Free Marketing Mix Model Calculator

The fundamental MMM idea: average ROAS lies

Every marketer knows their average ROAS by channel: Google 3x, Meta 4x, Email 8x. The intuition is to put more money in Email. The intuition is sometimes wrong. Email at 8x might be saturated: the next dollar produces 2x, while Google at 3x is still linear and the next dollar produces another 3x. The right metric for budget decisions is marginal ROAS (the next dollar's return), not average. Optimizing total revenue means equalizing marginal ROAS across channels until adding $1 to any channel returns the same as any other.

The math, top to bottom

For each channel, assume revenue follows a power-law in spend:

revenue(s) = a * s^e
where:
  a = current_revenue / current_spend^e   (fit from current state)
  e = elasticity (1 - saturation/100, capped at 0.95)

Marginal revenue (dRev/dSpend) at the current spend equals e × current_ROAS. With elasticity 0.5 and current ROAS 3x, marginal ROAS is 1.5x.

To maximize total revenue at a fixed total budget B:

optimal_share_i = a_i^(1/(1-e)) / sum_j(a_j^(1/(1-e)))
optimal_spend_i = optimal_share_i * B
predicted_rev   = sum_i(a_i * optimal_spend_i^e)

The exponent 1/(1-e) is what makes high-marginal channels grow faster than low-marginal ones in the optimal allocation. At e=0.5 (50% saturation), the exponent is 2, which is why an Email channel with 8x average ROAS often gets a much larger share than a Google channel with 3x average ROAS — the marginal advantage compounds.

Why this is not real MMM

Real Marketing Mix Modeling is a multivariate regression on 12 to 24 months of weekly data per channel. It models adstock (how spend in week 1 still produces revenue in week 5), seasonality, external factors, and channel-specific saturation curves. It produces confidence intervals for every coefficient. This calculator skips all of that: single global saturation assumption, no adstock, no regression. The advantage is you can use it without 12 months of data. The disadvantage is the lift number is directional, not exact.

How to read the result

Three useful interpretations:

• Channels marked +X% are underfunded relative to their marginal returns. Test growing them gradually.
• Channels marked -X% are saturated. Cut budget and watch revenue drop less than proportionally.
• The lift number is the upper bound assuming the saturation guess is correct. Real-world lift after testing is usually 30 to 60% of the predicted number.

The rebalance-halfway rule

Never move 100% to the suggested allocation in one period. The saturation curves are wrong by surprising amounts in practice — channels that look saturated sometimes scale further than expected, and channels that look untapped sometimes hit hidden ceilings. The standard practice in paid marketing is to move 30 to 50 percent toward the suggested mix, measure for 30 to 60 days, re-run the model, and iterate. Three or four iterations get you most of the way to the optimal mix while avoiding the catastrophic rebalances that come from trusting any single model output completely.

A worked example

You have $100K/quarter across 5 channels. Google Ads ($40K → $120K, 3x), Meta Ads ($30K → $90K, 3x), LinkedIn ($15K → $30K, 2x), Email ($5K → $40K, 8x), Content ($10K → $20K, 2x). Total revenue: $300K. At 50% saturation, the optimizer suggests Google $34K, Meta $26K, LinkedIn $6K, Email $30K, Content $4K. Email's high marginal ROAS pulls a lot of budget toward it. LinkedIn and Content get cut sharply because their average ROAS is only 2x and they look saturated. Predicted total revenue at the new mix: $324K, an 8 percent lift. The honest read: 4 to 5 percent lift is a realistic post-test outcome, with the direction (grow Email, cut LinkedIn) being more reliable than the exact number.