Stop guessing what creative testing framework to run.

Meta & TikTok · subscription apps, e-commerce, gaming · $40K+/mo accounts.

1. 3 distinct concepts — different hooks/angles, not 3 cuts of one ad.
2. 3 ad sets, identical targeting + budget ($50–150/day each), 1 ad each.
3. Run 72h, no edits.
4. Day 4: pause ad sets at CPA > 1.5× target; scale the winner.

Fast · cheap · forces 3 distinct concepts.

Noisy under $100/day. 3 days = before learning phase exits. Tests creative only, not audience fit.

3 versions of one ad isn't 3-3-3 — variation's too narrow and Andromeda picks the wrong winner.

1. Template — save a 3-3-3 structure: 3 ad sets, 1 ad each, identical targeting + budget.
2. Bulk launch — swap in 3 Drive creatives, launch all 3 ad sets in 2 clicks.
3. Auto-pause — Day 4, cut any ad set at CPA > 1.5× target.
4. Auto-promote — winner clones into your scaling campaign at 2× budget.
5. Slack — daily status + winner called on Day 4.

Shopping/Sales · broad audiences · post-Andromeda accounts.

1. 1 broad campaign, demographics-only targeting.
2. Single ad set, single objective.
3. Load 7–20 creatives — different formats and concepts.
4. Phase 1 — at 48h, pause any creative with CPA > 1.5× target.
5. Phase 2 — at 72h, scale survivors with CPA ≤ target (2× + add to BAU); pause the rest.

Algorithm-aligned. No manual audience-creative matching. Scales — more creatives = more pockets.

Needs 7–20 creatives. Hard to explain wins to stakeholders. Hard to attribute to specific choices.

Loading 20 near-identical creatives. Andromeda needs real diversity (format + concept + angle), not 20 carousels.

1. Template — broad targeting, single ad set + objective.
2. Bulk launch — drag the full batch; all creatives load into one ad set.
3. Phase 1 auto-pause — cut any creative with CPA > 1.5× target at 48h.
4. Phase 2 auto-promote — scale 2× + clone to every BAU campaign for survivors with CPA ≤ target at 72h.
5. Slack alert — new winner posted the moment a creative passes Phase 2.

Testing creative communication hypotheses · 3 angles × 2 variations · a primary + backup winner in 5 days · $40K+/mo accounts.

1. 3 ad sets × 2 creatives, 5-day run.
2. Identical audience + budget ($150–300/day each — higher budget compresses signal).
3. 2 creatives compete inside each ad set.
4. Day 5: pick winner per ad set by CPA; move top + #2 to scaling.

Faster than the 14-day version. In-ad-set variation suits Andromeda.

Tighter signal = more noise. Needs a daily budget of at least 3× your target CPA per ad set.

Throwing random ads into each ad set. Each ad set should test one hypothesis — e.g. same angle, different CTAs — so you learn why the winner won, not just which ad won.

1. Template — 3 ad sets, 2 ads each, $150–300/day, identical targeting.
2. Bulk launch — load 6 Drive creatives (2 per ad set) in 1 click.
3. Auto-pause — Day 3, cut the worst ad sets at CPA > 1.5× target.
4. Day 5 — Scalemate surfaces winner + #2 per ad set automatically.
5. Auto-promote — top winner + backup clone to scaling at 2–3× budget.

Video-heavy teams (mobile UA, DTC) · refining a proven concept · $5K+/mo accounts.

1. Take one winning video.
2. Make 3–5 variants — different first 3 seconds only.
3. Keep body + end identical.
4. Run in an Advantage+ CBO campaign — 1 ad set per hook, 1 ad each.
5. After 1K+ impressions/variant, cut bottom 50% by hook-rate.

Isolates the hook cleanly. Compounds a proven concept. Fast once you have a base winner.

Needs editing capacity. Refines, doesn't discover. Hook-rate ≠ ROAS.

Comparing hook rates across different campaigns instead of within one test — hook-rate is context-dependent.

