THESIS: There are two markets for attention, and they price in opposite directions. The auction (Meta, Google, TikTok Ads) prices attention by competition: every new brand in your niche bids against you, so your CPM only goes up — that's not a bug, it's the business model. The algorithm (TikTok FYP, Reels, Shorts organic) prices attention by content quality: it delivers your video to your exact buyer for free if the content earns the watch time — and gets cheaper as you scale, because volume buys learning and your library keeps working after you stop paying for it. Paid ads scale spend. The organic engine scales attention. The entire system below is one arbitrage: moving your demand generation from the market where price rises forever to the market where it falls.
CONTEXT: The honest numbers on both sides. The auction: Meta's median CPM sits around $13–14 across industries, with competitive eCom niches, retargeting, and Q4 auctions pushing $30–40+, median eCom CPA ~$30 — and the trendline points one way, because auction pricing is a function of competitor count, not your skill. The algorithm: at full volume — a creator network shipping ~1,000 native clips a week — organic reach stabilizes at $0.03 CPM in program math (all-in creator + ops cost over delivered impressions). A ~1,000× spread. Across 20+ brand programs: billions of organic views, CPA falling as volume grows, $0 spent on paid scaling — with the strongest single-brand outcome adding $1M in monthly revenue after the first campaign. The catch, stated plainly: the auction takes your money and delivers reach tomorrow; the engine takes a system and delivers compounding reach in weeks. This doc is the system.
Four parts — the network, the algorithm, the compounding, the resilience — then the operating math.
═══
What it is: The engine's surface is not your brand account — it's a managed network of creator accounts posting native short-form about your product: real people, real use, their own handles, their own voices. The brand sits behind the network the way a supplier sits behind a store shelf.
Why it's the unit of scale: One brand account is one signal stream the algorithm can classify (and one ban away from zero — Part 4). Twenty creator accounts are twenty independent doors into the recommendation engine, each reaching a different pocket of buyers, each reading as a person rather than an advertiser. The network structure:
The cost side that makes $0.03 possible: creators at $100–500 per short-form video (or hybrid retainers), one ops person per ~50–80 creators, tooling in the hundreds — against millions of algorithm-delivered impressions once the network finds its formats. No media line item at all: the impressions are earned, not bought.
═══
What it is: The counterintuitive core: you don't target anyone. TikTok, Reels, and Shorts exist to match content to the people most likely to watch it through — and their matching, trained on billions of sessions, is better than any audience you could buy. Your "targeting spend" is replaced by making content the algorithm wants to deliver.
How the machine reads your content: every clip enters a test batch (~200–500 viewers); completion rate, saves, shares, and rewatches decide whether it escalates. The bars that matter in 2026: ~70% completion for viral distribution on TikTok, saves>likes as the value signal, DM-sends-per-reach as the top Reels expansion signal, ~70%+ retention on shopping Shorts. Clear the bars and the platform delivers your product video to exactly the users who watch product videos like yours to the end — which is a purchase-intent filter no interest-targeting matches. Roughly 40% of Shorts viewers are actively shopping-oriented; the algorithm finds them for you.
The operating discipline: hook in the first second (no logo, no title card), silent legibility (muted autoplay is the default), platform-native exports (never one file to all three — that costs 50–70% of reach on the two it wasn't cut for), and a weekly test grid: hook variants × creators × platforms, ranked on completion and conversions, winners templatized and flooded across the network. The algorithm is the media buyer; the test grid is how you brief it.
═══