campaign-optimization

# Campaign Optimization System Systematic methodology for optimizing paid acquisition campaigns from launch through scale. Covers the full lifecycle: learning phase management, creative testing, budget allocation, and scaling. ## Week 1-2 Optimization Playbook ### Days 1-3: Learning Phase (DO NOT TOUCH) Meta's algorithm needs 50 optimization events per ad set per week to exit the learning phase. During days 1-3: - **Do NOT** change budgets, targeting, or creative - **Do NOT** kill underperforming ads (insufficient data) - **Do NOT** duplicate ad sets - **DO** monitor delivery status (should show "Learning") - **DO** verify pixel events are firing correctly - **DO** check that all creative variations are serving impressions Minimum thresholds before making any decision: | Metric | Minimum data needed | |--------|-------------------| | Impressions per ad | 1,000 | | Clicks per ad | 50 | | Conversions per ad set | 15 | | Spend per ad set | 2x your target CPA | ### Day 4-5: First Read Review performance but only make minor adjustments: - Pause any ad with 0 clicks after 500+ impressions (creative is broken, not underperforming) - Check frequency -- if above 2.0 already, audience is too small - Verify landing page load time (>3s kills conversion rates) ### Day 7: First Decision Point After 7 days with sufficient data: 1. **Kill losers**: Pause ad sets with CPA > 2x your target after spending 3x target CPA 2. **Keep runners**: Ad sets within 1-2x target CPA continue for another week 3. **Note winners**: Ad sets below target CPA are candidates for scaling ### Day 10-14: Optimize Winners - Increase budget on winning ad sets by 20% (never more than 20% at once) - Create new ad sets with winning creative + new audiences - Start testing new creative variations against the control - Build lookalike audiences from converters if you have 100+ conversion events ## Key Performance Metrics ### Primary Metrics (Optimize For) | Metric | Formula | Good benchmark (EdTech LATAM) | |--------|---------|-------------------------------| | CPA (Cost Per Acquisition) | Spend / Conversions | $5-25 (lead), $50-200 (paid user) | | ROAS (Return on Ad Spend) | Revenue / Spend | 3x+ for sustainable growth | | CAC (Customer Acquisition Cost) | Total marketing cost / New customers | < 1/3 of LTV | ### Secondary Metrics (Diagnose With) | Metric | What it tells you | Alarm threshold | |--------|-------------------|-----------------| | CTR (Click-Through Rate) | Creative resonance | < 1% = creative problem | | CPC (Cost Per Click) | Audience + creative quality | > $2 LATAM = targeting issue | | CVR (Conversion Rate) | Landing page effectiveness | < 2% = page problem | | CPM (Cost Per 1000 Impressions) | Auction competitiveness | > $15 LATAM = audience too narrow | | Frequency | Ad fatigue signal | > 3 in 7 days = rotate creative | | Hook Rate (video) | First 3s retention | < 25% = weak hook | | Hold Rate (video) | Full video views / impressions | < 5% = content issue | | Thumb-Stop Ratio | 3s views / impressions | < 20% = visual not stopping scroll | ## Creative Testing Methodology ### Testing Hierarchy Test in this order (highest impact first): 1. **Angle/Message** -- What pain point or desire are you leading with? 2. **Format** -- Image vs video vs carousel 3. **Hook** -- First line of text, first 3 seconds of video 4. **Visual** -- Colors, composition, faces vs no faces 5. **Copy length** -- Short vs medium vs long primary text 6. **CTA** -- SIGN_UP vs LEARN_MORE vs different CTA text ### Statistical Significance Never kill a creative without sufficient data: | Daily budget per ad set | Min days before decision | Min conversions | |------------------------|------------------------|-----------------| | $20-50 | 5-7 days | 15 per variant | | $50-100 | 3-5 days | 25 per variant | | $100+ | 2-3 days | 50 per variant | ### Winner Criteria A creative is a "winner" when: - CPA is at or below target for 3+ consecutive days - At least 30 conversions tracked - CTR is above 1.5% (shows broad appeal, not just niche) - Frequency is below 3 (not just retargeting effects) ### Iteration Process 1. Run 3-5 creative variants per ad set 