How to Reallocate Marketing Budget Using Attribution

The 5-Step Reallocation Process

Step 1: Baseline Analysis

Gather Current Data

Pull the following for each channel over the last 90 days:

-- 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)

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:

Estimating Marginal ROAS

Methods to estimate marginal returns:

# 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

Step 3: Reallocation Proposal

Calculate Optimal Allocation

In theory, optimal allocation equals marginal ROAS across all channels:

Build Reallocation Scenarios

PROPOSED REALLOCATION (CONSERVATIVE)

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

Account for Interdependence

Before finalizing, check for channel dependencies:

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:

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

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:

Step 5: Implementation

Phase the Rollout

Don't make all changes at once:

IMPLEMENTATION TIMELINE (8-WEEK EXAMPLE)

Monitor Leading Indicators

Don't wait for conversions—watch early signals:

Validate with Holdouts

Run a geo-holdout or randomized experiment to validate:

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

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 — Using attribution for planning
- How to Choose the Right Attribution Model — Model selection for budget decisions

Holly Mehakovic

Co-Founder, mbuzz

Holly Mehakovic is Co-Founder of mbuzz. With 10+ years in marketing including roles at Westpac, Avon, and Forebrite, she's obsessed with making measurement actually useful.

Harvard Extension School Forebrite Westpac Avon