Written by Bilal Ahmad and Claude. Yes, Claude co-wrote this one. It did most of the actual work on the project, so it earned the byline. I'm giving away the whole playbook here, so if it helps you, a repost is appreciated.
For months, almost every pitch ended the same way.
I'd show a brand what we could make for them. They'd lean in, a little impressed. Then someone would say the quiet part out loud. "It looks a bit AI." Meeting over. And honestly? I couldn't blame them. Everyone's seen the bad stuff by now. Google, Meta, OpenAI and the rest have poured billions into this, so your mum has already made an AI photo and your competitors have already flooded the feed with the soulless kind. The tech was never the question. its the trust thats in question rn. and There's real research showing AI content can make a brand feel less authentic, and a big chunk of people just don't trust an ad the second they smell a machine behind it.
So I stopped arguing. You can't win that one with a slide deck. We found one brand instead, did the whole thing for them, start to finish, and let the work talk.
This is exactly what we did. Every tool, every step, the part that you mostly not get to read on internet, and the part that finally clicked.
They're a real, growing company with a big catalog and products people genuinely love, and they were already going a little viral on their own. But they're in a competitive market, and so are we. The people we compete with read posts like this one. So I'll happily show you exactly how we work.
All it takes is a one cold DM to land
This brand was spending real money on Meta ads, and the ads had stopped working.
Not because the product was bad. The product was great. The problem is quieter than that, and it hits every brand eventually: ads wear out. People see the same thing a few times and stop noticing it.
And it's gotten brutal. Meta's newer algorithm burns through an audience about three times faster than it used to. An ad that once stayed fresh for three weeks now fades in five to seven days. Someone sees it four times and the clicks can drop by almost half. A brand spending fifty grand a month can quietly lose eight to fifteen thousand of it to stale creative, if they're refreshing by hand. The treadmill never stops. Most brands fall off it. This one had.
They'd tried to fix it already. An AI tool that made clean product shots nobody clicked on. Another that made videos with the same fake person everyone else was using. A freelancer who was good but slow. An agency that took two weeks to hand over a few files. A lot of money spent. Nothing moved. They were halfway to deciding AI just doesn't work for them.
That's when we outreached to this brand and purposed a brief plan if you want to have the Cold DM template + the plan we purposed comment "DM" will send over to you
And then my team, i and claude got to work. mentioning claude again and again is really annoying at this point i know but i really want people to understand that this is real and people are actually building multiple figure revenue agencies doing these.
The lesson we learned
I won't pretend it was smooth.
We said yes to a brutal deadline. Four days to rebuild the creative for a whole store, plus a new website, plus a first batch of ads ready to run. Most teams would ask for a month. We had four.
So we did the only thing you can do when the clock's against you. We barely stopped. We kept Claude running close to round the clock, taking turns to keep it moving, checking every photo, fixing the ones that weren't right, pushing the next batch through. We hit Claude's limits more than 3 times and just upgraded our plans on the spot to keep going. the pressure was immense we had to deliver not only these creatives but the whole creative store OPTIMIZED for CONVERSIONS and with AI crazy task but trust me AI have gotten so good that at this point everything is almost at autopilot and i think it will soon go to autopilot which will sadly make us operators or others obsolete....
Why we had to build our own engine that we recently launched to public too
Quick backstory, because it's the whole reason this worked.
A while back we kept hitting the same wall. The AI tools could make something that slop kinda, and no you cant offer slop to brands they will never pay serious money for this. A shot can be sharp and clean and still do nothing for a brand. Worse, the second you needed the same product to stay the same across a whole catalog, the tools fell apart. The bottle changed shape. The label drifted. By the tenth image it wasn't the same product anymore. That drift makes a catalog look fake, and fake is exactly what people scroll past, and thats expensive for brands.
So we built our own engine, eComrads, and trained the part that actually matters: a judgement layer we call AdDNA. and this is actually important and not another buzzword
We fed it millions of real ads that actually sold. Not award-winners. but Money makers. And we taught it to judge its own work the way a tough creative director would, scoring every output on whether it'll sell, and killing the ones that won't. It doesn't stop at making an image. It builds the ad, and it holds the product locked and identical across every shot, which is the one thing general tools can't do.
That judgement layer is the whole game. It's why the engine gets it right the first time instead of handing you a pile of near-misses to sort through. So when it says a creative is good, it's good. It's been measured against thousands that already worked.
