Ad spend allocator
Skill narayan-metaflow/metaflow-marketing-skills/skills/ad-spend-allocator
Optimize budget allocation across paid media channels using Marketing Efficiency Ratio (MER), marginal ROAS analysis, diminishing returns modeling, and incrementality-informed decisions.From its SKILL.md
npx -y skills add narayan-metaflow/metaflow-marketing-skills --skill ad-spend-allocatorAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 18 stars18 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
17.0 KB, as published. Nobody here has run it
Process
- Collect channel data — Spend, revenue/conversions, and time period per channel
- Calculate core metrics — MER, channel ROAS, blended CAC
- Assess marginal efficiency — Where is each channel on its diminishing returns curve?
- Assign channel roles — Primary driver, scaling channel, support channel, fill channel
- Model reallocation scenarios — What happens if you shift $X from channel A to B?
- Factor incrementality — Adjust for platform-reported vs actual incremental lift
- Build forecast — Quarterly plan with seasonality and CPM trend context
- Recommend moves — ICE-prioritized action plan with expected outcomes
Core Metrics Framework
MER vs Platform ROAS
Marketing Efficiency Ratio (MER) = Total Revenue ÷ Total Ad Spend (across ALL channels)
MER is the north star metric. Platform-reported ROAS overcounts because of attribution overlap — Google and Meta both claim credit for the same sale.
| Metric | What It Measures | Use For |
|---|---|---|
| MER | Blended efficiency across all channels | Strategic planning, board reporting |
| Platform ROAS | Platform-attributed revenue ÷ platform spend | Within-platform optimization |
| Incremental ROAS | True lift over holdout baseline | Channel investment decisions |
| Marginal ROAS | ROAS on the last dollar of spend | Allocation decisions |
| Blended CAC | Total spend ÷ new customers | Efficiency per new customer |
| LTV:CAC | Customer lifetime value ÷ CAC | Long-term channel viability |
Rule of thumb: If platform ROAS sums to more than your actual MER, you have attribution overlap. The gap = overclaimed conversions.
Target MER Benchmarks (Starting Points)
| Business Type | Healthy MER | Notes |
|---|---|---|
| DTC ecommerce (physical products) | 3.0-5.0x | Lower for high-margin, higher for low-margin |
| Subscription/SaaS (LTV-driven) | 1.5-3.0x on first purchase | Needs 6-12 month LTV analysis |
| Lead gen (service businesses) | CPL-based, not MER | Use CAC targets instead |
| Luxury/high-AOV | 5.0-10.0x | Can tolerate higher thresholds |
Confidence note: MER benchmarks vary significantly by industry, margin structure, and LTV. These are starting points — calibrate to your actual unit economics.
Marginal ROAS & Diminishing Returns
The key insight: ROAS is not linear with spend. Every channel has a curve.
The Diminishing Returns Curve
| Zone | Characteristic | Marginal ROAS Behavior | Action |
|---|---|---|---|
| Under-invested | Budget capped, high impression share gaps | Marginal ROAS > average ROAS | Scale up aggressively |
| Efficient frontier | Sweet spot — spend matches demand | Marginal ROAS ≈ average ROAS | Maintain, test small increments |
| Saturating | Diminishing returns setting in | Marginal ROAS < average ROAS but above target | Scale carefully, watch for decay |
| Over-invested | Wasting spend on low-intent audiences | Marginal ROAS < target | Cut budget, reallocate |
Identifying Your Position on the Curve
Signals of under-investment:
- Google Search impression share <30% (Lost IS Budget)
- Meta ad set frequency <1.5 (not reaching enough of audience)
- CPA/ROAS stable as spend increased (no efficiency loss)
- Low reach % of eligible audience
Signals of saturation:
- Google Search impression share >80-95%
- Meta frequency >4 with rising CPA
- ROAS declining as spend increased
- Diminishing absolute conversion growth per budget dollar
Practical calculation of marginal ROAS:
Marginal ROAS = (Revenue at spend level B − Revenue at spend level A) ÷ (Spend B − Spend A)
Example: Channel grew from $10K/mo ($40K rev, 4.0x ROAS) to $15K/mo ($55K rev, 3.67x ROAS).
- Marginal revenue = $55K − $40K = $15K
- Marginal spend = $5K
- Marginal ROAS = $15K ÷ $5K = 3.0x (vs 3.67x average)
- Interpretation: The last $5K came at 3.0x. If target is 3.5x, this channel is saturating.
