# How to Reallocate Marketing Budget Using Attribution A practical workflow for using multi-touch attribution data to reallocate marketing budget. Includes the 5-step reallocation process, sensitivity analysis, common pitfalls, and a downloadable budget template. - Canonical: https://mbuzz.co/articles/budget-reallocation-attribution - Published: 2026-07-23 - Author: Holly Mehakovic, mbuzz (https://mbuzz.co) --- > **TL;DR:** Reallocate budget using attribution in 5 steps: (1) Calculate current spend and attributed revenue per channel, (2) Compute marginal ROAS for each channel, (3) Identify over/under-invested channels, (4) Propose reallocation with sensitivity analysis, (5) Implement gradually and measure. Key principle: shift budget from channels with declining marginal returns to channels with higher marginal efficiency. Validate changes with incrementality tests before making large shifts. ## The 5-Step Reallocation Process
STEP 1
Baseline Analysis
→
STEP 2
Marginal Analysis
→
STEP 3
Reallocation Proposal
→
STEP 4
Sensitivity Analysis
→
STEP 5
Implementation
## Step 1: Baseline Analysis ### Gather Current Data Pull the following for each channel over the last 90 days: ```sql -- Baseline metrics by channel SELECT channel, SUM(spend) AS total_spend, SUM(attributed_revenue) AS total_revenue, SUM(attributed_revenue) / NULLIF(SUM(spend), 0) AS average_roas, COUNT(DISTINCT conversion_id) AS conversions, SUM(spend) / NULLIF(COUNT(DISTINCT conversion_id), 0) AS cpa FROM channel_performance WHERE date >= CURRENT_DATE - INTERVAL '90 days' GROUP BY channel ORDER BY total_spend DESC; ``` ### Example Baseline

BASELINE ANALYSIS (LAST 90 DAYS)

| Channel | Spend | Revenue | Avg ROAS | Conv | CPA | |---------|-------|---------|----------|------|-----| | Paid Search | $150,000 | $525,000 | 3.5× | 1,050 | $143 | | Paid Social | $100,000 | $280,000 | 2.8× | 700 | $143 | | Email | $20,000 | $240,000 | 12.0× | 600 | $33 | | Display | $50,000 | $75,000 | 1.5× | 250 | $200 | | Retargeting | $30,000 | $120,000 | 4.0× | 300 | $100 | | **Total** | **$350,000** | **$1,240,000** | **3.5×** | **2,900** | **$121** | > **Warning:** > **Don't stop at average ROAS:** Email looks amazing at 12.0x, but this is misleading. Email converts existing leads—it doesn't generate new demand. Reallocating budget TO email won't produce 12.0x returns. You need marginal analysis. ## Step 2: Marginal Analysis ### Why Marginal ROAS Matters **Average ROAS** = Total revenue / Total spend **Marginal ROAS** = Additional revenue from additional spend These are very different:

PAID SOCIAL

Current spend: $100,000

Current revenue: $280,000

Average ROAS: 2.8×

If we add $20,000:

Estimated additional revenue: $40,000

Marginal ROAS: 2.0× (lower than average)

Why? Diminishing returns. We've already captured high-intent audiences.

EMAIL

Current spend: $20,000

Current revenue: $240,000

Average ROAS: 12.0×

If we add $20,000:

We can't "buy more email". The audience is fixed.

More spend = more frequency = fatigue = worse performance

Marginal ROAS: 0.5× or worse

Why? Email doesn't scale with spend the way paid does.

### Estimating Marginal ROAS Methods to estimate marginal returns: | Method | How It Works | Best For | |--------|--------------|----------| | **Historical spend variation** | Compare ROAS at different spend levels | Channels with spend variance | | **Geo experiments** | Increase spend in test markets | Paid channels | | **Time-series analysis** | Regress ROAS against spend over time | Stable channels | | **Platform curves** | Use platform's marginal cost projections | Google, Meta | ```ruby # Estimate marginal ROAS from historical data def estimate_marginal_roas(channel, spend_increase_pct: 0.20) historical = get_historical_performance(channel, months: 12) # Fit diminishing returns curve: Revenue = a * Spend^b (b < 1) model = fit_power_curve(historical[:spend], historical[:revenue]) current_spend = historical[:spend].last proposed_spend = current_spend * (1 + spend_increase_pct) current_revenue = model.predict(current_spend) proposed_revenue = model.predict(proposed_spend) marginal_revenue = proposed_revenue - current_revenue marginal_spend = proposed_spend - current_spend { current_roas: current_revenue / current_spend, marginal_roas: marginal_revenue / marginal_spend, saturation_point: model.find_inflection_point } end ``` ### Marginal ROAS by Channel