How we actually did it
This is the part I'm proudest of, and this is first time we are making something like this of us public so i'll try tp keep it simple.
eComrads plugs into Claude through something called MCP. Don't worry about the name. All it means is you can connect a tool to Claude with one link, no code and no passwords, and after that Claude can use it like it's holding it in its hand.
And Claude's gotten frighteningly good. It works on its own for hours now, breaks a big job into pieces, runs them side by side. So we weren't babysitting every click. We pointed it, checked its work, let it run.
We plugged in two tools.
First, the brand's own store. There's a Shopify connection that lets Claude look right inside a shop, so we hooked it up and told Claude to walk the whole catalog. And it did, one product at a time. Every product, every color and size, what was in stock, what had sold out, which few items were the real bestsellers carrying the sales, which ones didn't have a single good photo yet. A few minutes later we had a clean map of the entire store. Two hundred and fifty products. A hundred and seventy-three of them live.
Claude here. The part Bilal makes sound easy, reading the whole store, is the part I actually love. I went product by product, and the messy bits are where it matters. A bestseller sitting there with one blurry photo. A sold-out item we shouldn't waste a shoot on. A human skims a catalog and gets bored by item thirty. I read every line of all two hundred and fifty without blinking. That map told us exactly what to shoot. It's what made the next four days fast.
Second tool: eComrads itself, plugged in the same simple way. Now Claude had both in its hands. One tool to read the store, one to make and judge the creative.
So we just talked to it. "Take these products and make the photoshoots." It went down the list, pulled each product, ran it through the engine, and AdDNA scored each result and kept the ones that would sell. "Now turn the best ones into ads." Done. "Now build the website." Built that too, same setup. The five of us steered it. Nobody jumped between ten apps or dragged files around. We talked to Claude. Claude read the store and used eComrads. The finished work came out the other side.
Four days later, here's what we handed over:
• 635 finished product photos, covering all 173 live products.
• A batch of ready-to-run Meta ads built from the best of them.
• A brand-new website, genuinely beautiful, built the same way.
All on-brand. All built to sell. All of it through Claude, eComrads, and the Shopify connection. Yes all with AI and they paid us $30k for this because brand knows how hard it is to find competence in market nowadays everyone is Expert so its hard to find the real ones..
The part I'm proudest of: we almost never hit redo
Most AI tools make you redo an image five or six times before one comes out right. You make it, it's wrong, you try again. That waste piles up fast, in money and in hours.
Because of AdDNA, our engine got it right about 95% of the time. First try.
So instead of burning 50,000 credits to get those 635 photos, we used about 13,000. Same photos. Roughly a quarter of the cost, a sliver of the time. The machine wasn't sitting there guessing all day. Neither were we.
Claude here. This is the difference nobody talks about. I can generate images all night, but if four out of five come back wrong, you've just moved the bottleneck from a designer to me. The whole thing only works because the engine knows what good looks like and won't pass me junk. That's what let five tired humans keep up with me for four days.
And it doesn't just make photos
One more thing, because it matters for what you're about to see. The engine doesn't stop at still product shots. It makes the whole campaigns ready to make live on meta, in whatever format fits the brand.
some examples: Every clip below was made by eComrads in a single try with not very complex prompts:
Street interview. Podcast. Keynote. Testimonial. Surreal scroll-stopper. Different faces, different settings, different moods. Every one of them AI. That's the part that breaks people's brains, and it's exactly why the old "looks fake" reflex is already out of date - you can make same outputs at ecomrads.com or use it via Claude MCP ecomrads.com/mcp
PRO TIP: open claude code connect ecomrads mcp and ask it for the keywords given above and make same for your brand simply connect ecomrads mcp and give this prompt to claude "Make the street interview ugc for my product" upload your product image thats it...we made it ready made templates for anyone looking to speed up and a lot more you can do with eComrads so your outputs doesnt look like slop in first try.
So why pay brand paid us, when the tools are right there?
This is the real question, and I want to answer it straight, because it's the thing everyone gets wrong.
Yes, the tools are public. Anyone can sign up. So why did a serious brand pay us thirty grand to do something they could, in theory, do themselves?
Because typing "make a product photo" gets you garbage, and they know it. Output that's actually on-brand, actually consistent, actually real instead of AI-looking, isn't a prompt. It's a craft. And at catalog scale it's a punishing one.
Here's a number that says it all. To keep a single product looking the same across a catalog with a general AI tool, people burn one hundred to two hundred hours of prompt-wrangling for just fifty products. Fifty. And to do one product right, you're not writing a sentence. You're writing closer to three thousand words, locking the shape, the material, the camera, the lighting, the shadows, the background, the color, and every detail that has to stay identical shot to shot, or the product morphs and the whole thing looks fake.
