Alvin Ding

i keep seeing the same pattern with performance marketing teams. they ship 30 ad variants. two weeks of work. three days later, 28 of them are dead. the problem isn't the creative quality. it's that nobody mapped the buyer's awareness stage before writing a single headline. here's what that means in practice: someone who doesn't know your category exists needs a completely different ad than someone comparing your product to competitors. obvious, right? but most teams write all their ads for the same stage - usually the one closest to purchase. the result: reach campaigns fail, prospecting burns budget, and the only ads that work are retargeting. the fix is simple but almost nobody does it: 1. map your audience by awareness stage before you brief creative 2. write different hooks and angles for each stage 3. let performance data tell you where to double down teams that adopt this approach see higher ad hit rates, creative fatigue improve, cpa lowered. creative quality drives 47% of ad-driven sales according to nielsen. most marketers estimate it at 20%. closing that gap starts with psychology, not production. good news: buying stage is built-in as foundational function in prism - the creative tool that i built and use to power ad testing for clients. and its free to use. try it for free.

here's a pattern i keep seeing with teams that scale creative production: they hit 100+ ad variants per week. output is great. but cpa starts climbing and nobody can figure out why. the culprit is almost always brand drift. when you're producing at volume, small inconsistencies compound. slightly wrong colors here. off-brand messaging there. different font weights across platforms. each individual ad looks fine. but collectively, your brand signal gets noisy. the algorithm can't learn what's working because the variants aren't controlled experiments - they're unintentional brand chaos. the fix is what i call the brand kit method: 1. lock your visual identity into production templates - colors, fonts, spacing, logo placement 2. create messaging guardrails with approved copy patterns and tone rules 3. build brand compliance into the system so it's automated, not manual teams using this approach see two things happen: first, production speeds up because there's less review friction. designers work within constraints instead of reinventing the brand on every brief. second, ad performance improves because every variant reinforces the same brand signals. your data gets cleaner. your testing gets more meaningful.

six months ago i replaced our entire two-week creative process with ai. production speed tripled. ad variation output went from 5 per sprint to 40+. cost per acquisition dropped 22% in month one. but i made two mistaken assumptions that were demoralizing to overcome. first, i assumed the tools were plug and play. they're not. roughly 90% of ai output was unusable - generic copy, uncanny visuals, text that didn't quite make sense. getting to usable output required serious prompt engineering and brand kit integration. this has been an ongoing struggle for 4 months. turns out nano-banana can get sloppy with iterating on images and its difficult to build a production-grade service without blowing up database costs. second, i thought ai could produce near-final ads. in reality it gives you a 60-70% starting point. you still need humans refining copy, fixing visual hierarchy, making sure the creative connects emotionally. teams that skip this step ship forgettable ads. i built an instant-feedback button into prism so that users could stay in the loop. see screenshot today, after 4 months, this thing is really quite usable - but it took a ton of iterating. if i started over today i'd build a three-tier system: tier 1 (60% of volume): full ai automation for direct-response ads, format adaptations, a/b variants. brand compliance automated. this is currently somewhat live - but a certain level of human control is still required to cull nonsensical copy / images. tier 2 (30%): ai-assisted production with human creative director reviewing every batch before launch. this is fully build and we're actively using it at rise. tier 3 (10%): human-led brand campaigns. ai handles research and production scaling after approval. this is not yet built - but also lowest priority as most our clients are DR and ROAS focused. the biggest mistake teams make is applying one workflow to all three tiers. direct-response ads don't need the same review rigor as a brand campaign. and brand campaigns can't be automated without losing the judgment that makes them effective.

spent the last 4 months building an ad platform that doesn't suck. put all of my 15+ years of growth marketing experience into it. have been using it to drive creative iteration on existing client accounts (meta). dogfooded the product. here are 2 examples of ads for itself. anyone can try it for free.

your creative team isn't slow. your process is. only 43% of content teams have standardized workflows. 52% of creatives experienced burnout last year, and most of it came from operational friction, not creative workload. the real bottleneck is everything surrounding the design work. vague briefs that create revision cycles. feedback scattered across slack, figma, and loom. approvals that sit untouched for days. and you can't hire your way out of this. a senior designer costs $120k+ fully loaded. onboarding takes months. and they'll hit the same friction points as your current team. the fix: standardize briefs, centralize feedback, add a strategy layer before production starts. costs a fraction of a new hire. impact is immediate. makes every existing team member more productive (leverage).

