Heeyong

[Heeyong @HeeyongCh]

it’s been 30 days since we launched the brand. some thoughts on where we’re at:

it was definitely a slower start than i wanted. took a while to get CAC down to a reasonable number where we could justify scaling even a little bit.

CPMs were extremely high (still sitting around $150–$200), but there’s not much we can do except just work through it. instead of juggling different angles and trying to guess which concept would win, i just made ads for all of them. high volume, high-intent, long-form video straight to the PDP.

switching our ad account structure also helped. we were putting out a lot of ads on a small starting budget, and a single CBO wasn't giving us enough data to get any real feedback loop. switched to ABO for more intentional testing.

lastly, AI in marketing. creative strategy and copywriting still need a human in the loop, but AI has massively sped up both my volume and quality. key takeaway here is that you need a repeatable system that can generate an ad end-to-end without human supervision.

we’re not first-order profitable (didn't expect to be), but we’re now profitable on CAC against a projected 3-month LTV, which was our main launch goal.

still focusing almost entirely on creatives and ads. looks like the next short-term bottleneck is going to be inventory. retention is one area that we haven’t dialed in, so that’s another thing we’ll be focusing more on.

[Heeyong @HeeyongCh]

@kamal_razzak true

[Heeyong @HeeyongCh]

@EcomKostnchko_ its an openAI model

[Heeyong @HeeyongCh]

It's concerning but expected, and I'm not sure if it's necessarily bad. We seem to be accumulating technical debt faster than we can recognize it, and accountability isn't evolving as quickly as capability - yet we're witnessing unprecedented acceleration in intelligence in these domains.

I could argue that AI-generated code could be another abstraction layer, like software engineers already rely on without fully understanding every detail (e.g., OS internals or library implementations). Difference is that traditional abstractions have documented interfaces, predictable behavior, and stable implementations.

[Heeyong @HeeyongCh]

@sxnchzw this is simply not true - it's the same underlying model, meaning core generation properties like movement, lighting, physics, and prompt adherence remain identical

[Heeyong @HeeyongCh]

a founder's "personal taste" is often one of the biggest liabilities in e-commerce. it definitely was for me.

show a heavily CRO’d 9-figure site to people with "good taste" and they’ll say it looks cheap or overdone. they’re judging a commercial engine like it’s an art gallery

hyper-minimalist, editorial sites only work when your audience is tiny and hyper-curated. the second you want real scale, you’re selling to a broad tam that doesn't want subtle - they want obvious, simple, and frictionless.

unlearning my own aesthetic bias and trusting the data behind "ugly" simplicity was one of the hardest - and most profitable - lessons i had to learn.

[Jacob @jforjacob]

See too many people justifying their lack of success by saying "I have a real brand though"

First off, you don't. None of us do

Secondly, how do you define a real brand?

Is it through repeat purchase rate? If so, the DR demon supp brands have you beat

Is it LTV? Again, DR demons have you beat

Revenue? DR demons again

Is it pretty images on a website? Everyone can have that with a few clicks

In my mind having a real brand is simply that you are associated with something and that association sits deep in the collective conscious of the consumer market

Again, none of us on here have this. I don't care how cool your brand guide is

Mcdonalds has this, Coca Cola has this, Rolex has this

The people I see coping about 'brand' are almost always just bad at ads and funnels and they need some way to make themselves feel better instead of just facing reality

Skill issue

[Heeyong @HeeyongCh]

openai released hundreds of mathematical results tackling problems that have challenged researchers for decades. meanwhile, ai still struggles to write a genuinely good ad

the contrast feels a bit absurd

i think it shows something about the difference between intelligence and human intuition. good copywriting isn't just about mastering language, but understanding the subtleties of how people perceive and respond to it. ai can model the patterns of good writing without fully grasping what makes it resonate.

my guess is that the next leap in creative ai won't come from scaling intelligence alone, but from training models on the consequences of language, not just language itself

[Heeyong @HeeyongCh]

@nicktheriot_ opus 5.5 felt the best for me, astra and sol couldn’t handle aggressive copy very well

[Heeyong @HeeyongCh]

starting a new brand and decided to skip the shopify app store entirely this time. no aftersell, no intelligems, and we even cut klaviyo.

the common advice is that this is a trap because maintenance will eat you alive. but that was true back when building and maintaining software took weeks of manual work.

with current models running against proper test setups, building native checkout extensibility, custom cart transforms, and our own event-driven email flows barely took an afternoon.

the upsides are hard to ignore.

mobile site speed doesn't get destroyed by third-party tracking scripts, which directly protects conversion rates on paid traffic.

you get raw first-party behavioral data piped straight to your own database instead of trapped across half a dozen closed dashboards.

and you stop paying an escalating tax on your total contact list or gross sales just for someone else's ui wrapper around basic logic.

people treat klaviyo like untouchable infrastructure, but for triggered flows, piping events through resend or postmark with proper dkim, dmarc, and instant bounce suppression is straightforward.

you only really need to worry once you're sending massive 50k+ subscriber marketing blasts at the exact same minute and need deliberate queue throttling.

for standard storefront cro and lifecycle flows, paying thousands a month for apps that are just thin layers over shopify and email apis feels completely obsolete.

[Heeyong @HeeyongCh]

@kes11av It would depend more on your LTV than AOV

[Heeyong @HeeyongCh]

@EcomKostnchko_ use claude/codex and connect to shopify via cli and handle it there

[Heeyong @HeeyongCh]

@codyplof i generate then with gpt image 2.5

[Heeyong @HeeyongCh]

@danrobertsecom @LukasPakter @cjfeth would love to watch the call

[Heeyong @HeeyongCh]

@markbuildsbrand Interesting take. What kind of software are you building?

[Heeyong @HeeyongCh]

@codyplof fixed, sorry