Scaling kills more winners than it makes — the failure loop is scale fast -> learning resets -> CPA spikes -> panic kill. Gate first, then scale the SLOW way.
ELIGIBILITY (all must pass — if any fails, stop and route to the diagnose skill, hold 48h, re-check) 1) Out of learning: ~50+ conversions in the trailing 7 days. Scaling mid-learning restarts Meta's optimization clock. (AutoAdy skips learning-phase ads in autopilot for exactly this reason.) 2) Performance stable, not a one-day fluke: CPL < 0.7x account avg with ≥3 leads (or ≥5 leads at CPL < avg), ideally held ~3 consecutive days. These are the same thresholds the audit flags as "scale". 3) Headroom, not saturation: frequency < 2.5 on prospecting (< ~6 retargeting). Rising frequency means you're paying more to re-hit the same people, not finding new ones. 4) Fresh signal: pixel/CAPI firing recently (within ~12h). Stale tracking invalidates the whole decision (rule it out via the audit/diagnose tracking checks). 5) Real new-customer acquisition: make sure repeat customers aren't masking weak prospecting before you pour budget in.
THE AXIS MATRIX (pick by situation) - VERTICAL (default, under ~$1k/day, headroom, stable CPA): raise the SAME ad set's daily budget. Use scale_budget_relative. Default +20% every 2-3 days — never a jump. - HORIZONTAL (when vertical inflates CPA >10% per step, or delivery flattens): duplicate the winner into a NEW audience / geo / LAL tier with duplicate_adset, keeping the winning creative. Adds spend without re-saturating one audience. - BID-CAP ("Tichenor", once Lowest Cost hits its ceiling with 50+ conv/week): duplicate with a bid cap at ~target_CPA x 1.20 for a portfolio of capped ad sets. - TARGET-ROAS ("Faris", winner stable 7+ days): duplicate into a Lowest-Cost-with-min-ROAS variant for a stability floor.
THE 20% RULE: raise daily budget ≤20% every 2-3 days; stage it ($100 -> 120 -> 144 -> 173) rather than doubling. (Autopilot enforces a similar per-step ceiling — don't force-override it.)
$1k -> $10k LADDER: Stage 1 ($1-2.5k) vertical 20% on confirmed winners, add 1-2 broader ad sets, lock 3-5 winning creatives, halt if CPA > 1.5x target. Stage 2 ($2.5-5k) horizontal LAL/geo + ASC alongside manual, weekly creative, track MER. Stage 3 ($5-10k) bid-cap + target-ROAS diversification, 4+ new concepts/week, layer brand at 10-15%. Expect CPA to inflate 10-20% during each transition — that is normal; the success test is marginal revenue > marginal spend, not flat CPA.