Channel Role Assignment (Power Pack Thinking)
Not every channel needs to be an efficiency champion. Assign roles:
| Role | Purpose | ROAS Expectation | Example Channels |
|---|---|---|---|
| Primary driver | Bulk of revenue, proven efficient | At or above MER target | Google Search non-brand, Meta Advantage+ Shopping |
| Scaling channel | Unlocking new audiences/growth | Slightly below primary, high incremental | Meta prospecting, Google PMax |
| Support/mid-funnel | Consideration and warming | Lower direct ROAS, assist-heavy | YouTube, Meta video, Demand Gen |
| Fill/hygiene | Brand defense, retargeting | Very high ROAS but low volume | Google brand, retargeting |
| Experimental | Testing new channels | Unknown, budget-capped | TikTok, Pinterest, LinkedIn new launch |
Google Power Pack (2025): Demand Gen → AI Max → PMax working together. Demand Gen creates interest at the top, AI Max captures it in search, PMax scales at the bottom. Evaluate as a system, not as siloed channels.
Agency mistake to avoid: Killing a "low ROAS" channel that's actually feeding your "high ROAS" channel. Run incrementality tests before defunding a suspected assist channel.
Incrementality-Informed Allocation
Platform attribution lies. Incrementality testing tells the truth.
Incrementality Hierarchy (highest to lowest evidence quality)
| Method | Evidence Quality | Effort |
|---|---|---|
| Geo-based holdout tests | Highest | High — requires geo split |
| Conversion lift tests (Google, Meta native) | High | Medium — use platform tools |
| Marketing Mix Modeling (Google Meridian) | High for strategic | High — needs data science |
| Ghost ads / PSA tests | Medium-High | Medium |
| Pre-post analysis on pause events | Medium | Low (opportunistic) |
| Platform-reported ROAS alone | Low | Zero — but unreliable |
When Incrementality Testing Is Worth It
- Before major budget shifts ($10K+ reallocations)
- When considering pausing a channel
- When two channels overlap heavily (brand search + Meta retargeting)
- Annually for MMM refresh
Google Meridian (2025): Google's open-source Marketing Mix Modeling framework. Incorporates granular video and Search signals. Calibrate with conversion lift tests. Best for organizations with data science capability or larger spenders ($1M+/year).
Applying Incrementality to Allocation
If platform-reported ROAS is 5.0x but incrementality test shows 60% lift, actual incremental ROAS is 3.0x. Use the incremental number for allocation decisions.
| Channel | Reported ROAS | Incrementality | True Incremental ROAS | Action |
|---|---|---|---|---|
| Google brand | 15x | 20% lift | 3.0x | Defensive only — don't scale beyond natural demand |
| Google non-brand | 4x | 85% lift | 3.4x | Primary driver — scale aggressively |
| Meta prospecting | 3x | 75% lift | 2.25x | Scaling channel — monitor marginal |
| Meta retargeting | 8x | 30% lift | 2.4x | Support — don't pay to reach already-converting users |
Budget Forecasting
Quarterly Budget Modeling
Build a forecast that accounts for:
- Historical seasonality — Previous year's monthly index
- CPM trend — Meta's avg ad price climbed 9% in 2025 (Meta Q1 2025 earnings); similar pressure on Google
- Planned initiatives — New product launches, geo expansions, creative refreshes
- Diminishing returns — Don't forecast linearly past saturation zones
Simple Forecast Template
Q[X] Forecast:
Base spend = [previous quarter adjusted for seasonality]
CPM inflation factor = +9% (2025 Meta), +[X]% (Google)
Expected MER = [trend analysis + planned improvements]
Forecast revenue = Forecast spend × Expected MER
Confidence range: ±15% (narrower if stable, wider if volatile)
Scenario Planning
Always build three scenarios:
| Scenario | Spend Change | Expected Outcome | Risk |
|---|---|---|---|
| Conservative | Flat or -10% | Maintain current efficiency | Missing growth opportunity |
| Base case | +15-20% | Scale within efficient frontier | Typical execution risk |
| Aggressive | +40-50% | Capture growth, risk efficiency decay | Higher CPA, learning resets |
CPM Trend Context (2025-2026)
| Platform | 2025 CPM Movement | Implication |
|---|---|---|
| Meta | +9% avg ad price (Q1 2025 earnings) | Rising costs compress MER — not always a performance issue |
| Google Search | Competitive pressure on non-brand | Growing Smart Bidding adoption pushes auctions |
| YouTube | Relatively stable | Still cost-efficient for reach |
| TikTok | High volatility | Check quarterly |
| Premium pricing | Viable for B2B only |
Implication for allocation: If channel ROAS declined 10% YoY but CPMs rose 9%, the underlying efficiency is roughly flat. Don't cut a channel that's fighting macro pressure successfully.
