The Growth Den Lab · lesson one

Creative is
the targeting

See all those dots? Every single one is a person, with their own secret stuff they care about. Tap any dot to meet them. Then scroll — and watch what a good ad does to a crowd.

~2,000 people, each with hidden traits the interface can't see. Tap any dot to inspect one. Then scroll: why broad targeting + strong creative beats interest-stack micromanagement.

(This is a toy, not the real machine — Meta's actual computer is a zillion times messier. But the idea is the real idea. 😉) (Mechanism-level mental model, not a leaked spec — the internals are ML soup. But the behavior it predicts is the behavior practitioners observe. 😉)

01 · the old way

Drawing circles around people.Hand-drawn interest circles.

People used to choose who sees an ad by hand — drawing circles around labels: “parents”, “likes gadgets”, “ages 25–34”.

But watch the circles form. They scoop up loads of people who don't care at all… and miss the exact people who do — because people are so much more than their labels.

Interest stacks are proxies. “Parenting” catches the empty-nester who liked one reel; it misses the exhausted dad who never told the platform he's a dad.

Hand-drawn boxes optimize for what's legible to you — not for where the actual signal lives.

✗ 0 don't-care people caughtnon-fits captured ◌ 0 perfect people missedtrue fits missed

02 · the new way

The ad is a whistle.The creative broadcasts; delivery finds resonance.

Now watch this. A really specific ad — “POV: your kid actually brushes” — is like a whistle only certain people can hear.

You show it to everyone… and the right people come to it. You didn't find them. The ad did.

Delivery optimizes on response. Ship a specific concept broad, and early engagement teaches the system exactly who resonates.

Targeting didn't disappear — it moved into the ad. The creative emits the signal; the machine finds the resonance.

📣 this ad is out there right now: “POV: your kid actually brushes”broadcasting: “POV: your kid actually brushes”

03 · the bland tax

A whistle that makes no sound.Weak signal = expensive scatter.

But what if the ad could be for anybody? “The best toothbrush. Period.” Stock photo of a smiling family. “Trust us.”

Blow the whistle… nothing. Nobody gathers, because it wasn't for anyone. Now the computer has to guess — and guessing costs money.

“The best toothbrush. Period.” is a flat response surface: the system can't learn who resonates, because nobody particularly does.

So it buys scatter — broader exploration, longer learning, higher CPA. Bland isn't the safe choice; it's the most expensive brief in the building.

💸 cost per new customer💸 est. acquisition cost
$—

04 · the lever

Make more whistles, not more settings.More concepts, not more dials.

So the trick isn't fiddling with the settings. It's making more different whistles — one for the worried mom, one for the gadget nerd, one for the person whose dentist yelled at them.

Look: three different ads, three different crowds. Same product. Same settings.

The lever is creative volume and diversity: distinct hooks for distinct motivations, each recruiting its own pocket of the market you'd never have drawn a circle around.

Audience settings are one dial. Concepts are a keyboard. Three ads below, three cohorts — identical settings.

“POV: your kid…”parent hook “I tested 5 of these”reviewer hook “My dentist yelled at me”confession hook

05 · the floor is yours

Now you build the whistle.Now you build the signal.

Pick a hook, a look, a voice, and some proof. Watch who shows up. Then — this is the fun part — make it boring on purpose and watch everyone wander off.

Assemble a concept; the cohort assembles itself. Then select the generic chips and watch resolution collapse — that's your control group.

HOOK · the first words· the first 2 seconds

VISUAL · what you see· the frame

TONE · the voice· the register

PROOF · why believe you· the receipts

Who showed upAssembled cohort

pick some chips and see…select chips to resolve a cohort…

people gatheredcohort size—
how tight the huddle iscluster tightness
cost per new customerest. acquisition cost—

next lesson → The Auction, In Human Terms So the right people gathered… now who wins the seat? Balls, magnets and wind await.Your cohort assembled — now see how total value decides who actually wins the impression.