MARGINAL ROAS ANALYSIS

| Channel | Avg ROAS | Marginal ROAS (at +20%) | Status | |---------|----------|-------------------------|--------| | Paid Search | 3.5× | 2.5× | Room to grow | | Paid Social | 2.8× | 2.0× | Room to grow | | Email | 12.0× | 0.5× | Can't scale with $ | | Display | 1.5× | 1.8× | Underinvested | | Retargeting | 4.0× | 1.2× | Near saturation |

INSIGHT

Display has a higher marginal ROAS (1.8×) than its average (1.5×). We're underinvesting in display relative to the opportunity.

đź“Ś PLACEHOLDER: Marginal ROAS Curve Visualization

Add chart showing diminishing returns curves for each channel. X-axis = spend, Y-axis = revenue. Show current position on each curve and the marginal slope at that point. Visual makes the concept immediately clear.

## Step 3: Reallocation Proposal ### Calculate Optimal Allocation In theory, optimal allocation equals marginal ROAS across all channels:

OPTIMAL ALLOCATION PRINCIPLE

Move budget until: Marginal ROAS (Channel A) = Marginal ROAS (Channel B)

Paid Social marginal (2.0×) > Display marginal (1.8×)

→ Status quo is approximately optimal for these two

Display marginal (1.8×) > Retargeting marginal (1.2×)

→ Shift budget from Retargeting → Display

### Build Reallocation Scenarios

PROPOSED REALLOCATION (CONSERVATIVE)

| Channel | Current | Proposed | Change | Rationale | |---------|---------|----------|--------|-----------| | Paid Search | $150,000 | $157,500 | +$7,500 | High marginal ROAS | | Paid Social | $100,000 | $110,000 | +$10,000 | Good marginal, more room | | Email | $20,000 | $18,000 | −$2,000 | Can't scale with spend | | Display | $50,000 | $62,000 | +$12,000 | Underinvested (high marginal) | | Retargeting | $30,000 | $2,500 | −$27,500 | Near saturation | | **Total** | **$350,000** | **$350,000** | **$0** | **Budget neutral** |

Maximum shift: 20% of any channel's budget. The Retargeting reduction is larger because its marginal ROAS is below the acceptable threshold.

### Set Shift Limits > **Note:** > **The 20% rule:** Never reallocate more than 20% of a channel's budget in a single cycle. Larger shifts are too risky—you can't easily reverse if performance drops. Exception: channels below break-even marginal ROAS can be cut more aggressively. ### Account for Interdependence Before finalizing, check for channel dependencies: | If You Cut... | Watch For... | |---------------|--------------| | Paid Social | Email list growth slowing | | Display | Branded search volume dropping | | Retargeting | Longer conversion windows | | Content/SEO | Fewer MQLs entering pipeline | ```ruby def check_interdependence(proposed_cuts) risks = [] proposed_cuts.each do |channel, cut_amount| dependent_channels = get_dependent_channels(channel) dependent_channels.each do |dep| impact = estimate_downstream_impact(channel, dep, cut_amount) if impact[:revenue_loss] > cut_amount * 0.3 risks << { cut_channel: channel, affected_channel: dep, estimated_loss: impact[:revenue_loss], recommendation: "Reduce cut amount or test with holdout" } end end end risks end ``` ## Step 4: Sensitivity Analysis ### Model Revenue Impact Calculate expected revenue change from reallocation: ```ruby def calculate_reallocation_impact(current, proposed) total_impact = 0 proposed.each do |channel, new_spend| old_spend = current[channel] spend_change = new_spend - old_spend if spend_change > 0 # Increasing spend: use marginal ROAS revenue_change = spend_change * marginal_roas(channel, direction: :increase) else # Decreasing spend: revenue loss at marginal rate revenue_change = spend_change * marginal_roas(channel, direction: :decrease) end total_impact += revenue_change end total_impact end ``` ### Scenario Table

