TL;DR – what happened in 7 weeks
- Over 7 weeks, 38 AI-assisted LinkedIn posts lifted our total monthly reach about 25x (roughly 2,200 to roughly 54,000 per month), by posting around 4x more often and reaching around 6x more people per post.
- The first post the system produced hit 25,560 impressions – the highest in our company’s history.
- Average reach per post rose 6.2x (338 to 2,113); even after stripping out our two viral posts, it still rose 2.8x.
- The honest limit: our median post moved only 1.3x (242 to 309). The system’s value is finding the winners and showing up consistently, not making every post go viral.
- A human still reviews, vet, accept/reject and approves every post before it ships. AI amplifies Clarity. It does not create it.
Everyone says AI content kills your LinkedIn reach. The first post I built with an AI system hit 25,560 views – the most in our history.
This is the honest before-and-after from a real Singapore SME – my own. Not “AI is magic.” Here is the system, the numbers, and the limits, including what did not work.
What is Claude Code, and why would a founder use it for marketing?
Claude Code is a command-line tool built by Anthropic that lets you run their Claude AI models against your own files, scripts, and instructions. Marketers know Claude as a chatbot. Claude Code is the same intelligence, but wired into a working environment where it can read your data, follow your rules, and produce finished work.
I am not a coder. I run Scaling-Up! Ventures, a strategy consulting, service design, growth culture and execution operationalisation firm. I picked Claude Code for marketing because it lets me do one thing a chatbot cannot: build a repeatable system that remembers our brand voice, our audience, and our strategy, then runs it with the same precision every time.
That last part matters more than the technology. A tool is only as good as the strategy behind it. This was never a tech flex. It was a bet that I could encode our Playing to Win™ strategy into a system and let it execute the parts I could not get totally right.
Why I stopped being the bottleneck on our own marketing
For years I was the bottleneck on our own marketing. Every post waited on me. I would draft something, sit on it, forget it, and three weeks would pass with nothing published.
This is pain point number one for most SME founders I work with: you are working IN the business instead of ON it, and you are the single point of failure for your own voice in the market.
The usual fix is to hire someone or pay an agency. For an SME our size, that is real money for output that often misses the voice and the strategy. Worse, when you outsource your own marketing, you outsource your Clarity. The agency does not carry your voice and conviction. They carry a content calendar.
I did not want more content. I wanted our actual strategy showing up consistently, in my voice, without me being the chokepoint.
How I built the AI marketing system (agents + skills)
I will keep this at the operating-system level, not the config level. The test I held myself to: could a non-technical founder follow this?
The agents
I set up a small team of specialised AI agents, each with a job description and one job. A strategist agent writes briefs. A writer agent drafts LinkedIn posts in my voice. An editor agent reviews them against our performance data. An analyst agent reads the numbers. Each agent knows our brand rules, our five pain points, and our content pillars, because I wrote them down once and the system reads them every time for every post.
The skills and the pipeline
On top of the agents sit repeatable workflows – what I call skills. A daily skill checks my idea sheet. A weekly skill scans Singapore business news for stories worth reacting to. Another schedule approved posts. The pipeline moves a post through clear stages: idea, draft, review, approved, scheduled, published, and update marketing dashboard metrics weekly. Then, ingest metrics, review, generate insights, what works, what didn’t and refine. Nothing skips ahead.
The human-in-the-loop QA gate
Here is the part that keeps it honest. Every draft hits an adversarial editor gate before it can move forward, and then it waits for a human – me or my colleague Mark – to approve it. No post ships without a person saying yes.
The system forced a discipline I would otherwise skip. It enforces the strategy. It does not invent one. That is the whole point: the AI executes a strategy we set, against rules we wrote, with a human signing off.
The results – what the data actually showed
Two windows.
BEFORE: 1 January to 9 April 2026, manual posting, 21 posts over 14.1 weeks.
AFTER: 10 April to 24 May 2026, the AI system, 38 posts over 6.4 weeks.
The windows are different lengths, so I will only compare per-post or per-month numbers. Raw window totals would be misleading, so I am not using them.
Total monthly reach went from about 2,200 to about 54,000 – roughly 25x.
That 25x is the headline, and it would be dishonest to leave it there. It came from two things stacking:
- We posted about 4x more often. Cadence went from 1.5 posts a week to 5.9 posts a week.
- Each post reached about 6x more people. Average reach per post went from 338 to 2,113.
Multiply 4x by 6x and you get roughly 25x. It is not that every post got 25x better. It is that we showed up far more, and each appearance landed harder.
Now the most important number in this whole piece: our median post moved only from 242 to 309 – about 1.3x.
Read that again. Half of our AI posts performed about the same as before. The average is pulled up by a small number of big winners. So the real value of the system is not that it makes every post go viral. It is that it helps us show up consistently and find the winners.
