How We Built an AI Growth Director That Turned 10 Years of Work into 38 Posts in 4 Months

The Problem

IB Solutions had 10+ years of integration work, dozens of client projects, and real systems running in production. None of that was turning into content.

The issue was not a lack of expertise. It was not knowing where to start. What's worth sharing? Which project has an angle that would land with the right audience? How do you turn a 2-year client engagement into a LinkedIn post that says something useful in 150 words?

Before this system, IB Solutions had zero consistent content output. No posting schedule, no content backlog, no process for turning experience into published work. The founder knew the work was strong. Translating it into content that reached the right people was a different skill entirely.

What We Built

We built an AI-powered content system where Claude operates as the Growth Director. Not a tool we open when we need a draft. A persistent system connected to everything the practice knows, running on a schedule, surfacing ideas from data we would never have combined manually.

The system connects Claude to nine sources of information that together represent everything the practice knows:

• Growth strategy and marketing strategy defining positioning, ICP, content themes, and the engagement model
• Client case studies and work pages from real engagements with real numbers
• Client work tasks tracked in Notion, where recent project decisions, implementation details, and technical choices live as they happen
• LinkedIn performance data showing which content formats, topics, and proof points drive engagement, saves, and follower conversion
• Meeting recordings and raw notes where the founder's thinking happens before it becomes structured
• Growth-brain skills stored in GitHub: codified workflows for content creation, competitor tracking, website tasks, and Strapi CMS management
• A style guide defining voice, tone, formatting rules, and the specific language patterns that make the content sound like the founder, not like AI
• Published Insights on Strapi CMS so the system knows what has already been said in long form and avoids repeating angles
• The Notion content board where every post idea, draft, and scheduled post lives with status, category, publish date, and linked insight slug

Claude does not just draft when asked. A content miner runs on a schedule, scanning across these sources to surface post ideas that connect information the founder would not have combined on his own. A client work task closed last week combined with a pattern from the marketing strategy becomes a Tuesday Teach post. A performance stat from last month's analytics paired with a meeting recording about a technical decision becomes a Thursday Proof post.

How a Content Session Works

The founder opens the Claude Project. All context is already loaded. No re-explaining the strategy, the voice, or what has been published recently.

A typical session:

1. Pull the current schedule from the Notion content board. Claude shows what is confirmed, what is in draft, and which slots are empty for the coming week.
2. Review ideas the content miner has surfaced. These are not generic suggestions. They are specific angles drawn from real data: a client proof point that has not been used recently, a performance insight from the analytics, a pattern the founder mentioned in a meeting.
3. Draft posts together. The founder provides direction and corrections. Claude drafts against the style guide, checks the hook against voice rules, validates the integration anchor, and runs the full quality checklist before presenting the draft.
4. Approve and schedule. Approved posts go directly into the Notion content board with status, category, publish date, and content. No spreadsheet, no copy-pasting between tools.
5. Capture new ideas. Anything that comes up during the session that is worth saving goes into the backlog as an Idea, with context attached so the next session starts informed.

The Weekly Cadence

The system maintains a 3-post-per-week cadence: Tuesday, Wednesday, and Thursday, published between 9:00 and 9:30 AM CET.

• Tuesday: Teach (Systems Thinking) names friction in language a pre-integration buyer recognizes. Stage 1 accessible, no jargon.
• Wednesday: Teach (Integrator's POV) sharp, short observations from inside the practice. What we see that most vendors miss.
• Thursday: Proof (Real Work) a real client story with a real number. The non-negotiable proof slot, and the highest-performing category by every metric.

Once a month, one Thursday rotates into a "How We Work" post that makes the engagement model visible to prospects. The system tracks the rotation.

What the System Changed

The biggest shift was not speed. It was capability.

Before this system, the founder had no content output. Not inconsistent content. Zero. The expertise was there. The projects were there. The translation layer between "work I have done" and "content worth publishing" did not exist.

The AI Growth Director created that layer. It holds the full picture: what has been published, what performed, what the strategy says to focus on, what the client work proves, what tasks were completed this week, and what the founder said in a meeting that could become a post. No human content manager could hold all of that context simultaneously and surface the right connection at the right time.

Almost every content idea over the last four months came from the system, not from the founder staring at a blank screen. The system does not replace the founder's judgment. It makes the judgment possible by putting the right information in front of him at the right moment.

The Numbers

• 38 posts published in 4 months
• 3 posts per week, consistent, no missed weeks
• Thursday Proof posts average 4x the impressions of any other category
• Content output went from zero to a sustained publishing operation in under a month
• The content miner runs on a schedule, surfacing 5-10 post ideas per cycle from existing data
• Every post is validated against a style guide, voice rules, integration anchor test, and quality checklist before the founder sees it

Where This Pattern Applies

The underlying pattern is not specific to content. Any recurring knowledge work that currently depends on one person holding all the context in their head is a candidate for this kind of system.

A marketing director could connect an AI workspace to campaign briefs, brand guidelines, and performance data. A founder running sales could connect it to CRM data, proposal history, and meeting notes. The mechanism is the same: give the AI persistent access to the information the human already has, and let it surface the connections the human would miss because they cannot hold it all at once.

The prerequisite is that the information exists somewhere structured. If the expertise is only in the founder's head and nowhere else, the system has nothing to draw from. The first step is always getting the knowledge into documents, databases, and recordings that the AI can read.

Four months ago, IB Solutions had 10+ years of integration work and nothing to show for it on LinkedIn. Today the system publishes 3 posts per week, every week, each one grounded in real work and validated against a style guide the founder wrote. The constraint was never the expertise. It was not having a system that could see all of it at once and say what was worth sharing next.