agentleFS
minhnv0807minhnv0807/ai-business-skills/skills/en/55-scaling-ads-global/SKILL.md

55-scaling-ads-global

Skill644 starsChanged 6 months agoMITView on GitHub

Use when a WINNING ad has to be scaled without breaking it — readiness checks, vertical scaling at plus 20 to 30 percent per step, horizontal scaling into new audiences, creatives, and channels, learning-phase reset risk, how Advantage+ and Performance Max behave when scaled, creative refresh cadence, and the signals that mean stop. Trigger on 'scale ads', 'increase the ad budget', 'scale a winner', 'CPA went up after I scaled', 'how fast can I scale', 'it worked at 50 a day and broke at 200'. Not for — fixing an account that is not working yet, see `03-performance-eval-global`; warm audience plans, see `56-retargeting-plan-global`; proving the winner first, see `19-ab-test-setup-global`.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 8 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask up to 4 questions:

    line 34
  2. Distinctive:

    Which platform, and how many spare creatives exist?: Never scale on a single creative.

    line 39
  3. Distinctive:

    Monitor 24-48 hours after every increase: before the next one.

    line 45
  4. Distinctive:

    If any item fails, do not scale.

    line 63
  5. Distinctive:

    After each increase, monitor for 24-48 hours:

    line 74
  6. Distinctive:

    Duplicate the winning ad set into a new audience: Keep the creative identical.

    line 82
  7. Distinctive:

    New creative on the winning audience: Keep the audience identical, add 2-3 variants changing hook or format.

    line 83
  8. Distinctive:

    They need signal, not structure: When performance drops during scaling, the fix is usually more or better creative assets and cleaner …

    line 102
  9. Very specific:

    Every step must still clear the learning floor from 52-account-structure-global: (50 events x target CPA) / 7.

    line 119

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Scaling Ads (Global)

Scaling means increasing spend while holding or improving efficiency. Scaling badly means CPA spikes and budget burns. Hard rule: never increase budget while CPA is bad. Fix the problem first (run 21-ads-audit-global), then scale.

Information gathering

Ask up to 4 questions:

  1. Is there a proven winner? How many consecutive days has CPA been stable? At least 7 are needed.
  2. What are the current numbers? CPA, ROAS, frequency, spend, and reach against total audience size.
  3. Current daily budget and target daily budget?
  4. Which platform, and how many spare creatives exist? Never scale on a single creative.

Principles

  1. Bad CPA means no scaling. Adding budget to a broken campaign accelerates the loss.
  2. +20-30% per step, never double overnight. Sudden jumps reset the learning phase and CPA spikes. The safest cadence is +20% every 72 hours.
  3. Monitor 24-48 hours after every increase before the next one.
  4. Do not edit a winning ad set while scaling. Any significant change — budget shock, audience edit, creative swap inside the ad set — can reset learning.
  5. Always hold a creative reserve. At least 2-3 fresh creatives ready before scaling. Scaling accelerates fatigue.
  6. Vertical first, horizontal in parallel. Raise the winner's budget while duplicating audience and creative so the account never depends on a single point.
  7. Scaling raises cost per result even when done right. Broader delivery means less efficient impressions. Decide in advance how much CPA erosion is acceptable at the higher volume.

Workflow

1. Readiness check — all must pass

  • Running stably for at least 7 days
  • CPA at or below target on at least 5 of the last 7 days
  • Frequency below 2.5 (audience not saturated)
  • Reach below 50% of the addressable audience (room left to grow)
  • Not currently in the learning phase
  • Tracking verified correct — never scale on bad data (53-tracking-setup-global)
  • At least 2 fresh creatives in reserve

If any item fails, do not scale. Go back to optimizing or preparing.

2. Vertical scaling — raise the budget

ModeIncreaseCadenceCondition
Conservative+20%Every 72hCPA holding at target
Moderate+30%Per step, >= 72h apartCPA clearly better than baseline
Aggressive+50%Once only; exception to the +20-30% step rule, needs a named approver through rule 7Very stable campaign, risk accepted
Never2x overnight—Guarantees a learning reset and CPA spike

After each increase, monitor for 24-48 hours:

  • CPA up by 20% or less: hold, let it stabilize, then increase again.
  • CPA up more than 20%: HOLD, do not increase further.
  • CPA up more than 40%: revert to the previous budget and wait 48 hours for stability.

3. Horizontal scaling — expand

  1. Duplicate the winning ad set into a new audience. Keep the creative identical. Test the same demographic against a different interest theme, lookalike 1% -> 3% -> 5%, or broad. Exclude the original audience to avoid overlap.
  2. New creative on the winning audience. Keep the audience identical, add 2-3 variants changing hook or format. Run these in the testing campaign, not inside the winning ad set.
  3. New channel with the winning message. Port the proven hook from Meta to TikTok or YouTube (or the reverse). Keep the core message, change the execution.
  4. New market. The cheapest volume expansion is often a second country rather than a wider audience at home. Check the cost baseline first: per references/benchmarks-global.md, a Tier 2 market can cost a fraction of Tier 1 CPM, but conversion value and AOV usually drop too. Re-run 10-reverse-kpi-global for the new market before committing budget.