Reallocation Decision Framework
Step 1: Calculate Current State
| Channel | Spend | Revenue | ROAS | Marginal ROAS | Incrementality | Role |
|---|---|---|---|---|---|---|
| [channel] | $[X] | $[X] | [X]x | [X]x | [X]% lift | [role] |
Step 2: Identify Imbalances
- Channels with marginal ROAS > target → candidates for scaling
- Channels with marginal ROAS < target → candidates for cutting or restructuring
- Channels at saturation → hold, don't scale
- Channels with weak incrementality → evaluate true value
Step 3: Size the Move
Conservative reallocation: Shift 10-15% of budget at a time. Observe for 14-21 days before next move.
Why not 50%? Big reallocations destabilize learning on both sides. The receiving channel enters learning phase, the losing channel loses momentum. Gradual > dramatic.
Step 4: Expected Impact Calculation
Expected new revenue =
(Remaining spend on channel A × ROAS_A) +
(New spend on channel B × Marginal_ROAS_B) +
(Original spend on channel B × ROAS_B)
If expected new revenue > current revenue, the shift is value-creating.
Output Format
## Ad Spend Allocation Analysis
### Current State
| Channel | Spend | Revenue | ROAS | Marginal ROAS | Role |
|---------|-------|---------|------|--------------|------|
**Blended MER:** [X]x (target: [X]x)
**Total Spend:** $[X]/month
**Total Revenue:** $[X]/month
### Key Findings
- [Channel X] is [under-invested / saturating / over-invested]
- Marginal ROAS gap: [specific]
- Incrementality concern: [if applicable]
- CPM trend context: [Meta +9%, etc.]
### Reallocation Recommendations
| Move | From | To | Amount | Expected Impact |
|------|------|-----|--------|----------------|
| 1 | [channel] | [channel] | $[X] (X% of budget) | [specific outcome] |
### Quarterly Forecast
| Scenario | Total Spend | Expected MER | Expected Revenue |
|----------|------------|-------------|-----------------|
| Conservative | $[X] | [X]x | $[X] |
| Base | $[X] | [X]x | $[X] |
| Aggressive | $[X] | [X]x | $[X] |
### Testing & Validation Plan
- [Incrementality tests to run]
- [Monitoring cadence]
- [Rollback criteria]
Example
Input: "DTC skincare brand. Monthly spend: Google $18K (Search + PMax), Meta $22K (ASC + prospecting), TikTok $5K, YouTube $3K. Revenue last month: $165K. I'm told by the CMO to either cut 20% or justify why not. Where should I push and where should I cut?"
