Frameworks, workflows, and real implementation details from someone who builds AI marketing systems every day. Not theory. Practice.
There are two types of marketers right now: those who ignore AI entirely, and those who add "AI-powered" to their slide decks without changing how they actually work. Both are losing.
The marketers who will win are the ones using AI as infrastructure, not decoration. Not "we used ChatGPT to write a blog post." More like: we built an autonomous system that researches competitors, drafts positioning, generates campaigns, and optimizes in real-time.
That takes a framework. Here's the one I built and use every day.
IDM is a framework I developed based on a simple insight: intent is the new source code for everything. The same governance model that makes software development reliable also makes marketing reliable.
Most marketing teams generate content in a vacuum. There's no structured connection between strategy, execution, and learning. IDM fixes that with a five-step cycle:
The critical rule: no generation without intent. Every piece of content, every campaign, every ad traces back to a structured intent document that defines the goal, audience, messaging boundaries, and success criteria. AI agents read these files before generating anything.
IDM organizes intent into four layers. Lower layers constrain higher layers — generated output must be consistent with strategy.
| Layer | What It Holds | Who Owns It |
|---|---|---|
| L0: Strategy | Why we exist, goals, market position | Leadership |
| L1: Domain Truth | Product facts, personas, positioning | Domain experts |
| L2: Active Work | Campaigns, calendar, channel config | Team leads |
| L3: Output | Content, ads, emails, landing pages | AI + human review |
This means an AI agent generating a LinkedIn post (L3) must check it against your brand voice (L1) and campaign goals (L2) before publishing. Claims must trace to product facts. Messaging must stay within positioning boundaries. The system enforces quality through structure, not hope.
IDM is part of a larger framework called IMF (Intent Management Framework) that unifies how intent flows across development and marketing. The same pattern powers both:
| Step | Development (IDD) | Marketing (IDM) |
|---|---|---|
| Define intent | Interview → spec file | Campaign INTENT.md |
| Decompose | Task execution master | Campaign master agent |
| Quality gate | Tests (TDD) | Claim verification, voice check |
| Execute | Code generation | Content creation via skills |
| Verify | E2E tests | Distribution confirmation |
| Learn | Sync back to intent | Metrics → update intent |
Why this matters: When development and marketing share the same intent layer, product truth flows directly into marketing content. No more "marketing said something engineering didn't build." The source of truth is the same file.
Forget the 200-tool listicles. Here's what actually runs the system, organized by function.
Model Context Protocol (MCP) is the standard that lets AI agents use external tools. Instead of copying data between systems, your AI reads from Google Analytics, writes to your CRM, posts to social, and sends emails — all through structured tool calls.
This is the difference between "AI helps me write" and "AI runs my marketing system."
Here's what a real marketing MCP setup looks like:
| MCP Server | What It Does | Marketing Use |
|---|---|---|
| Google Ads | 43 tools: campaigns, keywords, creatives, performance | AI-optimized ad spend, automated creative testing |
| GA4 Analytics | Real-time reports, custom dimensions, audience data | Automated performance monitoring, anomaly detection |
| Search Console | Search queries, indexing, click performance | SEO monitoring, content gap identification |
| Social APIs | Post, schedule, engage, analyze across platforms | Autonomous content distribution and engagement |
| Email (SendGrid) | Send, template, list management, campaign stats | AI-personalized email nurturing at scale |
The key shift: MCP turns your AI from a content writer into a marketing operator. It can read performance data, make optimization decisions, and take action — with human approval gates for anything involving spend or public publishing.
The difference between mediocre AI output and genuinely useful marketing work comes down to prompt quality. Here are real prompts I use in production — not toy examples.
These prompts reference structured intent files. That's the IDM difference: AI output is constrained by your strategy, not floating in a vacuum.
An agent monitors competitor websites, pricing pages, and job postings. Synthesizes changes into a weekly brief with positioning implications. No more 4-hour manual research sessions.
