Scale-or-Kill Decision Diagnostic — A Healthcare Framework
The 5 factors
| Factor | Scale signal | Kill signal |
|---|---|---|
| CPL vs specialty median | Top quartile | Below median + worsening |
| Hold Rate (3-sec view) | >45% | <35% |
| Frequency over 28-day window | <3.0 | >4.0 |
| Audience saturation | Low (<30%) | High (>60%) |
| CPL trend (week-on-week) | Stable or improving | Worsening 2+ weeks |
Scale decision
- 4-5 scale signals: Scale 20-30% this week. Re-evaluate after 7 days.
- 3 scale signals + 2 neutral: Hold steady, monitor.
- 2 scale signals + 3 mixed: Stay neutral. Don't scale.
Kill decision
- 3+ kill signals: Kill within 48 hours. Document outcome.
- 2 kill signals: 1-week observation period. Re-evaluate.
- 1 kill signal: Watch but don't act.
Why 20-30% scale increments (not 50%+)
Meta's algorithm requires re-learning when budget jumps significantly. 50%+ scale typically resets the learning phase, producing 7-14 days of elevated CPL during re-learning. 20-30% increments stay within the learning tolerance, preserving performance.
Across 100+ healthcare ad set scale events in the Meta Catalyst IQ dataset:
- 20-30% increments: CPL preserved in 78% of scale events
- 40-50% increments: CPL preserved in 51% of scale events
- 60%+ increments: CPL preserved in 28% of scale events
The lesson: incremental scaling produces consistent results; aggressive scaling produces volatility.
The "early kill" mistake
Most teams kill too early. New ad sets need 14-21 days minimum to:
- Exit the learning phase (Meta requires ~50 events typically)
- Establish a representative CPL baseline (random variance smooths out)
- Reveal audience response patterns
Killing on day 5-7 because "CPL looks bad" — the diagnostic shows kill signal = 1 (CPL high). Other factors (Hold Rate, Frequency, Saturation, Trend) need 14-21 days of data to assess. Premature kills throw away ad sets that would have stabilized.
The "late kill" mistake
Equally common: keeping ad sets alive long after they should be killed. Reasons:
- "It used to work, maybe it'll come back" — anchoring bias
- "We spent so much building this audience, can't kill it now" — sunk cost
- "Manager said don't kill without 30-day data" — bureaucracy
3+ kill signals sustained over 2 weeks = unambiguous kill. Keeping it alive wastes budget.
Healthcare-specific calibration
The 5-factor framework adjusts for specialty:
- IVF: longer evaluation window (28+ days) due to high consideration journey
- Dermatology (cosmetic): faster evaluation (14 days) due to lower consideration time
- Hair transplant: medium evaluation (21 days)
- Dental aesthetic: medium (21 days)
- Eye (LASIK): longer (28 days)
The compound learning effect
Teams using consistent scale-or-kill framework over 6-12 months produce:
- Consistent CPL improvement (8-15% YoY)
- Reduced wasted spend (Money Wastage trend down)
- Compounding audience signal quality
- Faster identification of Core Performers (good ad sets identified weeks earlier)
Get your accounts diagnosed.
ICG runs Scale-or-Kill diagnostic on your active ad sets via Meta Catalyst IQ. Output: list of which to scale, which to kill, which to hold. Founder-led.
Book a free audit → WhatsApp ICG