mbfinotti/advertising-skills / skills/paid-media-scaling
paid-media-scaling
VerifiedDecide when a proven paid campaign has earned a budget increase, how large each step should be, how the ramp sequences over weeks and months, and how to avoid performance collapse on the way up - including readiness gates, vertical vs horizontal scaling, and rollback triggers. Use whenever the user asks whether to scale a campaign, how fast ad spend can rise, mentions a budget ramp, a scaling ceiling, or says performance collapsed after the budget was raised - even if they never say 'scaling'. Covers B2B and B2C. Do NOT use to split a fixed total across campaigns (mbfinotti/advertising-skills@ad-spend-allocation) or to track daily spend against a set budget (mbfinotti/advertising-skills@ad-budget-pacing).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 29.8 KB |
| evals/evals.json | config | 34.7 KB |
| references/step-size-figures.md | markdown | 10.7 KB |
| references/worked-ramp-examples.md | markdown | 7.5 KB |
references/worked-ramp-examples.md
View on GitHub# Worked ramp examples
Load this when drafting a Ramp Plan and the user would benefit from seeing a complete one first, or to check a draft against the negative example. Amounts and dates are illustrative; every folklore default shown must be recalibrated to the actual account.
## Worked example - B2C e-commerce, vertical ladder
Context from the interview:
- Paid-social prospecting campaign at $30K/month, 9 weeks of history, blended MER 3.2, marginal aMER on the last spend band 2.1 vs a 1.8 break-even (owner-approved).
- Evidence: triangulated (platform + blended revenue), no causal test yet.
- 30-day penetration ~14%.
- Creative: 7 proven ads, 4 tests/month capacity.
- Cash: net-30 card float, revenue lag ~5 days.
- Target: $90K/month "as fast as safe."
- Last scale attempt: none.
```
MEDIA SCALING RAMP - prospecting campaign, $30K → $90K/month over ~14 weeks
Gates : affordability PASS (payback 0.9 months, per-cohort) | data maturity PASS
(9 weeks > learning + 5-day lag) | marginal PASS (2.1 vs 1.8, triangulated)
| measurement 11/15 PASS | creative supply MARGINAL - 7 proven vs
~6 needed at $30K, but 18 needed at $90K (folklore ratio, calibrate)
| absorption PASS | rollback defined below
Evidence bar : triangulated. Upgrade planned: geo holdout at the $60K tier - beyond
2x start, attributed-plus-blended is no longer enough
Approach : vertical ladder - default rung held (penetration 14% → headroom, target
under 2x at the start), switching horizontal if penetration crosses
~30% mid-ramp; measure-first folded in at the 2x line instead of leading
Steps : W1 $36K (+20%) | hold 2 wks | marginal aMER ≥1.8, reach up, freq <3
W3 $43K (+20%) | hold 2 wks | same set
W5 $52K (+20%) | hold 2 wks | same set + creative count ≥11
W7 $62K (+20%) | hold 2 wks | launch geo holdout here
W9 $75K (+20%) | hold 2 wks | holdout readout gates next step
W11 $90K (+20%) | hold 2 wks | confirm at target
Rollback : marginal aMER <1.8 across a full hold (not one day) → revert one step,
stabilize 2 wks, resume +10% per step. Emergency cut only for runaway
spend, broken destination, or tracking corruption
Ceilings : creative supply binds first (18 proven ads needed at target; pipeline
produces ~4 tests/month at ~1-in-6 win rate - folklore - so creative,
not the auction, sets the ramp speed) | cash clears | penetration ~42%
at target is past the hold band → expect horizontal switch near $70K
Exit : target reached, OR marginal contribution margin ≤ $0 on a band, OR
penetration >35% with declining unique reach → remaining budget goes
horizontal
Open items : geo-holdout design; the ÷$5,000-per-proven-ad ratio is folklore -
recalibrate from this account's fatigue history by W5
```
Why this passes the threshold:
- Every step pre-commits its hold, monitor set, and rollback.