we took a dtc food brand from 1.2 to 1.9 new customer roas in four weeks. no video production. no copy-length experiments. no budget increase. just 10 static image variations. here's what happened: the brand was running five figures monthly on meta with the same three static images. frequency had crept above 4.0. creative refresh rate was zero. meta's own research shows that at four repeated exposures, conversion likelihood drops about 45%. so we built 10 new static variations across two format sizes: 4x5 for feed and 9x16 for stories/reels. each variation paired a distinct visual approach with a specific on-image text angle. the winner: a 4x5 product close-up with ingredient transparency messaging. bold on-image text calling out what's in the product. it tapped into the maha movement and the broader push toward healthier, real food. the losing concepts weren't bad. they were just vague. lifestyle shots that looked polished but didn't answer the buyer's actual question: "what's in this and is it worth the premium?" three takeaways: 1. creative fatigue hits faster than you think. at 4 exposures, you're losing nearly half your conversions. 2. format fit matters. 4x5 won in feed. 9x16 won in stories. produce both for every concept. 3. specificity beats polish every time. the ingredient transparency image wasn't the prettiest. it was the most clear. google's research says creative quality accounts for up to 70% of ad performance. targeting is largely solved. creative is the last real lever.

@eriktorenberg streaming what...their workday? doing deepwork and streaming and trying to be engaging are probably not compatible.

Why aren’t there more Twitch style streamers in tech? It feels like there’s white space in streaming for tech creators, especially relative to podcasts and newsletters which are more saturated. TBPN and a few others have shown there’s substantial demand for it when it’s good.

@CoryOnBrand i mean you can achieve that on meta too with the optimization setting changed to reach. not all impressions are built the same either. at the eod that meta ad is probably driving stronger attributable ROAS both have a time and place ;)

Your Meta prospecting is hitting a 1.8x individual frequency on a 7-day window Our CTV prospecting is hitting an even 1x household frequency on a 30-day window Not all prospecting is built the same 🔥

@rxhit05 poor execution. bad ideas can still pivot if executed rationally / well

What kills more startups: bad ideas or no execution?

@mdnlabs lol so true

Anyone gonna take the new AI courses from Anthropic? Or just make your AI watch it and make skills out of it?

so how many of us are here just waiting for claude to come back online? ✋🏼

i've been looking at the numbers on meta ad costs and they're telling. cpms averaged $10.88 in q1 2025. that's a 19% jump year-over-year. some months showed 25%+ increases. and they're not coming back down. here's what most people are getting wrong. they're trying to out-target their way out of rising costs. but meta's algorithm has gotten so good that targeting isn't a competitive advantage anymore. it's table stakes. the teams that are actually winning? they figured out that creative is the only lever left. 3 things they're doing differently: 1. they collapsed the gap between strategy and production. no more 47-slide briefs that take 2 weeks to execute 2. they're testing at 10x the volume because the cost of producing each variation dropped to nearly zero 3. they treat brand consistency as infrastructure, not a bottleneck. guardrails built into the system, not the review cycle creative production is becoming more like software engineering. automated where it can be. systematic where it matters. human where it counts.

→ try useprism.io to see the system in action.

creative review cycles are where velocity goes to die. 3 rounds of approval = 3–5 extra days per asset. totally get that protecting brand is important - but i argue that you can have your cake and eat it too...and im talking about even with AI (i know it can be a slop-fest) with prism: → load in your PDP → load in ss of your product, font examples → press a few buttons to select copy angles + visual templates + photography style → let our system combine everything in a way that doesn't suck → ???? → profit (assuming you have PMF and your site CR isn't terrible) we built this to be rocket fuel for marketers and creative teams. let producers ideate fast without going off-brand.

@mdnlabs @jackfriks I had to do this twice. Brutal. I almost cried. Make backups on git and disk backups

@jackfriks Happens so often lol. Yesterday I completely reinstalled openclaw from things breaking lol

openclaw is great but i feel like the first 2 days were the peak for me so far

@jackfriks lol talking to openclaw is like talking to a lying amnesiac

@JKClayton_FN @MitcheIl An EMP

@MitcheIl someone needs to make an ai tool that will stop ai from replacing us

made this launch video with just 1 prompt AI is gonna replace us its so over

@jameygannon @m_g_nichols Taste agents sounds like an oxymoron Then again there are lots of tasteless humans

@m_g_nichols ooooo

who’s gonna be the first company to hire a Head of Taste

@JamilWanders @jspeiser In that case we’re all bottlenecks

@jspeiser if it's taking you 2 hours you might be the bottleneck

Nothing like spending 2 hours to get/train openclaw to automate something that takes me 5 minutes to do once a week… 🤷