1. Template — Advantage+ CBO campaign, 1 ad set per hook, 1 ad each, primary metric = hook rate.
2. Bulk launch — drag 5 Drive variants into the ad set, launch.
3. Auto-pause — cut any variant under control × 1.2 hook-rate after 2K impressions.
4. Winner — reporting tags the top hook for the next batch.
5. Next cycle — Slack alert when one variant clears 1.5× control to prompt iteration.

Enterprise w/ Meta rep · $75K+/mo accounts · validating a big change before scaling 2–3×.

1. Request a Lift study via your Meta rep.
2. Meta splits users into test (sees ads) vs hold-out (doesn't).
3. Run 4+ weeks for a valid sample.
4. Define the conversion event.
5. Meta reports incremental conversions + cost per incremental conversion.

Highest-quality incrementality Meta offers. Separates 'ad drove this' from 'would've converted anyway.' Ends attribution debates.

Gated — needs a Meta rep. 4+ week cycle. $30K+ for the study. One decision at a time, not a daily tool.

Assuming it's the same as the A/B Test option. A/B compares two live variants; Lift uses a true hold-out. A/B can't measure incrementality.

1. Pre-Lift — Scalemate locks current variants (no rotation during the 4-week window).
2. During — dashboard reports CPA + ROAS in parallel so finance sees both.
3. Post-Lift — flip 'scaling mode' once incrementality is confirmed.
4. Auto-scale — +20%/day when ROAS > target × 1.1, auto-revert on a 15% drop.
5. Slack — daily scaling-action summary for team + finance.

High-volume R&D teams · 30+ creatives/cycle · AI-generation pipelines · $10K+/mo accounts.

1. 1 CBO campaign, 4–5 ad sets (broad or diversified lookalikes).
2. 3–15 ads per ad set (12–75 total).
3. Phase 1 (3–5d): let CBO allocate; cut bottom 50–70% on upper-funnel signal (installs / registrations / add-to-cart).
4. Phase 2 (3–7d): keep top 20–40%; judge on lower-funnel (purchases / trials / paying users).
5. Move final 5–10 winners to scaling.

Tests volume efficiently (12–75 vs 9). 2-phase saves budget. Built for AI creative pipelines. Matches how CBO allocates.

Needs 12–75 creatives. Phase 1 metric choice is critical — wrong signal cuts real winners. 6–12 day cycle.

Treating Phase 1 metrics as final. High install rate ≠ high purchase rate. Phase 2 exists to weed those out — don't skip it.

1. Template — CBO campaign, 4–5 ad sets, single objective.
2. Bulk launch — drag 12–75 Drive creatives; Scalemate spreads 3–15 per ad set.
3. Phase 1 auto-pause (Day 3) — cut bottom 50–70% on upper-funnel events.
4. Phase 2 auto-pause (Day 5–8) — cut survivors failing lower-funnel CPA.
5. Auto-promote — final winners clone to scaling at 2–3×; batch tagged 'winner library.'

New accounts · format transitions · vertical-specific calls (mobile UA, eCom Reels).

1. 3 ad sets: static-only, video-only, mixed.
2. Same concept across all 3.
3. Same audience + budget ($100–200/day each).
4. Run 7–14 days.
5. Compare CPA, CTR, hook rate, watch-through by format.

Ends the 'video always wins' assumption. Data-backed format strategy.

Needs both formats produced. Result varies by funnel stage (TOFU = video, BOFU = static).

Calling video the winner on engagement when static had better CPA. Optimize for the business metric, not engagement.

1. Template — 3 ad sets (static / video / mixed), identical targeting + budget.
2. Bulk launch — drag 6 static + 6 video + 6 mixed; 6 per ad set in 1 click.
3. Auto-pause — cut anything under 0.5% CTR after 500 impressions.
4. Format reporting — dashboard shows aggregate CPA per format, not just per ad.
5. Auto-promote — winning format's top 2 clone to scaling; lock the format default.

Any team running winners 30+ days · critical post-Andromeda.

