The whole idea in one line
Treat every campaign as an experiment with a control, not a slot machine you feed and hope. Six lessons. None of them cost anything.
Lesson 01
The black box cannot teach you anything.
The platform wants you to hand it your budget and let the optimizer decide. Sometimes that even works. The problem is you learn nothing: you cannot tell whether your win came from the offer, the creative, the audience, or luck. So you cannot repeat it.
Start with a controlled test you can read. Get the real numbers. Then hand the proven thing to the optimizer to scale. Trust the algorithm with what you have already verified, not with the question you are trying to answer.
Verify first. Scale second. Never the other way around.
Lesson 02
Always run a broad control.
Here is the test almost nobody runs. Alongside your clever, tightly targeted audience, run one broad audience as a control: minimal targeting, let the platform spread it wide. Same offer, same creative, same budget.
Now your clever targeting has to beat random. Half the time it does not, and that is the most useful thing you can learn, because it means you were paying a premium to feel smart. A control turns an opinion into a measurement.
Lesson 03
Creative is the new targeting.
The platform already knows who to show your ad to better than you do. What it does not know is which hook stops the thumb. So your job is volume: many angles, many openings, many formats, and let the data pick the two or three that work.
This is where AI changed the math for me. I can generate dozens of real variants (different hooks, images, even short videos) in the time it used to take to make one. I am the art director. The model is the studio. Then the test, not my taste, decides the winner.
One offer can carry twenty pieces of creative. Make them.
Lesson 04
Change one variable at a time.
If you swap the offer, the audience, and the creative all at once and the numbers move, you have no idea why. You just bought noise. Pick one thing to test per round. Hold the rest still. The boring discipline is the whole edge.
A clean sequence: round one tests the offer (which thing do people actually want), round two tests the audience against the broad control, round three tests creative inside the winner. Each round answers exactly one question.
Lesson 05
Read the number that pays you.
Likes, reach, and "engagement" are vanity. They feel good and mean little. The number that matters is cost per result: cost per lead, cost per sale, cost per booked call. That is the one you optimize, and the one you kill against.
Set your kill rules before you launch, while you are calm. Something like: if an ad spends a set amount with zero results, pause it. If cost per result runs past your ceiling, pause it. Decide the rules cold so you are not negotiating with a losing ad at midnight.
Lesson 06
The nurture is the conversion engine.
A lead from a paid ad is colder than someone who searched for you. They were not looking, they just stopped scrolling. So the ad is only the front door. What you do in the next few minutes, hours, and days is where the conversion actually happens: the fast reply, the helpful follow-up, the sequence that warms them up.
One tactical nugget while we are here: if you run ads for housing, jobs, or credit, the platform forces a Special Ad Category that strips age, gender, and tight-radius targeting (it is a fair-housing rule). Real estate agents trip on this constantly. Plan your targeting around it instead of fighting it.
The ad gets attention. The follow-up gets the customer.
The part the ad didn't show you
The playbook is free. The machine took months.
Everything above is yours, no catch. But reading a method and running it every week on real money are two different sports. The strategy is the cheap part. Doing all of it, fast, at volume, without lighting your budget on fire, is the part almost nobody pulls off. So here is a look behind the curtain.
That ad you clicked was not made by hand. It came out of a system we have been building for months, and the system is the real story:
Research
A team of AI agents
Swarms of agents pointed at the whole world of ad knowledge: podcasts transcribed, threads scraped, the people worth listening to ranked by their actual work instead of their follower count. Thousands of pages of distilled intelligence, on command.
Creative
A studio that compounds
Image and ad generation we've sharpened across dozens of builds, so every new campaign starts further ahead than the last one.
Voice
A voice engine, trained for months
Months of teaching the system one real brand's voice until the writing sounds like the brand, not a robot. The voice is a setting now, not a guess.
Build
A campaign that builds itself
Enough of Meta's plumbing wired up that the machine assembles the whole campaign (audiences, creative, kill rules, the lead form) and hands it over finished. One toggle and it's live.
Backstops
Guards on the money
Kill rules written before launch, so a losing ad pauses itself instead of quietly eating your budget at 2am.
Volume
Twenty pieces in two hours
Twenty real pieces of creative in a couple of hours, aimed at demographics most people never test. The volume Lesson 03 asks for becomes a Tuesday.
And here is the part that still gets me. Once the machine exists, the model barely matters, and neither does the brand. We build a voice engine for each business, fed on that company's own material (decades of records, old radio spots, the way they actually talk), and the same machine runs it. We have done it now across businesses that could not be more different: a free community, a $15 supplement, a $60 clothing drop, and a home-service company with a $20,000 average ticket. Online, on a shelf, on a job site. Same machine, a different voice every time.
That is the whole point. This adapts to whatever you are running, not just to us. The advantage was never one clever prompt. It was building the whole thing once, so it works no matter what you point it at.
So why hand all of this to you on a free page? Because hoarding it is exactly what the spam groups do, and we are tired of it. We would rather give it to a room full of people who actually want to build, and get sharper together. Every time we give it away, we get better at it.
Giving it away is the point.
New to all this, and you don't have the thing you'd run ads for yet? Start one door over, at the Founder Kit.
Why this is free
A lot of these groups got rough: spam, bad info, and people selling you what should be free. We wanted the opposite. A place to geek out about AI and ads, give away everything we figure out, and get sharper together. If you are the kind of person who actually runs the test, this is your room. Come build in the open with us.
