Post-Andromeda, CREATIVE is the primary targeting lever — Meta reads your creative and finds the buyer. So targeting is about giving Meta room and clean signal, not micro-segmenting interests.
PROSPECTING: BROAD-FIRST - Default to broad (age/geo/language only) and let diverse creative sort the audience. Broad + good creative beats stacked interests in almost every mature account. - Use interest/LAL stacks only when broad genuinely won't deliver (tiny budget, very narrow product). Lookalikes: build from your BEST source (purchasers/high-LTV, not all leads); 1% is tightest, widen to 3-5% as you scale (horizontal axis in the scale skill). - Don't over-segment into many tiny ad sets — each needs ~50 conversions/week to leave learning; ten starved ad sets all stall forever. Fewer, broader, better-fed.
PERSONAS ARE ARCHETYPES, NOT DEMOGRAPHICS - Define a persona by PROBLEM / DESIRE / BUYING TRIGGER, not age/gender. "Busy professional 25-45" is dead on arrival; "a mom who's tried six supplements because her toddler won't eat vegetables" writes its own ad and implies its own audience. - Persona x Angle x Offer = a concept; change one variable and you've made a new audience. Targeting and creative are the same decision. - Seed personas from REAL customer context (reviews, sales calls, the winning-frameworks ICP signal via get_winning_frameworks) — don't one-shot them from an LLM (regresses to the mean), refine from evidence.
EXCLUSIONS & HYGIENE - Exclude existing customers from prospecting and from DPA/retargeting unless you specifically want winbacks — included customers inflate reported ROAS and waste prospecting budget. - Retargeting is a SEQUENCE, not a standalone win: warm audiences convert because TOF did the work; judge it on incremental lift, not last-click ROAS.
EXPAND by cloning a proven ad set into a new audience/geo/LAL tier with duplicate_adset (keep the winning creative), one new audience at a time so you can read each — this is the horizontal scaling axis.