Quick Summary

Your paid social has stopped converting. We know why, we have seen it before, and we have a precise fix. This document covers what we will build, how it works, what it costs, and - critically - how it runs automatically after setup so you never have to think about it again.

Everything in this plan runs on free infrastructure (GitHub Actions). No SaaS subscriptions, no agency retainers, no dashboards to check every day. The system watches your ad accounts 24/7 and emails you when something needs attention.

What we need from you: API access to three platforms (Google Ads, Meta, Shopify). Setup takes 30-60 minutes on your end. Everything else we handle.

Section 1: Why Your Ads Stopped Converting

Based on your store data and common DTC failure patterns at your AOV (~$70-90), here are the five most likely causes:

1. Creative Fatigue (Most Likely - ~70% chance)

The Meta algorithm shows your ads to the same people repeatedly once it identifies your target audience. When someone has seen the same creative 3-5 times, they stop registering it. CTR drops. CPM rises. ROAS collapses.

The signal: Frequency above 3.0 on your best-performing ad sets, combined with CTR falling over the last 30 days on the same creatives.

The fix: Fresh creative, and a system that automatically flags fatigue before it becomes a crisis.

Industry benchmark: Creative refresh cycle should be every 4-6 weeks for top-of-funnel, every 8-12 weeks for retargeting.

2. Audience Saturation (Likely - ~50% chance)

If your campaigns have been running for 12+ months with the same targeting parameters, you have likely reached most of the people in your defined audience who were ever going to convert.

The signal: Reach declining or plateauing while frequency is increasing. Cost per result rising faster than market CPM increases.

The fix: Lookalike audiences built from your actual top-spending customers. Not lookalikes built from "website visitors" - those are too broad.

3. Wrong Audience Composition (Possible - ~30% chance)

If any campaigns target "plant-based diet" or "veganism" interest categories, you are targeting a small, expensive audience. Your real buyer is a meat-eater who is curious or health-motivated - a vastly larger, cheaper-to-reach pool.

The fix: Audience rebuild using flexitarian and omnivore interest stacks, with your real customer data as the targeting anchor.

4. Attribution Window Mismatch (Possible - ~25% chance)

Your products cost $60-160 and are frozen goods that require a considered purchase decision. People do not click and buy in the same session. If campaigns optimize for a 1-day click window, they miss late converters.

The fix: Set attribution to 7-day click / 1-day view for your price point.

5. Google Ads Budget Cannibalization (Worth Investigating - ~20% chance)

If Google brand campaigns are getting high budgets while non-brand prospecting is underfunded, you are paying for people who were already going to find you.

The fix: The Google Ads audit will show impression share, budget loss by campaign, and whether brand terms are eating prospecting budget.

Section 2: What We Know About Your Business

US Store (juicymarbles.com)

EU Store (eu.juicymarbles.com)

• Umami Burger (EUR 19.99 / 6 patties) - EU only, koji + miso, 30-day shelf life
• Lamb-ish is available in EU even though it is OOS in the US
• Fish products (Salmon, Cod): EUR 19.99 entry price vs $65.99 in the US - this is not a typo

The Fish Pricing Problem

Volume Discount Architecture

The 10-pack and 20-pack of Thick-Cut Filet cost the same per-unit ($8.00). That is a "smart buyer" signal that almost no brand leverages in ad copy:

"The 10-pack and the 20-pack cost exactly the same per filet. Buy 20, it feels like theft."

This works especially well as retargeting creative for people who have already visited the product page.

Free Shipping Threshold

Your free shipping threshold is $120 USD / EUR 120 EUR. A single 6-pack of Thick-Cut Filet at $64.99 does not hit it. Two packs do. Entry-price ads should hint at free shipping. Retargeting ads to single-item cart abandoners should show the "add one more, ship free" angle.

Brand Positioning

Your brand voice is rare: irreverent, chef-forward, and philosophically serious about food. "Plant-based, but big meat energy" is not just a tagline - it is a product thesis that actively rejects the category conventions.

Your target customer is a home cook who eats meat and is open to interesting alternatives. Not a vegan. Running ads that look like they target vegans is a systematic audience mismatch.

Section 3: How This Actually Gets Built - Claude Code

Everything in Section 12 - the analytics pipelines, the smart policies engine, the ML optimizer, the RSA generator - is real, tested, production software running for other clients right now. The "adaptation" required to make it work for Juicy Marbles is not a development project. It is a conversation.