2. After 7 days, identify the winner 3. Create 3 new variations inspired by the winner (same angle, different execution) 4. Replace the losers with the new variations 5. Repeat weekly -- creative fatigue is the #1 scaling killer ## Budget Allocation Framework ### The 70/20/10 Rule | Allocation | Purpose | What goes here | |------------|---------|----------------| | 70% | Proven performers | Ad sets with CPA below target, scaling | | 20% | Testing | New audiences, new creative, new angles | | 10% | Moonshots | Completely new approaches, new platforms | ### Budget Adjustment Rules - **Never increase budget by more than 20% per day** -- resets learning phase - **Never decrease budget by more than 50%** -- better to pause entirely - **Minimum daily budget per ad set**: 2x your target CPA (so Meta can optimize) - **If CPA spikes after budget increase**: wait 48 hours before reacting - **If CPA doesn't recover after 72 hours**: roll back budget to previous level ## Scaling Stages ### Stage 1: Validation ($20-50/day per ad set) Goal: Find winning creative + audience combinations. - 2-3 ad sets with different audiences - 3-5 creative variants per ad set - Success: at least 1 ad set with CPA at target after 7 days ### Stage 2: Optimization ($50-100/day per ad set) Goal: Optimize and expand what works. - Scale winning ad sets by 20% every 3 days - Add 1-2 new audiences per week - Introduce lookalike audiences (1%, 2%, 3%) - Success: 3+ ad sets with consistent CPA ### Stage 3: Growth ($100-500/day per ad set) Goal: Maximize volume while maintaining efficiency. - Switch to CBO for budget distribution - Expand to broader audiences (reduce targeting specificity) - Launch in new geos (test 1 country at a time) - Success: predictable daily conversion volume ### Stage 4: Scale ($500-1000+/day per ad set) Goal: Dominate your category. - Multiple campaigns by objective (conversion + retargeting + awareness) - Full-funnel approach (TOFU awareness, MOFU consideration, BOFU conversion) - Creative team producing 5-10 new variants per week - Risk: CPAs increase 20-40% at scale -- factor this into unit economics - Success: sustainable ROAS at high volume ## Common Pitfalls 1. **Touching campaigns during learning phase** -- Resets the algorithm, wastes spend 2. **Scaling too fast** -- 2x budget overnight kills performance 3. **Creative fatigue denial** -- If CTR drops 30%+, the creative is tired no matter how good it was 4. **Audience overlap** -- Multiple ad sets competing for the same users inflates CPM 5. **Ignoring frequency** -- Frequency > 5 = you're annoying people, not persuading them 6. **Optimizing for clicks instead of conversions** -- High CTR with low CVR = wrong audience 7. **No exclusions** -- Always exclude existing customers from acquisition campaigns 8. **Weekend panic** -- B2B campaigns naturally dip on weekends, don't panic-pause Friday afternoon ## Patience Paradox (April 2026) The single most important optimization rule in the Andromeda era: **when profitable, DO NOTHING.** ### The Rule If CPA <= 1.2x target for 5 straight days at $500+/day spend: **lock the ad set for 14 days.** During the lock period: - **No new creatives** added to the ad set - **No bid changes** - **No budget changes** (except scheduled increases already planned) - **No audience adjustments** - **No pausing individual ads** within the set ### Why This Works Andromeda's pattern recognition builds a compounding model of your ideal customer. Every edit -- even a "minor" one -- resets this model. Frequent edits interrupt the algorithm's ability to identify and expand high-value audience pockets. The top $100M+/year advertisers are patient. They let winning ad sets run undisturbed while concentrating all iteration in separate testing campaigns. The scaling campaign is a "set it and forget it" machine. ### When to Break the Lock Only break the 14-day lock if: - CPA exceeds 2x target for 3 consecutive days (not just 1 bad day) - Delivery drops to near-zero (algorithm has exhausted the audience) - External event changes the market (competitor launches, news cycle, etc.) ## Creative Fatigue Detection (April 2026) Meta introduced two new delivery statuses