Right then. Let's start with the thing most of you will be tempted to do, then regret.
Angela Siljegovic ran an experiment a lot of operators talk themselves into. She excluded her Klaviyo customer list on Meta, logic being: cut the existing buyers out, force Meta to go find fresh ones, watch new customer acquisition cost drop. Tidy theory. It blew up in her face. NCAC didn't fall, it climbed significantly, and performance dropped across campaigns, not just the one she was poking at. She pulled the exclusions this week and NCAC is already drifting back down and stabilizing. Her honest question to the room: does excluding customers actually break Meta's ability to find the right buyers, even when finding new ones is the whole point?
Derek Rasmussen had the partial answer before she even asked it. His tip: at the Ad Set level, set your exclusion and targeting rules to "suggestions" only. That lets Advantage+ go hunt for you. Yes, it'll ignore your targeting to a degree, but it still treats it as a signal rather than a wall. The reframe here is simple: your customer list isn't a fence you build to keep Meta out. It's a hint you hand it to point it in a direction. Hard exclusions starve the algorithm of the very data it learns from. Angela's numbers are the receipt.
Now the one everybody's going to be buzzing about.
Molly brought Devyn Merklin of XScale onto the GrowthOS call Tuesday, and by Paul Kalnitzky's account the room went a bit quiet by the end. Here's what they've quietly built. An AI brain loaded with over 20 years of Smart Marketer IP: a thous-plus videos, internal SOPs, client call recordings, Slack conversations, the lot, and it keeps updating. Sitting on top is Gary, a super agent that drops a business health report in your Slack every morning, flags your biggest risks, surfaces your top revenue opportunities, then tells you what it wants to do and asks for your approval before doing it. Under Gary, each brand gets its own agent with sub-agents handling CRO, ads, email, and more.
Devyn ran it live on his own brand. He sent a voice message asking for a Memorial Day promo. The agent built the whole thing: strategy doc, discount code live in Shopify, email design, Klaviyo segments, copy, all drafted and waiting for his review. Then he let it run his Google Ads autonomously for a few days with guardrails, and it drove an 85% revenue increase in one day off budget moves he says he never would have made himself.
The tell, and the bit worth actually thinking about: the difference between this and just rigging up your own Claude setup is the brain. It's not generic AI knowledge scraped off the internet. It's what's working across SM clients right now, with every brand that plugs in making the whole thing smarter. Your data stays siloed, you control what gets shared. So it's collective intelligence without handing your numbers to the bloke next to you.
Sam Clark also followed up on last week's CRO system. They turned the CRO walkthrough into a step-by-step SOP so you can spin up a CRO Agent in your own brand. The template doc is posted in 01-announcement. Go grab it.
On the grimmer side of operations: fraudulent orders are spiking, and one member reckons AI is making it worse, not better. Shopify's built-in fraud prevention isn't cutting it for him and he's eyeing Signifyd. Another operator confirmed it's an industry-wide problem that's getting worse. The thread's still open if you've got war stories or a fix that actually holds.
Coming up: Tuesday 6/23 at 11AM ET, Ben Bennett pulls the curtain back on what's genuinely working with AI in creative systems, aimed at people who want a real creative process rather than treating AI as a one-off novelty. Thursday 6/25 at 12PM ET, Business Strategy Office Hours with John.
Bottom line is this: the algorithm wants your data, not your fences. Stop building walls and start feeding it.