Two posts crossed 20,000 impressions and carry most of that average. The first was a post about a S$50K AI grant for SMEs, published 10 April at 26,874 impressions – our account record. The second was a story about a one-sign, four-word execution fix, at 20,941 impressions.
If you strip both of those outliers out, the average reach per post is 942 – still 2.8x the old baseline of 338. So even the skeptic’s floor is a genuine lift.
Two more strong AFTER posts, to show it was not only the megahits: a logistics story on effort versus outcome at 8,655 impressions, and “Not a Gardenia story, a Singapore story” at 14,449.
Engagement lifted too, on a per-post basis: comments went from 0.9 to 4.4 per post, and reactions from 4.8 to 8.8.
Two honest caveats before you take 25X to the bank.
First, the account was already climbing out of an October-November 2025 algorithm dip. Some of this lift is recovery, not AI. I cannot cleanly separate the two, so I am telling you up front.
Second, several of the AFTER posts are recent and their impressions are still maturing. So if anything, the AFTER numbers are conservative.
What drove the winners?
Three patterns the system enforced: reacting to Singapore business news within a day, using bolder and more contrarian hooks (those rose from under 5% of our posts to about a quarter), and posting in the Friday late-morning window that our own data says performs best.
What AI can and can’t replace in marketing
This is where I draw the line clearly, because it is the line most “AI marketing” pitches skip.
AI can replace the production grind: researching, scanning the news, shortlisting, drafting, scheduling, formatting, keeping cadence. That is the bottleneck for most founders, and removing it is what bought us 4x more posts.
AI cannot replace judgement, conviction, or a strategy worth executing. The median-post number proves it. When the AI Agents did not have a sharp angle or a real story, the post landed flat – 1.3x, basically where we started. The big winners came from real client situations, real Singapore news, and a strategy that told the system “Where to Play” and “How to Win”.
There is also a voice point worth taking head-on. Independent analysis (for example, Originality.AI’s reporting that a large and rising share of long LinkedIn posts are now AI-generated, while human-voiced content still tends to out-engage) lines up with what I saw: the posts that worked still sounded like a person who has done the work, not a content mill.
So no, AI did not run my marketing. Humans think. Humans plan. Humans design the system. Humans develop the processes. Humans automated the processes. Humans make judgement calls. Humans approve every post. AI makes it easier. AI amplifies Clarity. It does not create it.
How an SME leader can start (without being a coder)
You do not need to write code. You need three things.
- A strategy the system can enforce. Before any tool, get clear on who you serve (your ideal customer profile), where you play, and how you win. AI is an execution engine. Point it at a fuzzy strategy and you get fuzzy posts at scale. Garbage in, garbage out.
- Your rules are written down once. Your voice, your audience, your pain points, your do-nots. The system reads these every time so you do not have to repeat yourself.
- A human approval gate. Keep a person in the loop on every post. That is what protects your reputation and keeps the work honest.
Start small. One pillar, one posting day, one approval step. Let it run for a few weeks and read the numbers honestly – including the median, not just your best post.
If you would rather have us facilitate the Strategy and run it with you, you can work with us to scale your SME.
A system did not make us interesting. It made us consistent, and it found the moments worth amplifying. The Clarity was always ours to bring.
FAQ
Can AI really run a company’s LinkedIn content?
It can run the production – drafting, scheduling, news-scanning, and keeping cadence. Over 7 weeks it helped us post 4x more often and lifted total monthly reach about 25x. But a human approved every post, and the strategy was ours. AI executes a plan; it does not set one.
Do I need to know how to code to use Claude Code for marketing?
No. I am not a coder. The technical setup is one-time, and after that you work in plain language – your brand rules, your audience, your strategy, written down once. The system reads those instructions and produces drafts you review. The skill that matters is strategic clarity, not coding.
How much did this cost vs a marketing agency?
The honest answer for any SME: an AI system runs at a small fraction of a monthly agency retainer or a full-time hire, because the recurring cost is mostly software, not salaries. The bigger saving is keeping your voice and strategy in-house instead of outsourcing your own Clarity.
Will AI-written posts hurt my engagement?
They can, if they sound generic. Our flat median post proves bad AI content lands flat. But our per-post comments rose from 0.9 to 4.4 and reactions from 4.8 to 8.8. The difference was human voice and real stories – human-voiced content still out-engages content-mill output.
Can a Singapore SME with 20-100 employees do this?
Yes. We are one. The approach suits an established SME with a real strategy and a founder tired of being the marketing bottleneck. You need a clear point of view and a willingness to keep a human in the approval loop. The system handles the consistency you struggle to maintain manually.
How much time does an AI content system actually save?
Qualitatively, it removed me as the bottleneck. The grind that used to stall for weeks – drafting, scheduling, finding news angles – now moves daily without waiting on me. I have not put a verified hours-saved figure on it, so I will not invent one. The real win was consistency, not a stopwatch number.