that provide early warning of creative exhaustion: ### "Creative Limited" (Pre-Launch) - Andromeda detected visual similarity to existing ads in your account - Delivery will be **reduced before the ad even launches** - Action: Replace with a genuinely different visual. Text changes alone will not clear this flag. ### "Creative Fatigue" (Post-Launch) - The audience pool for this visual is exhausted - All exposures across your entire Page are counted (including organic posts) - Action: Retire the creative to the catch-all campaign. Create new visuals. ### Monitoring Cadence - Check creative status **daily** during the first 7 days of any new creative - Check **weekly** for all active creatives in scaling campaigns - If 3+ creatives are flagged "Creative Limited" in a single week, your creative production pipeline is not diverse enough ## EMQ Monitoring (Weekly Audit) Add Event Match Quality to your weekly optimization audit alongside CPA, ROAS, and creative health. ### The EMQ Check 1. Open Events Manager > Data Sources > Your Pixel 2. Check EMQ score for your top conversion events (Purchase, Lead, CompleteRegistration) 3. Goal: EMQ 8+ on 80%+ of events ### EMQ Red Flags - **EMQ <7 on >20% of events:** ARM is crippled. Fix CAPI immediately before any other optimization. - **EMQ declining week over week:** Something changed in your tracking stack. Investigate. - **EMQ varies by event type:** Common when Purchase events have better data than Lead events (different pages capture different parameters). ### CAPI Fixes (Priority Order) 1. Deploy server-side CAPI container (Stape or Weld) 2. Ensure `external_id`, `em`, `ph`, `fbp`/`fbc`, `ct`, `client_user_agent` are sent with every event 3. Implement redundant event deduplication (event_id matching between pixel + CAPI) 4. Test with Meta Event Debugger on live production traffic ## Blended MER (Marketing Efficiency Ratio) In the post-attribution-upheaval world (March 2026), platform-reported ROAS is unreliable. Use blended MER as your source of truth. ### Formula ``` MER = Total Ad Spend (all platforms) / Total Revenue (all sources) ``` ### Why MER > Platform ROAS - Every platform over-claims credit (Meta, Google, TikTok all take credit for the same conversion) - MER doesn't care about attribution windows, click definitions, or view-through counting - MER measures what actually matters: is your total marketing investment generating profitable revenue? ### MER Benchmarks | MER | Interpretation | |-----|---------------| | < 0.15 | Excellent -- spending $0.15 to make $1.00 | | 0.15 - 0.25 | Good -- sustainable for most margins | | 0.25 - 0.40 | Acceptable -- watch unit economics closely | | > 0.40 | Danger -- spending $0.40+ to make $1.00, likely unprofitable | ### Weekly MER Tracking 1. Sum ALL ad spend across Meta, Google, TikTok, LinkedIn, email tools 2. Sum ALL revenue from Stripe (or payment processor) 3. Calculate MER 4. Compare week-over-week. Trending down = improving efficiency. Trending up = investigate. ## Budget Scaling (Updated April 2026) Andromeda-era budget scaling is more conservative than pre-Andromeda. ### Rules - **Never increase budget by more than 20% in a single edit** (down from the previous "soft" 20% guideline -- now a hard rule because Andromeda's model is more sensitive to budget changes) - **Wait 3-4 days between increases** (not 2-3 days as before) - **Any increase larger than 20% triggers a learning phase reset** requiring 50 optimization events/week to exit - If CPA spikes after increase, wait 72 hours (not 48) before reacting ### Scaling Sequence ``` Day 1: CPA at target, spending $100/day Day 5: CPA still at target → increase to $120/day (+20%) Day 9: CPA still at target → increase to $144/day (+20%) Day 13: CPA still at target → increase to $173/day (+20%) Day 17: CPA still at target → increase to $207/day (+20%) ``` At this rate, you 2x budget in ~3 weeks without ever triggering a learning phase reset. Patience compounds. ### When NOT to Scale - CPA has been above target for any 2 of the last 5 days - Creative Fatigue flags are appearing on your top performers - EMQ has dropped below 7 - You haven't refreshed creative in 3+ weeks