4. Learning-phase reset risk

ActionResets learning?Safer way
Budget increase up to 20-30%Usually notStep up gradually, wait 72h
Budget increase above 30-50%LikelyBreak into several smaller steps
Change audience, placement, or bid strategyYesDuplicate into a new ad set instead of editing
Add or remove creative inside the ad setPossiblyAdd in a new ad set or the testing campaign
Pause and restart after a long gapPossiblyAvoid pausing winners; throttle with budget instead

5. Automated campaign types behave differently

Advantage+ (Meta) and Performance Max (Google) do not respond to scaling the way manual campaigns do:

  • They dislike frequent budget edits. Large or repeated changes push them back into a learning state. Same +20-30% discipline applies, with a longer settling window.
  • They need signal, not structure. When performance drops during scaling, the fix is usually more or better creative assets and cleaner conversion data, not more ad sets.
  • They cannibalize manual campaigns. Running Advantage+ or Performance Max alongside manual campaigns targeting the same pool means competing in the same auction and unreadable attribution. Decide which one owns the objective.
  • Exclusions and audience signals are directional, not hard rules. Do not assume an exclusion list keeps existing customers out the way it does in a manual campaign — verify with a customer-versus-new-customer report.
  • Scale by raising the budget on one campaign, not by duplicating the campaign. Duplicates split the conversion signal and both learn slower.

6. Scaling roadmap

Expressed as multiples of the starting daily budget so it holds in any market or currency. The budget column is cumulative; each individual step stays within +20-30% and at least 72h apart (heuristic).

WeekDaily budgetActionKPI checkpoint
1 (testing)1.0x baselineTest 3-5 ad setsCPA at or below target
21.4-1.7xTwo steps of +20-30% on the winner (>= 72h apart), propose pausing losers per the house ladder (rule 7)CPA stable 5 of 7 days
32.0-3.0xTwo more +20-30% steps (>= 72h apart), duplicate the winner into a new audienceVolume up, CPA held
43.0-5.0xHorizontal: lookalikes, broad, new channelCPA within tolerance
Month 2+5x+Full scale plus continuous new creativeROAS at or above target

Every step must still clear the learning floor from 52-account-structure-global: (50 events x target CPA) / 7.

7. Operating cadence

Daily, 15 minutes: yesterday's CPA versus target — scale, hold; if CPA is above 3.0x target once spend >= 3x target CPA (below that, not enough data), only PROPOSE a pause through rule 7 (house ladder in 03-performance-eval-global); frequency above 2.5 (heuristic, cold traffic, 7-day window; retargeting vs account history) means prepare new creative; check spend pace.

Weekly: pull 7 days of data and apply the decision rules — scale winners +20% (steps >= 72h apart), propose pausing losers through rule 7 with a written reason (house ladder), brief new creative wherever frequency is climbing. Compare against references/benchmarks-global.md for the market and industry before concluding a campaign is broken.

Stop-scaling signals

  • CPA up more than 30% versus baseline for two consecutive days
  • Frequency above 3.5
  • ROAS down more than 25%
  • CTR down more than 40% from its stable level
  • Delivery drops suddenly (check for rejections or "learning limited")

Action: revert to the last budget that worked, wait 48-72 hours for stability, and run 21-ads-audit-global if it does not recover.

Creative refresh schedule while scaling

SignalSeverityAction
Frequency above 2.5WarningBrief new creative now
Frequency above 3.5UrgentLaunch new creative today
CTR down more than 30%Week 2-3Refresh the hook, keep the body
RoutineEvery 2-3 weeksKeep 2-3 new creatives ready to swap

Output structure

File name: scaling-plan-[product]-[YYYYMMDD].md

# Scaling Plan — [Campaign/Product]
Baseline CPA: [amount] · Current daily budget: [amount] · Target daily budget: [amount] · Market: [x]

## 1. Readiness checklist     [7 items with status]
## 2. Vertical plan           [increase schedule, 24-48h checkpoints]
## 3. Horizontal plan         [new audiences / creatives / channels / markets, with exclusions]
## 4. Automated campaign notes [Advantage+ / Performance Max handling]
## 5. Weekly roadmap          [budget multiple, action, checkpoint]
## 6. Stop signals            [thresholds and the action for each]
## 7. Creative pipeline       [reserve creatives and deadlines]

Related skills

  • 19-ab-test-setup-global: a winner must come from a valid test before it is scaled.
  • 21-ads-audit-global: audit before scaling if CPA is unstable, and again if scaling breaks performance.
  • 03-performance-eval-global: daily and weekly checks and decision rules.
  • 56-retargeting-plan-global: scaling cold traffic grows the warm pool — raise retargeting budget alongside.
  • 52-account-structure-global: how to split the winner into its own scaling campaign.
  • 57-next-ads-plan-global: this period's scaling results feed next period's plan.

Quality checklist

  • All 7 readiness conditions checked before any budget increase is proposed
  • Increase steps stay at +20-30% with 24-48 hour monitoring points
  • Horizontal plan includes exclusions to prevent overlap with the original audience
  • Learning-reset risk table included with the safer alternative for each action
  • Advantage+ / Performance Max handled separately from manual campaigns
  • Stop-scaling signals stated with numeric thresholds
  • At least 2 reserve creatives confirmed before scaling starts
  • Budget path expressed in multiples, and every step clears the learning floor
  • The hard rule is restated: bad CPA means no budget increase

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.