- The causal upgrade arrives before the 2x line.
- The binding ceiling (creative) is named and slows the ramp rather than being discovered mid-collapse.
## Worked example - B2B long sales cycle, measure-first flavored ladder
Context:
- Professional-network lead-gen at $15K/month, cost per SQL $310 vs $400 break-even (ACV $12K × 25% SQL-to-close ÷ margin - per-cohort).
- Conversion lag click-to-closed-won ~5 months.
- Offline conversion loop: live (CRM stages flow back to the platform).
- Target: $40K/month within two quarters, CFO approves steps above 25%.
```
MEDIA SCALING RAMP - lead-gen line, $15K → $40K/month over 2 quarters
Gates : affordability PASS | data maturity PASS on leading indicators only -
closed-won verdict arrives ~5 months late, so steps are judged on
cost per SQL and lead-quality score, never last month's revenue
| measurement 12/15 PASS (offline loop live - precondition, or stop)
| creative supply PASS | absorption: cash float covers 60-281-day
pipeline lag at target - CFO sign-off attached | rollback below
Evidence bar : triangulated via CRM. Causal upgrade: audience-split holdout in Q2 -
TAM too small for a clean geo test (state this, don't fake one)
Approach : vertical ladder with monthly steps - default rung held; measure-first
not promoted despite the 2.6x target, since no clean geo test exists
at this TAM; horizontal (new segment) parked until penetration
signals fire - small TAM saturates fast
Steps : M1 $18K (+20%) | hold 4 wks | cost/SQL ≤ $340, quality score ≥6/9
M2 $21.5K (+20%) | hold 4 wks | same + penetration check
M3 $26K (+21%, CFO) | hold 4 wks | CRM-platform reconciliation -
when they disagree, the CRM wins
M4-M6 continue +15-20%/month to $40K, each gated on the prior hold
Rollback : cost/SQL >1.5x target across a full hold → cut 20-30%, stabilize
2 wks, resume +10%/month. Falling CPL with flat SQL volume = broken
proxy → freeze the ramp and fix the proxy, don't celebrate
Ceilings : TAM binds first - 30-day penetration 22% now; at ~35% the remaining
increase opens a second segment instead | sales capacity: SDR team
absorbs ~1.6x current lead volume before speed-to-lead degrades -
staffing gate at M4
Exit : target reached, OR penetration >35%, OR cohort ROAS at 180 days
(first readable cohort, M6) fails the boundary → hold at last
good tier until the cohort verdict
Open items : quality-score sample (~20 scored calls/month) to keep the leading
indicator honest; audience-split holdout design for Q2
```
Why this differs from B2C, and only where it should: same conjunction, same loop, same ceilings. What changes:
- Monthly holds.
- Leading indicators instead of revenue.
- The CRM as arbiter.
- Sales capacity as a gate.
## Negative example - annotated
A widely installed open-source ads skill reduces scaling to a single line - "budget reallocation: move budget to top performers" - and its sample output recommends increasing budget on the campaign with the best CPA in a summary table. What's wrong, line by line:
- **"Best CPA" is a blended, attributed average** - no marginal read, no evidence label, and the best-looking CPA line (usually brand or retargeting) is exactly where attribution overstates most.
- **No readiness gates.** Nothing checks affordability, data maturity, creative supply, cash, or measurement health before recommending the raise.
- **No step size, no derivation.** "Increase budget" with no magnitude, so the user defaults to a big jump - the documented learning-reset trap.
- **No hold period.** Nothing says when to judge the change or on what window; a bad first week triggers panic, a lucky one triggers another raise.
- **No rollback.** The recommendation has no down-rule, no trigger, no verification date - the single most reliable marker of an unsafe scaling instruction.
- **No ceiling or exit.** The logic recommends the same raise forever; nothing detects saturation or a binding non-media constraint.
The one-sentence test for any scaling recommendation: does it name the evidence, the step, the hold, the rollback, and the exit? Missing any one of the five, it is a budget edit wearing a plan's clothes.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |