How to scale a campaign

The goal of scaling isn’t to increase budget. It’s to increase results while preserving efficiency. The right question isn’t “how much can I increase?” — it’s “how much can the algorithm grow while keeping the indicators healthy?”

Metrics before scaling — by phase

There’s no universal threshold — it depends on product margin and current phase.

Test phase ($10-20/day)

Don’t scale. You’re validating whether the angle works. Relevant metric: link CTR (above 1.2%, indicator of stopping the scroll) and CPC compatible with the niche. ROAS here is secondary because volume is small and variance is high.

Validation phase ($20-60/day)

ROAS starts to matter. More important: CPA stabilized over 5-7 days. If it varies more than 25% day to day, it’s not validated yet — you’re getting lucky or unlucky. Don’t scale.

Scaling phase ($60-400/day)

• ROAS above margin minimum (2.5x for average margin; 3x+ for tight; 2x for fat)
• Frequency below 2.5
• Stable CTR
• Stable CPA
• Link CTR and engagement CTR moving in the same direction — if one rises and the other falls, the algorithm is confused

Max phase ($400+/day)

Previous metrics + buyer quality. You need to look at LTV of recent buyers. If LTV is dropping (worse buyers), scaling more will destroy long-term ROAS. Today’s ROAS may look good but you’re getting buyers who won’t repurchase.

Data nobody watches: percentage of new vs repeat buyers in the campaign. When this changes significantly after scaling, the algorithm changed audience without you noticing.

Analysis period

Analysis period must be larger than the learning cycle. Learning takes 7 days or 50 conversions. Decisions inside that window are decisions in fog.

The 5 scaling methods

Most traffickers choose vertical because it’s simple. Vertical has a ceiling. Mature scaling uses a mix.

A. Vertical scaling (same Ad Set, more budget)

• When: Ad Set with stable history, consistent performance, low frequency
• Limit: 20% increase at a time, max 2x per week
• Risk: exiting learning phase if you raise too much

B. Horizontal scaling (duplication)

• When: you don’t trust vertical to hold. Or want to test audience variations without touching the original
• Procedure: duplicate the winning Ad Set with one variable changed (slightly different audience, different placement). Don’t duplicate identically — it cannibalizes in the auction
• Limit: up to 3 duplications per campaign in parallel

C. CBO with multiple Ad Sets

• When: you have 2-3 audiences validated separately in ABO and want to give the algorithm freedom
• Procedure: create a new CBO campaign (don’t convert ABO to CBO — always start fresh). Place validated audiences as child Ad Sets. Total budget: sum of ABO + 30%
• Risk: CBO can concentrate budget on one Ad Set and ignore the others

D. Geographic scaling

• When: you’ve exhausted the audience within current geography
• Procedure: duplicate the campaign with expanded geography. Don’t change the original — open a new front

E. Channel scaling

• When: the campaign worked well on Reels, you want to test Stories or Feed
• Procedure: duplicate with isolated placement, creative adapted to the new format

RULE — Never increase an Ad Set’s budget by more than 30% at a time. Bigger jump and the algorithm treats it as a new Ad Set, restarts learning, burns history.

Pace — depends on what

The 20%-every-3-days rule is good for beginners. The detailed “depends”:

Depends on current budget size

At $20/day, 20% is $4. Algorithm absorbs easily. At $400/day, 20% is $80 — can pull out of learning. The bigger the budget, the smaller the safe percentage.

Depends on daily conversion volume

• 50+ conversions/day: can increase more aggressively, algorithm has data to recalibrate fast
• 5-10/day: any increase is risky, volume isn’t enough for the algorithm

Depends on product sales cycle

• Impulse product (immediate purchase, low ticket): scale faster, see results in hours
• Long consideration product (lead → call → sale in 2-4 weeks): scale slowly, real ROAS signal takes time

Depends on creative health

Recently validated creative has room. Creative with 2+ weeks at high scale is near saturation regardless of today’s numbers. When frequency exceeds 3, prepare the next creative before scaling more.

Practical pace

The 6 mistakes that burn an Ad Set

Mistake 1: Exiting learning and not stabilizing

• Symptom: ROAS plummets 24-48h after the increase, CPA spikes
• Cause: increase greater than 30% or simultaneous change of more than one variable
• Prevention: one variable at a time. Each change waits 3-4 days before the next
• Recovery: don’t revert. Reverting resets again. Wait 5-7 days for re-stabilization. If it doesn’t, duplicate the pre-change Ad Set and pause the current one

Mistake 2: Audience saturation

• Symptom: frequency rises above 3, CTR falls, CPA rises progressively
• Cause: audience too small for current budget, or tired creative
• Recovery: creative refresh first. If that doesn’t fix it, audience expansion. If still not, budget reduction — sometimes the answer is down, not up

Mistake 3: Algorithm optimizing for wrong volume

• Symptom: leads/purchases grow but quality drops. Immediate ROAS good, 30-day ROAS bad
• Cause: wrong optimization event
• Prevention: optimize for the deepest possible event with sufficient volume (50+/week)

Mistake 4: Burning Ad Set by editing

• Audience change: most destructive edit
• Creative change: second most destructive
• Budget change within 30% limit: least destructive

When you want to test something, duplicate and test on the copy — never edit what’s performing.

Mistake 5: Broken attribution after iOS 14.5

• Symptom: Meta’s ROAS different from backend ROAS (Shopify, Stripe)
• Cause: incomplete tracking, misconfigured CAPI, wrong deduplication
• Prevention: properly configured CAPI, deduplicated events, purchase event with correct value and currency. Monthly audit

Mistake 6: Ignored seasonality

• Symptom: campaign that was going well crashes on date X
• Cause: holiday, seasonal event, behavior shift
• Prevention: niche seasonality calendar. Don’t scale during inflection windows

Protocol: scale safely vs burn

Before scaling (checklist)

• ✅ Analysis period: 7 rolling days
• ✅ Stable metrics (CPA varies <15% across 7 days)
• ✅ Frequency below 2.5
• ✅ Backup creative ready
• ✅ Tracking audited within last 30 days

During the increase

• One variable at a time
• Increase max 20-30% (10-15% at high scale)
• Don’t touch creative, audience, or placement simultaneously
• Monitor first 24h, final decision at 72h

What NOT to do (guaranteed burn)

• Increase 100%+ budget at once
• Touch creative during learning phase
• Change audience because “we want to test”
• Pause and reactivate to “save during low-performance hours”
• Scale on Monday (competitive auction, bad data for decisions)
• Scale a campaign whose creative has been running 30+ days without refresh
• Scale without working CAPI (you’re blind)

The philosophical difference

Safe scaling treats the algorithm as a partner with inertia. You know it has learning latency, so you give time, space, consistent data. You don’t ask it to leap — you ask it to grow.

Burning an Ad Set treats the algorithm as a machine that obeys. You think if you increase budget, it’ll deliver proportionally more. It won’t. The algorithm needs to reallocate when budget changes — it’s looking for a different audience than before (because the previous one doesn’t scale). Give it too little time to find them, it delivers garbage. Give it too much budget at once, it delivers more expensive garbage.

The lesson that separates consistent scalers from those who burn Ad Sets weekly: the goal of scaling isn’t to increase budget. It’s to increase results while preserving efficiency. Budget increase is just the tool.

Those who understand this scale slowly and reach far. Those who don’t scale fast and zero out every 30 days.

Next steps

• The brief that becomes a great creative
• Your first campaign in 5 min
• When to rotate proxy IP