1. Track per-creative weekly: frequency, CPA, CTR, hook rate.
2. Replace when frequency > 3.0, CTR −20% from peak, CPA +30%, or hook-rate −15% (video).
3. Swap in a variation of the winner, not a brand-new concept.
4. Pause the fatigued creative immediately.

Stops budget waste on dying creative. Holds campaign CPA. Forces a refresh rhythm.

Needs a steady creative pipeline. Maintenance, not discovery.

Replacing with a new concept instead of a variation — fresh concepts need their own test; a variation inherits 70–80% of the winner.

1. Always-on monitor — watches every creative for fatigue triggers.
2. Slack — instant DM with creative ID + which metric broke.
3. Auto-pause — cuts the fatiguing creative the same hour.
4. Auto-swap — pulls the next variant from a tagged Drive queue. Zero campaign holes.
5. Weekly — which creatives fatigued, replacement performance, cycle time.

Teams with an established winner · iterating a replacement · app teams (IPM), web teams (CVR).

1. Use your proven winner as control.
2. 2 ad sets (identical audience) OR 1 ad set with 2 ads, standard delivery (not Dynamic Creative).
3. Run until both hit the same impressions (5K / 10K / 20K).
4. Pause both at target — don't let the leader run on.
5. Wait 24–48h for attribution, compare by IPM / CVR / lead CPA.

Cleanest head-to-head. Controls time-of-day, day-of-week, audience drift. Equal sample sizes.

Needs manual impression monitoring. Attribution lag delays the call. Not parallelizable (5 challengers = 5 runs).

Stopping one at target but letting the other run 'a bit longer' — re-introduces time bias. If one lags, raise its budget to catch up.

1. Template — 2 ad sets, or 1 ad set with 2 ads (no Dynamic Creative).
2. Equal-impressions auto-pause — both pause within minutes of the target, no manual watching.
3. Attribution lag — reporting holds 48h post-pause so late conversions count.
4. Decision metric — IPM (app) / CVR (web) / lead CPA per ad.
5. Winner — at ≥1.1× control, clone to BAU + auto-pause the old control.

Teams wanting native attribution + Meta-handled splitting + a clear verdict, no manual setup.

1. Ads Manager → Experiments → A/B Test.
2. Pick 2–4 ads / ad sets / campaigns.
3. Set duration (7–14d min) or a spend cap per variant.
4. Meta splits the audience — no overlap.
5. Judge by your chosen metric: hook rate, CTR, CVR, or CPA.

Meta handles the math. Auto-splits to remove overlap bias. Clear 'Variant X won, Y% confidence.'

7-day min even for small tests. Only 2–4 variants. Meta optimizes for confidence, not always your metric.

Setting the wrong primary metric — pick CTR and Meta crowns the high-CTR variant even if its CPA is worse. Set primary = the metric you'll act on.

1. Parallel monitor — Scalemate tracks CPA + ROAS + hook rate per variant alongside Meta.
2. Slack on verdict — Meta's call + Scalemate's full breakdown, side by side.
3. Disagreement check — flags when Meta's winner conflicts with your business KPI.
4. Auto-promote — if both agree, clone the winner to scaling.
5. Loser archive — losing variants tagged in Drive for future learning.

Mobile UA · $1K+/day per geo · high creative volume · proven cheap-geo→T1 correlation.

1. MAI campaign in T3 geos (Indonesia, Philippines, Brazil, Vietnam) or a WW MAI biased to cheap inventory.
2. Run creatives at 4–10× lower CPI than T1.
3. Judge by IPM + CPI.
4. Promote winners to a T1 AEO/ROAS campaign.
5. Small accounts (<$1–2K/day): move winners straight to BAU.

4–10× cheaper testing. More creatives per dollar. Fast install volume = fast signal.

Only works when cheap-geo winners correlate with T1. Strong for utilities/casual games, weak for premium/payment-heavy apps.

Promoting T3 winners to T1 without validating correlation. Some win cheap on low-intent T3 audiences and flop in T1. Validate the top 5 first.