SENSITIVITY ANALYSIS

| Scenario | Budget Shift | Expected Revenue | Expected ROAS | Risk Level | |----------|--------------|------------------|---------------|------------| | Status Quo | $0 | $1.24M | 3.5× | — | | Conservative (+5%) | $7,500 | $1.27M | 3.6× | Low | | Moderate (+10%) | $15,000 | $1.29M | 3.7× | Medium | | Aggressive (+15%) | $22,500 | $1.31M | 3.7× | High |

Note: the aggressive scenario has diminishing returns. The extra $7,500 shift only adds $20K revenue versus $30K in the moderate scenario.

### Set Success/Failure Thresholds Define in advance what would trigger reversal: | Metric | Success | Acceptable | Failure (Reverse) | |--------|---------|------------|-------------------| | **Overall ROAS** | +5% | ±3% | -5% | | **Increased channels** | +10% ROAS | Flat | -10% ROAS | | **Decreased channels** | N/A | ROAS stable | Other channels drop | | **Leading indicators** | On trend | Slightly down | Significant decline |

đź“Ś PLACEHOLDER: Sensitivity Analysis Calculator

Embed interactive calculator where users can: (1) Input current channel spend and ROAS, (2) Adjust proposed allocation, (3) See projected revenue impact with confidence ranges, (4) Download scenario comparison.

## Step 5: Implementation ### Phase the Rollout Don't make all changes at once:

IMPLEMENTATION TIMELINE (8-WEEK EXAMPLE)

WEEK 1–2
Phase 1 (50% of shift)
→
WEEK 3–4
Evaluation Period
→
WEEK 5–6
Phase 2 (remaining 50%)
→
WEEK 7–8
Validation
### Monitor Leading Indicators Don't wait for conversions—watch early signals: | Indicator | Healthy | Warning | Action If Warning | |-----------|---------|---------|-------------------| | **CPM** | Stable or down | Up >20% | Audience saturation—slow increase | | **CTR** | Stable or up | Down >15% | Creative fatigue—test new creative | | **CPC** | Stable | Up >25% | Competition or quality—review targeting | | **CVR** | Stable or up | Down >10% | Traffic quality issue—check sources | ### Validate with Holdouts Run a geo-holdout or randomized experiment to validate: ```ruby def design_reallocation_validation_test { test_type: :geo_holdout, duration: "4 weeks", treatment_geos: ["California", "New York", "Texas"], # 40% of traffic control_geos: ["All other states"], # 60% of traffic treatment: "New allocation (proposed)", control: "Old allocation (status quo)", success_metric: :attributed_revenue, minimum_detectable_effect: 0.05, # 5% lift analysis_plan: { primary: "Compare ROAS between test and control geos", secondary: "Compare by channel within test geos", validation: "Check for geo selection bias" } } end ```

đź“Ś PLACEHOLDER: Reallocation Case Study

Add mbuzz customer example: "Company X identified Display as underinvested (1.8x marginal vs 1.2x Retargeting). They shifted $15K/month from Retargeting to Display. After 8 weeks, overall ROAS improved 8% with no decline in Retargeting-attributed conversions—they had hit saturation."