What Claude Code Is

Claude Code is an AI that lives in the terminal and acts as an autonomous software engineer. It can read entire codebases, understand what they do, modify them, write new scripts, and run them - responding to plain English instructions.

It is not a chatbot. It opens your files, reads them, makes the changes itself, and shows you the diff before saving.

What This Means in Practice

Henrik opens Claude Code and types something like:

"Take the Google Ads export script and adapt it for Juicy Marbles. The customer ID is 123-456-7890. Output should go to /reports/google-ads/. Here is the store overview document - use it to understand the two stores."

Claude Code reads the existing script (which may be 500 lines long), reads the store document, understands both, and makes every change needed. That is not a 2-week development sprint. That is 20 minutes.

What This Means for Ongoing Changes

If your product catalog changes, the update takes minutes. If Meta changes their API, Claude Code reads the documentation and patches the scripts. The maintenance cost is proportional to how often things change - not to how complex the system is.

What This Does Not Do

Claude Code is the engineering layer, not the strategy layer. It cannot tell you whether your Salmon pricing in the US is too high to convert - that takes understanding your business, your margin structure, and your competitors. That is what Henrik provides.

Section 4: What We Will Build

Component A - Shopify Customer Intelligence Export

Pulls your full order and customer history from both stores, scores every customer on three dimensions (Recency, Frequency, Monetary spend), and produces ready-to-upload audience files for Meta and Google.

Script: scripts/export-shopify-data.py | Runtime: ~2-5 minutes per store

Component B - Google Ads Audit

Script: scripts/export-google-ads-data.py | Output: 16 CSV reports

Component C - Meta Ads Audit

Script: scripts/export-meta-ads-data.py | Output: 4 CSV files with creative fatigue detection

Component D - Audience Architecture Rebuild

Once we have the Shopify export and the Meta/Google audits, we build the full audience system from scratch.

Component E - Ad Copy Generator (Phase 2)

Generates 15 headline variants + 4 description variants per batch, pre-scored for brand voice match, compliance, and hook strength.

Section 5: Full Automation - How Everything Runs Itself

The Automation Stack (Zero-Cost)

No Zapier. No Make/Integromat. No Looker Studio. No third-party analytics tools. Everything runs on infrastructure you control.

Automated Schedule

Critical Alerts (Email Immediately)

Meta Automated Campaign Rules

Google Ads Scripts (Automatic)

Section 6: Creative Testing Framework

The Creative Lifecycle

Hook Rate Benchmarks (Food Category, Meta)

Seasonal Creative Calendar

Landing Page Strategy

Section 7: Google Shopping Setup

People who search "buy plant-based filet" or "plant meat online" are ready to purchase. Shopping ads capture them visually before they reach organic results.

Shopping Feed Title Format

• Juicy Marbles Thick-Cut Filet Plant-Based 20-Pack Frozen
• Juicy Marbles Baby Ribs Plant-Based 6-Rack Vegan Meat Alternative
• Juicy Marbles Kinda Salmon Plant-Based 4-Filet Frozen

Performance Max (PMax) Guidance

• Set up PMax for your top 3 SKUs
• Add brand exclusions (exclude "juicy marbles" - you do not want to pay for brand-name searches)
• Upload high-quality assets: 3-5 images per product, 2-3 video clips, all 15 headlines, all 4 descriptions
• PMax will cannibalize your other campaigns if budgets are shared
• PMax ROAS numbers include assisted conversions - actual incremental ROAS is usually 15-25% lower

Section 8: Timeline

The actual bottleneck is always API access setup. If you can get credentials to us within 48 hours of reading this, you can have both audit reports by end of the week.

Section 9: What We Need From You

Google Ads

Meta

Shopify

For each store (US and EU separately): Admin > Settings > Apps and Sales Channels > Develop Apps > Create an App. API Access: grant read_orders + read_customers only. Copy the Admin API access token. Share token + store domain.

These are read-only credentials. We cannot place orders, modify products, or access customer passwords.

GitHub

Create a free account at github.com. Share the username and we will set up the automation repository. All API keys stored as encrypted secrets - not visible to anyone once uploaded.

Section 10: Budget Framework

Meta Budget Split (Recommended)

Scaling Rule

Section 11: Performance Targets

Section 12: The Production System - What You Are Actually Getting

We run two other clients on fully automated ad intelligence systems: TEEI (a Norwegian education nonprofit) and YPAI (a professional development platform). Both run entirely without human intervention between weekly check-ins. What you are getting is a battle-hardened version of the same stack, tuned for an e-commerce DTC brand at your price point.

11A - The 11 "God Mode" Analytics Pipelines

These are diagnostic lenses - each one answers a specific question about why your account performs the way it does. They run on your Google Ads data after the weekly export.

How this works for you: Every Monday, all 11 pipelines run against your fresh data. The output is a prioritized action list - not a spreadsheet to interpret yourself.

11B - Intelligent Keyword System

• Auto-Pause Zero-QS Keywords: Any keyword with Quality Score below 3 gets automatically paused every day.
• Search Term Harvester: Any query that drove a conversion but is not yet a keyword gets promoted to Exact Match. Any query that spent more than EUR 5 with zero conversions gets added as a Negative keyword.
• Paid-Organic Gap Analysis: Identifies keywords you are paying for where you already rank organically in position 1-3. Monthly cost: ~EUR 0.20.
• Negative Keyword Architecture: Master shared negative list that pre-blocks job-seeking queries, DIY queries, competitor brand names, and irrelevant category terms.

11C - GTM + Conversion Integrity Audit

Before we trust any ROAS numbers, we verify the tracking is correct. The GTM audit inventories every tag, trigger, and variable in your GTM container and maps each conversion tag to the corresponding Google Ads conversion action. If your purchase event is not firing on every completed checkout, Google's algorithm is learning from incomplete data.

11D - Gemini-Powered Responsive Search Ad Generator

Every RSA goes through: Context Extraction from your landing page, Gemini 2.5 Pro ad generation (15 headlines + 4 descriptions), Quality Scoring on 4 dimensions (0-10), and a gate at 8.0 minimum. Any ad below 8.0 goes back for revision.

11E - Smart Campaign Policies Engine

• Geo Targeting Policy: Ensures every campaign is set to "Presence Only" (not "Presence or Interest" - a default that silently wastes budget).
• Device Bid Adjustment: Mobile/tablet bids suppressed -70% to -90% (frozen goods are not impulse buys on mobile). Calibrated after 60 days of your data.
• Bidding Governance: Blocks Target ROAS bidding if a campaign has <30 conversions/month. Below that threshold, it uses Maximize Conversions instead.

11F - ML-Powered Creative Variant Testing

• Thompson Sampling Bandit Optimizer: Bayesian algorithm that automatically shifts budget toward better-performing creative variants in real-time, without waiting for full A/B test significance.
• A/B Test Analyzer: Both frequentist (chi-squared) and Bayesian analysis with minimum sample size calculation.
• Anomaly Detector (Isolation Forest): ML model trained on your normal account behavior. Flags statistically abnormal changes in ROAS, CPM, or conversion volume immediately.

11G - Recipe Blog to Paid Traffic Pipeline (New for Juicy Marbles)

Your 80+ recipe articles are an enormous untapped asset. The system indexes your blog by product and occasion, cross-references with actual Google search terms, and builds shoppable landing pages. Recipe-led ad creative ("Orange Whisky-Glazed Ribs. Yes, plant-based.") captures people through the dish, not the product.

11H - Multi-Store Automation (US + EU Separation)

Every script is multi-tenant by design. US and EU campaigns are tracked, audited, and reported separately. Audience files are built per market. EU-specific additions include: Umami Burger campaign template, Lamb-ish restock alerts, fish product pricing context, and 19+ country geo allow-list.

11I - The Complete Automation Schedule

Daily (06:00 UTC)

Weekly (Monday 07:00 UTC)

Monthly (1st of month, 08:00 UTC)

Quarterly (Jan / Apr / Jul / Oct)

11J - Total Stack: What Gets Adapted for You

Source Systems (Already Built and Running):

• TEEI Google Ads Platform: 172+ tools, C# MCP server, 177 Python scripts, 8 API integrations
• YPAI Ads and SEO: 16-report GAQL exporter, smart policies engine, ML keyword scoring
• Apollo AI Outreach: XGBoost propensity model, Thompson Sampling bandit, Bayesian A/B testing, Isolation Forest anomaly detection

Zero new infrastructure to buy. Everything runs on GitHub Actions (free), Python (free), and APIs you already need for the account anyway.

Security Notice

Next Step

Send us whichever API access you can set up first - Meta or Google, your call. We will have the first audit report back to you within 24 hours of receiving credentials.

The automation setup (GitHub Actions + alerts) runs in parallel with the audits. By the time the audits are done, the monitoring system will already be live.

The first week will show you exactly what has been happening. The second week we fix it.