ScalingBudgetCheatsheetPlaybook

Budget Scaling Increments: When and How Much to Raise

Percent increments by budget tier, the signals required before a raise, cooldowns, horizontal vs vertical rules and rollback triggers.

Updated 2026-04-3011 min read

LanguageEnglish

Raising budget is not a vibe — it is a sized increment, a required signal, a cooldown and a rollback plan. This sheet is the increment table we use by spend tier: how much to raise, when the numbers allow it, when to scale sideways instead of up, and the triggers that mean roll the raise back.

A raise is a test with a known size, not a reward

Most accounts scale in one of two broken ways: they double a winner because yesterday looked good, or they drip 5% raises forever and never find out whether the creative can actually take more volume. Both destroy the signal. A sized increment (large enough to matter, small enough that learning does not fully reset) is the only raise that teaches you something. If CPA holds inside the expected variance after the increment, you earned the next step. If it does not, you roll back — you did not 'ruin the campaign', you ran a test that failed at that volume. The locked table sizes the step by daily budget so the same rule is not applied at $80/day and $4,000/day.

No signal, no raise — even if the creative 'feels' like a winner

The increment is the easy part. The hard part is refusing to raise until the ad set has actually printed a stable read: out of learning, enough conversions, CPA or ROAS at target for consecutive days, frequency not already cooked, and hook/hold still in band. Skipping that checklist is how a two-day lucky streak becomes a 40% budget jump into a learning reset. The signal table is a gate, not a vibe. If any row fails, the correct action is wait or fix creative supply — not a smaller raise 'just to be safe'. A small raise on a failing gate is still a raise on noise.

Vertical vs horizontal is a volume decision

Vertical scaling is raising budget on the same ad set. Horizontal scaling is copying the winner into a new geo, a new lookalike, a new placement cut or a new CBO branch so you buy more volume without shocking one auction. Thin budgets almost always go vertical because there is nothing to split. Fat budgets that keep raising one ad set eventually hit a wall: CPA steps up, delivery gets jumpy, and one bad day takes half the account with it. The sheet tells you at which tier you should prefer horizontal, and what 'a copy' is allowed to change so you are not accidentally launching a brand-new test dressed up as a scale.

Every raise needs a rollback trigger written down first

If you do not know in advance what would make you undo the increment, you will sit in a worse CPA for a week arguing about attribution. Rollback is not panic — it is the other half of the test. After a raise, you wait a cooldown; if CPA or ROAS has moved outside a numbered band versus the pre-raise baseline, you reverse all or half of the step and you do not raise again until a longer cooldown clears. Creative fatigue is a different problem: if the increment held but frequency and hook are sliding, you need new ads, not a smaller budget. That is the moment to batch replacements from a product URL in Klip Kanvas rather than keep squeezing the same three winners.

5 tables inside: raise % by daily-budget tier, the signal gate that must pass before any raise, cooldown timers after raises and rollbacks, horizontal vs vertical rules, and numbered rollback triggers versus the pre-raise baseline.

Vertical raise size by daily budget (same ad set)

$50-100/day: +15-20% per step · not more than one step per 72h
$100-250/day: +20% per step · one step per 48-72h
$250-500/day: +20-25% per step · one step per 48h
$500-1,000/day: +20-30% per step · one step per 48h · house default 20-30%/48h
$1,000-2,500/day: +15-25% per step · one step per 48h · prefer mixing in horizontal copies
$2,500-5,000/day: +10-20% vertical per 48-72h · horizontal first if one ad set already holds >30% of account spend
$5,000+/day: Vertical +10-15% per 72h max · default to horizontal (new ad set/geo/CBO branch)
Hard cap on a single step: Never >30% in one edit · 2x 'because it is working' resets learning and voids the test

Signal gate — all of these before any raise

Learning status: Learning Limited / Learning exited · do not raise inside active learning
Conversion sample on the ad set: ≥50 conversions at the optimization event · or ≥20 with directional CPA if budget cannot hit 50
CPA or ROAS vs target: At or better than target on 3 consecutive days · not a 3-day blend if one day carried it
7-day frequency: Still under 2.5 (cold) · under 4.0 (warm) · if already at ceiling, horizontal or new creative, not a raise
Creative health: Hook still ≥30% (or platform band) · hold still in the 12-20% healthy band · outbound CTR not down >25% vs 7-day peak
No other edit in the cooldown: No bid, audience, creative-add or pixel change in the last 48h · one variable per test
If any row fails: No raise · wait or fix supply · a smaller raise still counts as a raise on noise

Cooldowns (clock starts at the budget edit)

After a vertical raise, read window: Do not judge the step before 24h · full read at 48h (72h if <$250/day)
After a vertical raise, next raise allowed: Not before 48h · 72h below $100/day · 72h at $5,000+/day
After a rollback: No new raise for 72h · 96h if learning re-entered
After adding a creative to a scaling ad set: Treat as an edit · 48h cooldown before a budget step
After a horizontal copy launches: Fund the copy at 50-100% of parent daily · do not raise the copy for 48-72h
Stacked raises in 7 days: Cap at 2 vertical steps per ad set per 7 days · third step must be horizontal or skipped

Horizontal vs vertical — what you are allowed to copy

Prefer vertical when: Daily budget <$500 · one geo · creative set still taking volume without frequency >2.5
Prefer horizontal when: One ad set ≥30% of account spend · or $1,000+/day · or last two vertical steps each lifted CPA ≥15%
Allowed copy (true scale): Same creative + same offer · new geo, new 1% LAL, new placement cut, or new CBO branch
Not a scale copy: New hook, new landing page, new audience broad vs interest · that is a test, fund at testing budget
Copy budget: Start copy at 50-100% of parent daily · never at 2x parent 'to catch up'
Exclusion: Exclude copy audience from parent and vice versa · overlap tax is not 'scale'
Kill rule on a copy: Copy CPA >1.3x parent after 48-72h at ≥15 conversions · pause copy · keep parent

Rollback triggers vs pre-raise 7-day baseline

CPA +25-35% within 48h of the raise: Roll back 50% of the increment · 72h cooldown
CPA +50% or ROAS <70% of baseline within 48h: Full rollback to pre-raise budget · 72-96h cooldown
Learning re-entered after the edit: Do not raise again · wait until exited, then re-read 48h before any step
Delivery collapse (spend <60% of new budget for 2 days): Full rollback · check learning, bid cap and audience size before retrying
Frequency jumped above 2.5 (cold) with CTR down >25%: Do not rollback budget first · refresh creative; rollback only if CPA also tripped the +25% trigger
One bad day inside an otherwise stable 48h: No rollback · require 2-day confirmation except on the +50% CPA tripwire
After two failed raises on the same ad set: Stop vertical · go horizontal or replace creatives · the auction already priced this volume

Locked

Unlock the full cheatsheet

Enter your email and we'll instantly unlock the complete cheatsheet — plus send you fresh ad playbooks every month.

No spam. Unsubscribe anytime.

Ready to put this into practice?

Create your first AI UGC video ad in minutes — no filming, no actors, no editing.

Try Klip Kanvas free