1. Templates — T3/WW MAI test + T1 AEO/ROAS scaling, reusable.
2. Bulk launch in T3 — drag the batch; cheap impressions start flowing.
3. Auto-pause (Day 2) — cut creatives under IPM threshold; T3 signal is fast.
4. Correlation gate — one-time manual check: top 5 T3 winners vs T1 baseline.
5. Auto-promote — validated winners clone T3 → T1 with budget scaling.

Mobile UA w/ proven cheap-geo AEO correlation · enough conversion volume for AEO learning.

1. AEO campaign in T3 cheap geos.
2. Pick an AEO event tied to monetization (d3 retention, 1st purchase, level-5, 1st top-up).
3. Test creatives; score on AEO event rate, not just IPM.
4. Promote winners to a T1 AEO campaign, same event.

Cost savings + signal quality. AEO outcomes mean more than raw installs.

AEO needs volume — under ~50 events/week the learning phase never closes. Validate event throughput first.

Running AEO on thin volume. ~50 events/week minimum; at 10/week you get random delivery, not optimization.

1. Template — cheap-geo AEO campaign tied to your monetization event.
2. Bulk launch — load the batch into the AEO campaign in cheap geos.
3. Event-volume safety — auto-pause the campaign if AEO events < 50/week (switch to MAI Method 11).
4. Auto-pause — cut creatives under event-rate threshold after 5–7 days.
5. Auto-promote — validated winners clone to T1 AEO, same objective.

Mature mobile UA where cheap-geo correlation failed · creatives that must work in production.

1. Duplicate your BAU campaign exactly — optimization, audience, placements, geos.
2. Swap in the new creatives only.
3. Run 5–7 days alongside BAU.
4. Compare by BAU metric (IPM, CPI, AEO event rate, ROAS).
5. Winners replace fatiguing BAU creatives.

Most reliable signal of any UA method. Winners behave the same when scaled. No correlation gap.

Priciest per creative — full T1/BAU CPI. Not for high-volume iteration (cost scales linearly).

Testing too many at once dilutes BAU and spreads spend thin. Limit to 2–4 per mirror cycle.

1. One-click BAU clone — exact optimization, audience, placements, geos. No config drift.
2. Bulk swap — drag 2–4 new Drive creatives, replace only the creatives.
3. Auto-pause (Day 3) — cut new creatives under BAU baseline × 1.3 CPI.
4. Day 5–7 — per-creative CPA + AEO rate vs BAU baseline.
5. Auto-replace — winners clone into BAU; fatigued BAU creative auto-paused.

Teams wanting isolated $-gated reads · clean per-variant signal · no in-adset competition.

1. 1 CBO campaign · 1 ad set per variant · 1 ad each.
2. Phase 1 — at $60 ad-set spend, pause if CPI > 1.5× target.
3. Phase 2 — at $150 ad-set spend, pause if CPA > 1.5× target.
4. Survivors: scale 1.5× budget + clone as new ad into every active BAU campaign.

Clean per-variant read — no ads competing inside an ad set. Spend-based gates adapt to account pace. Survivors land in BAU as proven, not theoretical.

More ad sets = more learning phases to feed. CBO may underspend low-CTR variants before they hit Phase 1 gate. Fights Andromeda's 'feed many in one' instinct.

Setting CPI/CPA targets too tight. Use the BAU running average × 1.5, not an aspirational target — otherwise you cut variants that just needed a wider audience pocket.

1. Template — CBO campaign, 1 ad set per variant, 1 ad each.
2. Bulk launch — drag N creatives from Drive; Scalemate auto-creates N ad sets + 1 ad per set.
3. Phase 1 auto-pause — cut when adset spend ≥ $60 AND CPI > 1.5× target.
4. Phase 2 auto-pause — cut when adset spend ≥ $150 AND CPA > 1.5× target.
5. Auto-promote — survivor clones at 1.5× budget + adds as new ad to every active BAU campaign.