Lessons behind RM700k per month ad spend

I was talking to Jon Lai Cheng Seong, of Atomic Group, who manages brands like Nattome, Heartio, and The Perfect Derma, trying to learn how his AI systems were set up.

"How much do you spend on ads?"

"RM600-700k"

"Cool, so that's about RM50k per month".

"No, monthly"

"Oh wow, what? And how big is your media buying team?

1 guy.

Live Masterclass: Leveragr neuroscience to build trust

Our weekly live masterclass is back. This week, Steve Wee joins us to explain how to use neuroscience during sales calls to build trust and close deals.

I can relate to this. When I invite speakers and brands to participate in Underdog Con, nobody just says yes. Speakers weigh their time. Exhibitors don't just hand over sponsorship money. Most of these decisions are emotional. And neuroscience can explain why.

If your job involves convincing someone, this is for you.

managingJoin the live masterclass free here

In the past, managing even RM10,000 a month in ad spend required a whole media buying team. Today, Jon manages RM700k a month across six brands with just one person and a Telegram bot he named Jarvis.

Here's what I learnt from our conversation.

Lesson 1: Build a Company Brain

One of the biggest wastes in business is forgotten knowledge. A team has a meeting, makes decisions, and just a week later, nobody can recall what was decided.

Same with marketing campaigns. A marketing campaign runs, results come in, and the lesson disappears with whoever ran it.

Atomic Group uses Notion as a single source of truth. Everything from customer feedback, product knowledge, meetings, to ad campaign data – all of it lives in Notion.

Then, using Notion MCP, Notion is connected to Claude. The team can query their own "company brain" for new launches, past decisions, and build AI systems trained on their own knowledge.

Lesson 2: Find your customers' pain & frustrations

By the time Jon launched his brands in 2021, the market was already crowded. But he noticed something: most brands were listing products. Very few were educating their customers.

He dug into Shopee reviews and Reddit forums. What were the fears, frustrations, and exact words customers used?

Then he created educational carousels using their own language, not marketing speak.

This is one reason I love direct-to-consumer (DTC) brands. Good marketing starts with understanding the customer.

Lesson 3: AI agents amplify you

If your process is broken, AI will amplify a broken process. But if you have a proven process, AI agents can help you scale it.

As Atomic Group grew, complexity followed. Managing hundreds of ad sets across multiple accounts became unmanageable without a big team.

Jon started experimenting with AI systems early. And his media buying agent was built by a media buyer, not a techie. He was problem-first, not app-first.

Here's what the agent does:

• Ad analyzer: Every morning at 9 AM, it scrapes paid ad data and sends a report to Telegram as an HTML file.
• Ad Recommendation: It analyses CPL, frequency, and a 7-day performance window to flag whether to pause or increase budget.
• Ad Execution: Braden, their media buyer, no longer opens ads manager. He just clicks "OK" and the agent makes the change.

Cost to run this? RM40 a month.

But the agent isn't the point. The point is that it's trained on Atomic Group's own processes. Before you build an AI agent, make sure you have a proven process worth amplifying.

Lesson 4: Don't stop talking to customers

I was surprised to learn that over 90% of Atomic Group's ads are optimized for leads, not conversions.

Their ads are built to start conversations, not close sales immediately.

This struck me as interesting, as most e-Commerce brands run straight to sales objectives. But what Jon found is that a proper customer support conversation leads to larger order sizes. Especially, after understanding the prospect's problem before recommending anything.

Their customer support channels are powered by AI agents that handle repetitive messages. Real humans step in when high purchase intent is detected.

The human work always comes first. AI comes into scale.

WhatsApp automation but human

One thing that amuses me is how much marketers want to AI everything. Yet, when it's them as a customer, they want to talk to humans.

The solution is not to make your AI chatbot impersonate a human. The solution is to be transparent about where AI is in your customer flow, and where it's not.

OtoWA is a WhatsApp broadcast and automation tool I recommend for businesses that use WhatsApp a lot.

Write "Daily CMO" in the referred by field to get bonus credits in your trial account.