4 hours/week saved. Insights are real-time, not quarterly.
Feed CRM data and win/loss notes into an LLM. It identifies patterns humans miss: which titles convert fastest, what signals predict high LTV, which messaging resonates by segment.
Discovered 3 ICP segments with 3x higher conversion rates.
Not "ask ChatGPT to write a blog." A structured pipeline: research phase, outline with competitive hooks, draft with brand voice rules, edit pass, SEO optimization, automated distribution through n8n. Intent files constrain every step.
5x content output. One person doing what took a team of 3.
n8n + LLMs + CRM + social APIs. A campaign brief goes in. Copy variants, scheduled posts, email sequences, and attribution tracking come out. You review and approve. The system executes.
Generic drip sequences are dead. LLMs personalize email content based on industry, role, company stage, and behavior. Not mail-merge personalization — actual narrative personalization that reads differently for every prospect.
50% higher pipeline conversion. 25% better retention.
Tactical AI tools are the starting point. The real power comes from connecting them into loops that compound — where each cycle makes the system smarter.
Keyword research → create content → distribute to platforms → measure engagement → learn what works → better research → repeat.
Content attracts visitor → form captures lead → email nurtures → score rises → demo offered → sale closed → testimonial becomes ad creative → repeat.
Organic content performs → boost with paid → measure CPA by channel → shift budget to winners → creative engine refreshes underperformers → repeat.
Each loop feeds the others. Content that drives organic traffic identifies topics for paid amplification. Leads that convert reveal which content works best. Paid data shows which audiences respond, improving organic targeting.
The learning agent closes the loop: it measures results from every campaign, every post, every email — and syncs insights back into intent files so the next cycle starts smarter than the last.
Most marketing teams run disconnected activities. Content team publishes. Demand gen runs ads. Events team does events. Nobody learns from each other's data. The three loops architecture connects everything through shared intent files and a learning agent that updates strategy based on actual results.
You don't go from zero to autonomous marketing overnight. Here's the phased approach I recommend — each phase builds on the last and delivers value immediately.
Get your team using Claude or GPT-4 for daily work: research, outlines, first drafts, competitive analysis. Focus on speed gains with existing workflows. Don't publish raw AI output — use it to 3x your speed while maintaining quality.
Connect n8n or Zapier to your content calendar, social accounts, and email platform. Automate distribution so one input creates many outputs. Set up MCP servers for your analytics and ad platforms so AI can read performance data directly.
Write your IDM intent files: product truth, personas, positioning, brand voice. This is the highest-leverage step — it transforms AI from a generic content machine into a system that produces on-brand, on-strategy output every time.
Build agent teams that run campaigns with human approval gates. Campaign master decomposes intent into tasks. Content creator generates. Quality gate verifies. Distribution agent publishes. You approve — the system executes.
Deploy the learning agent. Every campaign, every post, every email gets measured. Results sync back to intent files. The next cycle starts smarter. This is where the system compounds — it gets better without you doing more work.
| Phase | Key Metrics | Target |
|---|---|---|
| 1. Individual | Time-to-publish, content volume per person | 3x speed improvement |
| 2. Automation | Manual hours saved, distribution reach | 10+ hrs/week reclaimed |
| 3. Intent | Brand consistency score, claim accuracy | Zero off-brand content |
| 4. Autonomous | Campaigns launched per month, pipeline velocity | 4x campaign throughput |
| 5. Learning | Cost per lead trend, conversion rate trend | Month-over-month improvement |
The principle: AI doesn't replace marketers. It replaces the manual, repetitive parts of marketing so you can focus on strategy, creativity, and the human connections that actually close deals. The goal is a system that compounds — getting better every cycle, not just bigger.
I implement this as a fractional Head of Marketing for B2B SaaS and AI companies. I build the system, not just advise on it. From intent files to running campaigns.
Let's Talk →