Output:
Ad Spend Allocation Analysis
Current State
| Channel | Spend | Revenue (platform-reported) | ROAS | Est. Marginal ROAS | Role |
|---|---|---|---|---|---|
| Google Search (non-brand) | $8K | $32K | 4.0x | 3.5x | Primary driver |
| Google Search (brand) | $2K | $28K | 14.0x | 5.0x (incremental) | Fill/defense |
| Google PMax | $8K | $26K | 3.25x | 2.5x | Scaling |
| Meta ASC | $15K | $48K | 3.2x | 2.4x | Primary driver |
| Meta prospecting | $7K | $14K | 2.0x | 1.8x | Scaling |
| TikTok | $5K | $8K | 1.6x | Unknown | Experimental |
| YouTube | $3K | $5K | 1.67x | Unknown | Support/MOFU |
Blended platform ROAS (sum): 4.0x ($161K/$48K) Actual MER: $165K ÷ $48K = 3.44x Attribution overlap: ~14% (platform sum overcounts by this much)
Key Findings
- TikTok and Meta prospecting are below target — marginal ROAS underwater. These are the obvious cut candidates IF incrementality doesn't justify them
- Google brand search is defensive — high reported ROAS (14x) but likely 80% would happen organically. True incremental ROAS ~3.0x. Don't scale, don't cut
- Google Search non-brand is primary — healthy marginal ROAS at 3.5x, room to scale
- Meta ASC is the workhorse — 3.2x with 15K spend. Still room before saturation
- YouTube is underfunded for meaningful signal — $3K/mo can't support a real test. Either commit to $10K+ or kill it
- CPM context: Meta +9% in 2025 means Meta ASC efficiency is holding up against pressure — don't punish it for macro costs
Reallocation Recommendations (don't cut 20% — shift 20%)
| Priority | Move | From | To | Amount | Expected Impact |
|---|---|---|---|---|---|
| 1 | Kill TikTok test OR commit $10K+ | TikTok | Google Search non-brand | $5K | +$17-18K revenue at 3.5x marginal |
| 2 | Reduce Meta prospecting | Meta prospecting | Meta ASC | $3K | ASC has better marginal; prospecting feeds ASC but at 1.8x marginal, it's too expensive |
| 3 | Kill YouTube at this scale | YouTube | Google Search non-brand | $3K | $3K is below YouTube's signal threshold — reallocate |
| 4 | Net result: | No cut; reallocation | Expected MER lift to 3.7x |
Counter-Argument to Cutting 20%
Cutting 20% ($9.6K) proportionally would:
- Cut $1.6K from Google non-brand (the best performer) — value destructive
- Cut $3K from Meta ASC — value destructive
- Cut $1.6K from Google brand — unnecessary, brand defense is cheap
Instead: Shift the same $9.6K from underperformers to overperformers. Keep total spend at $48K, raise expected revenue to ~$177K, MER moves from 3.44x → ~3.69x.
Quarterly Forecast (next 3 months)
| Scenario | Monthly Spend | Expected MER | Expected Revenue | Notes |
|---|---|---|---|---|
| Conservative (CMO's cut) | $38.4K | 3.5x | $134K/mo | Value destructive |
| Base (reallocation) | $48K | 3.7x | $178K/mo | Recommended |
| Aggressive (scale winners) | $58K | 3.5x | $203K/mo | Requires creative refresh pipeline |
Incrementality Testing Plan
Before any major move, run:
- Google brand holdout test (1 month) — confirm true incremental value
- Meta prospecting geo-split test — verify it's feeding ASC vs wasted
- YouTube investment test — if keeping, commit $10K for 60 days to get real signal
Monitoring Plan
- Weekly MER tracking vs 3.7x target
- Marginal ROAS calculation monthly per channel
- CPM trend overlay — separate cost pressure from performance decay
- Rollback: if reallocation drops MER below 3.3x after 30 days, revert
Guidelines
- Don't sum platform ROAS and call it blended ROAS. Platforms overclaim. Use MER (total revenue ÷ total spend) as the north star.
- Don't make allocation decisions on platform-reported ROAS alone. Platform ROAS overcounts by 10-30% typically. Incrementality testing is the gold standard.
- Don't shift more than 20% of a channel's budget at once. Destabilizes learning on both sides. Gradual reallocation (10-15% moves) preserves performance.
- Don't kill a "low ROAS" channel without testing incrementality. It may be feeding your "high ROAS" channel. Run a holdout test first.
- Don't ignore diminishing returns. Scaling past the efficient frontier wastes money. Monitor marginal ROAS, not just average.
- Don't forecast linearly. Diminishing returns are nonlinear. 2x budget ≠ 2x revenue in most channels.
- Don't conflate CPM inflation with performance decay. Meta +9% CPM in 2025 means some "decline" is macro, not campaign failure.
- Don't run $3K monthly tests and expect signal. Each channel has a minimum signal threshold. Under-funded experiments produce noise.
- Don't assume MMM output without incrementality calibration. Google Meridian and other MMMs need incrementality experiments to calibrate. Uncalibrated MMM is directionally useful but not decisive.
- Cross-references: For within-platform optimization, use the google-ads-optimizer and meta-ads-optimizer skills. For channel-specific analysis, use the campaign-analyzer skill. For reporting the allocation story to stakeholders, use the paid-media-reporter skill.
- Confidence: Target MER benchmarks vary by industry, margin, and LTV. CPM trend data from Meta Q1 2025 earnings may shift quarter to quarter. Incrementality ranges are illustrative — your account will differ.