I'm going to show you exactly what I built, what broke, what I had to scrap and rebuild, and what it actually cost. Not the polished version you get in a webinar. The real one.
The Protocol runs on a full AI marketing pipeline. I built it myself, over about eight months, in parallel with running the business. It's not perfect. There are bits I'd redo. But it works, and it works consistently. Content goes in one end, leads and clients come out the other. Here's how it's structured.
The Starting Point: I Had No Stack, Just Subscriptions
When I started building this properly, I had what most business owners have: a collection of disconnected tools I was paying for. ChatGPT for writing. Canva for graphics. MailChimp for email. Meta Ads that I was running manually with no creative system. No integration between any of it. Every piece of content required me to start from scratch.
The problem wasn't the tools. It was the absence of a pipeline, a connected system where each stage feeds the next without manual intervention. That's what I set out to build.
Layer 1: Content Creation, The Viral-Script Skill
Viral-Script Skill (Claude-powered)
The first layer is content strategy and scripting. I built a custom skill, a structured prompt framework running on Claude, that takes a topic brief and outputs: three title variations, three hook options, a full Reel script, and a ManyChat CTA. The output is already formatted for production. No editing required, just review and approve. This is what I meant when I said AI amplifies leverage, one brief becomes a full piece of content in under two minutes.
Cost: ~$100/mo (Claude Max)The key thing that makes this work is that the skill is trained on my specific voice, the directness, the UK context, the specific methodology I teach. Generic AI scripts sound generic. Trained AI scripts sound like you. That training took about two weeks of iteration, and it was the highest-leverage two weeks I've spent on marketing.
Layer 2: Video Rendering, Remotion
Remotion (React video framework)
Once the script is approved, it goes into a Remotion template that renders a branded caption video. On-screen text, animated captions, Protocol Green colour scheme, the emblem in the corner. Consistent every time. The render takes about 25 seconds. Manually placing captions in CapCut takes 45–60 minutes per video. With Remotion, you review the script, hit render, and the video is production-ready. I have multiple Remotion templates: a talking-head style for educational content, a text-only kinetic typography version for hook videos, and a stats callout template for social proof content.
Cost: Free (open source)"Every piece of the stack was built to solve a specific bottleneck. Not because it was technically interesting, because it was costing me time I didn't have."
Layer 3: Instagram Automation, ManyChat
ManyChat (Instagram DM automation)
Every Reel ends with a keyword CTA, something like "Comment FREE below and I'll send you the framework." ManyChat catches that comment, sends a DM automatically within seconds, and walks the person through a short flow: deliver the freebie, ask one qualifying question, capture their email. The email then gets pushed to MailerLite via webhook. The first ManyChat flow went live and immediately demonstrated the model, a single Reel with a keyword trigger starts building your list automatically from day one. In my experience, comment-to-DM flows convert at significantly higher rates than landing pages, because there's no friction and no page load between the comment and the resource.
Cost: ~$15/mo (Essential plan, check current GBP pricing)Layer 4: Email Marketing, MailerLite
MailerLite
Every email that comes through ManyChat triggers a welcome sequence in MailerLite. Five emails over ten days: introduction, core methodology, case study, common mistakes, offer. All written with Claude, reviewed by me, timed and sequenced in MailerLite's automation builder. In my experience, email lists built from engaged content viewers outperform cold or scraped lists significantly, because every subscriber actively wanted what you offered. The quality of the lead source compounds into the quality of the email performance.
Cost: Free (free plan, up to 1,000 subscribers)Layer 5: Paid Traffic, Meta Ads
Meta Ads (Facebook & Instagram)
Once organic content has proven which messages resonate, measured by comment rate and ManyChat conversion, I put ad budget behind the winners. Not the other way around. The ad creative gets generated at scale using Fal.ai (more on that below). The campaign structure follows the Charley T framework: broad targeting, three to four creatives per ad set, optimising for lead events, with a kill/scale rule based on cost per lead after 48 hours. This is the part I've rebuilt most. The first version was too conservative, I was pulling ads before they'd had enough data to learn. The second version was too aggressive, scaling before confirming lead quality. The current version is more systematic: let it breathe for 48 hours, check CPL against target, scale winners by 20% every 48 hours if they're performing.
Cost: Ad spend variable (~£500–1,500/mo budget)Layer 6: Visual Creative, Fal.ai + Creative Engine
Fal.ai + GPT Image 2
Ad creatives and social graphics are generated through Fal.ai running GPT Image 2 via API. I built a script I call the creative-engine that takes a brief, pulls in brand reference images (logo, emblem, headshot, lifestyle shot from Cloudinary), and renders a batch of images at medium quality for review, then high quality for the approved ones. Before this, getting a set of ad creatives meant briefing a designer, waiting three days, getting something that was 80% right, going back and forth. Now I generate ten concepts in about eight minutes, pick the best three, and they're ready to go straight into Ads Manager.
Cost: ~$0.04–0.17 per image (pay-per-use)Layer 7: Agent Automation, OpenClaw
OpenClaw
OpenClaw is my personal AI agent running on the VPS. It handles automations, webhook processing, triggered tasks, background work that runs without me in the loop. The marketing pipeline generates a lot of moving parts: content gets created, leads come in, sequences trigger, creatives get generated. OpenClaw is the connective tissue that keeps it all moving. I route a significant portion of my automations through it rather than through third-party tools that would otherwise charge per task or require separate subscriptions.
Cost: VPS-hosted (server cost included in infrastructure)Layer 8: Code Intelligence, Codex
OpenAI Codex
Codex runs alongside Claude as a second agent in my development workflow. When I build something, a new automation, a new script, a new funnel integration, Codex cross-references the logic independently. Two AI agents reviewing the same code catch more than one does. It’s also useful for quick implementation tasks where I want a second opinion before shipping. The ~$20/mo is one of the highest-leverage subscriptions in the stack relative to cost.
Cost: ~$20/moLayer 9: Local AI Models, Ollama
Ollama
Ollama lets me run language models locally, useful for tasks where I want processing to stay on-machine, for bulk jobs where API costs would compound, and for testing prompt logic before routing it to a paid API. Having local model capability alongside cloud models gives the stack flexibility: not every task needs to go through an external API, and for the ones that don’t, local inference is both faster and cheaper.
Cost: Lowest planSupporting Infrastructure
WhisperFlow, Voice to Text
WhisperFlow converts voice to text and pastes it directly into whatever app is active. I use it to brief AI tools by speaking rather than typing, faster for longer context, and the output is more natural. When I'm walking through what I want a script to do, or describing a creative direction, speaking it out is significantly quicker than writing it. It's one of those tools that sounds trivial but changes how you actually interact with AI day-to-day.
Cost: Low monthly planAirtable, Content Pipeline
Every piece of content generated, images, videos, ad creatives, Remotion renders, gets logged to an Airtable base automatically. Brief, output URL, status, prompt used. It acts as the master log for the content pipeline: I can see at a glance what's been generated, what's approved, what's running as ads. Without this, the pipeline produces content into a void. With it, everything is auditable and trackable from one place.
Cost: Free tierNotion, Client Project Management
When a client purchases a website or funnel build, they fill a detailed intake form. That submission goes directly into a Notion database, client name, contact details, project type, brand assets, what they're selling, their audience, their goals. Everything in one place before a single line of code is written. Notion is the handoff point between the sale and the build, and it keeps every client project organised without emails going back and forth.
Cost: Free tiern8n, Automation Engine (On Deck)
n8n is self-hosted on the VPS and wired in, but not yet running active workflows. It's there for when the automation requirements outgrow what ManyChat and webhooks handle natively, multi-step sequences, scheduled tasks, cross-platform triggers, and complex automation logic. The cost is essentially zero since it runs on the same server as everything else. When I need it, it's ready.
Cost: Free (self-hosted on VPS)Contabo VPS, The Engine Room
Everything that needs to run 24/7 lives on a Contabo VPS. OpenClaw, n8n, the webhook that catches form submissions and fires them to Notion and Telegram, the lead CSV, Remotion server renders, all of it. A VPS is not optional for a stack like this. Cloud tools with their own servers are fine for simple use cases, but once you're running agents, custom webhooks, and background processes, you need a machine you control. At under £10/mo, it's the highest-leverage line item in the stack.
Cost: ~£8–10/mo (Contabo)What the Whole Stack Actually Costs
| Tool | Use | Monthly Cost |
|---|---|---|
| Claude Max | Scripts, copy, strategy | ~$100 |
| Remotion | Video rendering | Free (open source) |
| ManyChat Essential | Instagram automation | ~£15 |
| MailerLite | Email sequences | Free |
| Fal.ai (image gen) | Ad creatives, visuals | ~£20 |
| Cloudinary | Brand asset CDN | ~£0 (free tier) |
| Codex | Code review + agent | £20 |
| OpenClaw | Agent automation (VPS) | VPS-hosted |
| Ollama | Local AI models | ~$20/mo |
| WhisperFlow | Voice to text | Low plan |
| Airtable | Content pipeline log | Free |
| Notion | Client project management | Free |
| n8n (self-hosted) | Automation engine | Free |
| Contabo VPS | Infrastructure (24/7) | £16 |
| Total (excl. ad spend) | ~£155/mo | |
Ad spend is separate and variable, but the infrastructure cost of the entire stack is £72 a month. That's less than most people spend on a single software subscription they barely use.
All prices approximate and subject to change, verify current pricing with each provider.
What Broke (And What I'd Do Differently)
A few honest notes on what didn't work first time:
- ManyChat flows can break when Instagram updates their API, it’s happened to plenty of users. Worth building fallback flows and monitoring for API errors weekly.
- The Remotion templates took longer than expected to get right. The first three versions looked wrong, the typography was too small, the timing felt rushed. Budget more time for template development than you think you need.
- A common sequence mistake: starting too long. If open rates drop after email three, trim the sequence to five or fewer and move the offer earlier.
- I tried to scale ads too early. The lesson I would give anyone starting: validate your offer organically before touching paid.
The overall build time was about eight months of part-time work. If I were starting again knowing what I know now, I'd have it running in eight weeks. The mistakes weren't wasted, but they were avoidable with better guidance earlier.
Book a free call if you want to shortcut some of these decisions.
Want to talk through how this applies to your business?
Book a free 30-minute call. We'll look at where you are, what's worth automating first, and whether The Protocol is the right fit.
Book a Free Call