## Common Reallocation Mistakes ### Mistake 1: Chasing Average ROAS Email might show 10x ROAS, but you can't "buy more email subscribers" by increasing spend. Always use marginal ROAS for allocation decisions. ### Mistake 2: Ignoring Interdependence Cutting paid social by 50% might cause email ROAS to drop 20% a month later—fewer people entering the list. Model dependencies before cutting. ### Mistake 3: Moving Too Fast Large, rapid shifts make it impossible to diagnose what worked. Phase changes over 4-8 weeks. ### Mistake 4: Not Setting Failure Criteria Without pre-defined reversal triggers, teams let bad allocations persist too long. Define "failure" before implementing. ### Mistake 5: Reallocating Based on Platform Data Google says Google is great. Meta says Meta is great. Both are biased. Use your own multi-touch data for allocation decisions. ## Summary Reallocate budget using this 5-step process: 1. **Baseline Analysis** — Current spend, revenue, average ROAS by channel 2. **Marginal Analysis** — Estimate diminishing returns curves, find highest opportunity 3. **Reallocation Proposal** — Build scenarios with 20% max shift limits 4. **Sensitivity Analysis** — Model revenue impact, set success/failure thresholds 5. **Implementation** — Phase over 4-8 weeks, monitor leading indicators, validate with holdouts **Key principles:** - Use marginal ROAS, not average ROAS - Account for channel interdependence - Never shift more than 20% at once - Validate with incrementality tests - Quarterly cycles, not monthly ## Further Reading **On Attribution for Budgeting:** - [How to Build a Bottom-Up Revenue Forecast with MTA](/articles/bottom-up-revenue-forecast) — Using attribution for planning - [How to Choose the Right Attribution Model](/articles/how-to-choose-attribution-model) — Model selection for budget decisions

đź“Ś PLACEHOLDER: Downloadable Budget Template

Add download link for Excel workbook with: (1) Current allocation input sheet, (2) Marginal ROAS estimator, (3) Reallocation scenario builder, (4) Sensitivity analysis with charts, (5) Implementation timeline template.

## Key takeaways - Focus on marginal ROAS, not average ROAS, for reallocation decisions - Never reallocate more than 20% of a channel's budget at once - Validate with incrementality tests before major shifts - Account for channel interdependence—cutting introducers affects closers ## FAQ **How often should I reallocate budget based on attribution?** Quarterly for most businesses. Monthly reallocation can chase noise and doesn't allow enough time to see effects. Weekly reallocation is almost never appropriate—you need conversion cycles to complete. Exception: during major campaigns or seasonal peaks, you might adjust within-channel spend more frequently. **Should I use first-touch, last-touch, or multi-touch for budget decisions?** Multi-touch (linear or position-based) for cross-channel allocation. It provides the most balanced view of contribution. Use first-touch to understand pipeline sourcing and last-touch for conversion optimization, but neither alone is appropriate for budget allocation—they're too biased. **What if my attribution data conflicts with platform data?** Common situation. Platform data (Google, Meta) uses their own attribution, typically crediting themselves generously. Trust your multi-touch data for allocation decisions, but use platform data for within-platform optimization (bid strategies, audience targeting). **How do I handle channels with no attribution data?** Channels like podcasts, billboards, or TV rarely have direct attribution. Use MMM (media mix modeling) or incrementality tests for these. Don't assume zero contribution—they may be driving unattributed conversions that show up as 'direct' or 'branded search.' **What's the difference between reallocating budget and optimizing within a channel?** Reallocation moves budget between channels (Google → Facebook). Optimization adjusts within a channel (bid strategies, audiences, creative). Attribution data is good for reallocation; platform data is often better for within-channel optimization. **How do I prioritise budget allocation when performance data conflicts across Google, Meta, LinkedIn and programmatic?** Work in this order. First, stop summing platform numbers, because each platform claims the same conversion and the total always exceeds actual sales. Second, get one deduplicated cross-channel view so every channel is judged by the same attribution logic. Third, look at the spread between models rather than a single number: if a channel's credit holds steady across last-touch, linear and Markov, that is a robust signal; if it collapses when you change models, the channel's apparent performance was a model artefact. Fourth, allocate on marginal return, not average return, because the next dollar into a saturated channel is worth less than the average dollar already in it. Fifth, for your largest spend lines, confirm with a holdout or geo test before committing a big shift. Use platform data only for within-platform optimisation, never for cross-channel comparison. **Which channel should I trust when every platform claims credit for the same conversion?** None of them individually, because each platform can only see its own touchpoints and applies attribution logic designed to credit itself. Google Ads uses a 30-day click window, Meta uses 7-day click and 1-day view, and a single buyer who touched both gets counted twice. The resolution is not picking a winner among the platforms but adding a neutral layer that sees all touchpoints and applies one consistent model, then treating the platforms as directional inputs for optimisation inside each channel.