Now do that for a 173 products. On-brand. In four days.
operators sitting down to write three-thousand-word prompts all day. And the few people genuinely great at this are rare and expensive, if you can find one at all. That gap, between "the tools exist" and "someone can actually make them sell," is the entire business. eComrads is that skill, packaged. The engine carrying the taste.
The math, since I'm sharing everything
Here's the part people always want and rarely get. What it cost us, and what we made.
The AI for that four-day build cost us under $750. About $450 in eComrads credits to generate everything, a few hundred in Claude after we hit our limits and upgraded mid-sprint. That's the hard cost. All of it.
We charged $30,000.
Before anyone yells "markup," read the last section again. The thirty grand was never for $750 of compute. It was for knowing how to turn that compute into 635 ads that sell, on-brand, in four days, a skill we spent years and a mountain of failed images training into the engine. The pixels are cheap. The judgement isn't.
Then there's the part that compounds. The retainer costs us about $565 a month in AI to make roughly 300 fresh creatives. They pay us $5,000. That's about $4,400 in profit every month, from one client, for work that mostly runs itself.
Add up a year and this single brand is about $90,000 in revenue against roughly $7,500 in AI costs. One client. One cold DM.
How I know it worked
They paid us $30,00 dollars for that first build. One time. Then they asked us to stay.
Now they pay us $5,000 a month, on a retainer, to keep their Meta ad creative fresh. We make around 300 new creatives for them every month, product photoshoots, static ads, short videos, plus fresh shoots for every new featured product they drop. Same engine. Same setup. Same 95% first-try hit rate, which is the only reason that volume is even possible.
And I'll be honest about why it's working, because it matters. This brand is good on its own. They're already a bit viral, and they run real human creators alongside our AI work. We're not the whole engine. We're the part that keeps the creative tank full, cheaply and fast, so everything else can run.
And you're still early
Don't let the noise fool you into thinking you missed it. You didn't.
Everyone's talking about AI, sure. But talking and doing are miles apart, and the numbers prove it. About 89% of retailers say they've adopted AI. Only around 7% have actually scaled it. Only about a quarter of companies have AI properly wired into how they market, and the rest are poking at a dozen disconnected tools that don't talk to each other, don't hold a brand voice, never compound. Roughly a third of ads are made with AI today, climbing toward 40%. And the brands that got in early are already pulling around 40% more revenue than the ones who didn't.
Read that again. Almost everyone is dabbling. Almost nobody has a real system. That gap, between dabbling and a system that actually runs, is the whole opportunity. And it's wide open right now.
One honest warning: this won't stay this cheap
I want to be straight with you, because it's the most important part of this whole thing.
Right now, AI is the cheapest it'll ever be. Not because it's cheap to run. It's brutally expensive. The companies behind it lose money on almost every request. Anthropic, the company that makes Claude, reportedly pulled in around five billion dollars last year and spent ten. OpenAI is said to burn over a million dollars a day just keeping the lights on. They're selling it to us below cost on purpose, to win the market while it's still up for grabs.
That can't last, and the people who study this don't think it will. The honest read is prices climb in the next year or two, maybe thirty to fifty percent, maybe a lot more. Even Sam Altman, who runs OpenAI, recently called the rising cost a "huge issue." The cheap era is quietly closing.
So when I say move now, it's not a sales trick to be honest i dont have to sell you anything its still you who should move the needle and It's just math. The exact work we did for thirty grand, the work that costs us a few hundred in compute today, will cost real money once the subsidies dry up. The brands that build their creative engine now, while it's this cheap, lock in an edge the latecomers will pay full price to copy. Take advantage of it while it lasts. I mean that genuinely.
So try it for yourself
If you take one thing from this, take this. Go try eComrads with Claude on your own brand. Plug it in, point it at your store, see what comes out the other side. That's the whole pitch. We'd honestly rather you see it work than listen to us talk about it and start selling content to BRANDS with or without eComrads before you miss on this opportunity as well. and send that DM
And if you've got feedback, questions, anything at all about how we did this, comment under this post or DM me. I read everything, and I reply to everyone. We genuinely want the feedback, the kind words and the brutal ones, because that's exactly how we keep making the product better.
Thanks for reading if you made it till the end please dont just bookmark this take action use AI properly learn new things everyday this is the last wealth transfer happening around the world. take this seriously.
