarding playbook for the new head of growth
CHAPTER IV · HOW WE WORK

Mine. Train. Hand off.

Sprint in month 1 because that's where the build and the historical mining happen. Calibration through months 2 and 3, the system runs and just needs nudging. Slack-embedded throughout, kit drops as I ship new modules.

CHAPTER V · INVESTMENT

One number. Three months.

$12,0003 months committed · billed monthly · USD
MONTH 01 · MINE
$5,000
MONTH 02 · TRAIN
$3,500
MONTH 03 · HANDOFF
$3,500

Heavier in month 1 because that's where the build and the historical mining happen. Lower after because the system runs itself with light calibration once it's in. 3-month minimum committed up front. This replaces what you're paying a performance creative agency, and the system is yours to keep running afterward.

What happens after month 3?
Optional ongoing collab at $2,500/mo, Slack access, kit drops, and one 60-min call per month. Month-to-month, 30 days notice to cancel. Kit drops and Adlib access end on cancel. You keep every file already on your machines.
FINE PRINT

What's not in the box.

So nobody's surprised in month 2. If any of these become important, we scope them separately.

  • Media buying or ad-account management, your team and any retained agency keep that
  • Final QC on shipped ads, Jupiter humans stay in the loop
  • Hiring the new head of growth, I'll send candidate intros from my network, but recruiting and comp are on you
  • Work for direct competitors of Jupiter in anti-dandruff or scalp-care, mutual exclusivity in the SOW
CHAPTER VI · KICKOFF

From signed to first AI statics shipped in 30 days.

The sprint pace. Each item below is what's true at the end of that week.

▶ DAY 0

Signed + Stripe.

SOW countersigned, M1 invoice paid. Slack channel created. Kickoff call scheduled inside 48 hours.

▶ WEEK 01

Access + 5 years ingested.

Meta API access provisioned. All historical Jupiter ads ingested and tagged by persona, angle, emotion, asset type, format. Repo skeleton committed.

▶ WEEK 02

Historical analysis delivered.

The "what worked / what didn't / why" report you've been asking your team for years lands in Slack. Brand bible and spec cards drafted.

▶ WEEK 03

Production pipeline live.

Brief-generation agents, prompt converter, and fal.ai static generation all running on Jupiter's repo. First test briefs through the 10-agent scoring loop.

▶ WEEK 04 · SPRINT END

First AI statics shipped to Jupiter's Meta account.

Statics generated, QC'd, and live in the account. Sprint cadence ends. We move to bi-weekly + Slack for months 2–3.

▶ MONTHS 02–03

Train. Optimize. Hand off.

Whole-team teach-ins through M2. Iteration-on-winners pipeline + brand-voice agent through M3. Onboarding playbook for the new head of growth delivered at month-end.

▶ READY TO START? ◀

Let's mine five years
and ship the first statics.

If this is a go, I send the SOW + Stripe link inside 24 hours. Kickoff call scheduled inside 48. First statics in your Meta account inside 30 days.

Will Sartorius
Will Sartorius · April 2026