Use the return on the next marketing dollar when changing the budget

Average ROAS divides a campaign's attributed or modeled revenue by its spending. A budget change asks a different question: how much additional revenue and contribution would another increment of spending create? When response diminishes as spending grows, the next dollar can earn less than the historical average.

Google Research's media-mix modeling paper examines carryover and response-shape effects and distinguishes ROAS from marginal ROAS. The financial application below uses an invented saturating response curve. It illustrates the decision logic without claiming that the curve describes a particular channel or advertiser.

Keep the revenue definition explicit. A curve intended to guide incremental spending should have a defensible connection to incremental outcomes. Applying a sophisticated marginal calculation to an attribution total does not, by itself, turn that total into a causal estimate.

Suppose incremental revenue over a fixed response window is modeled as $1,000,000 multiplied by spend, divided by $100,000 plus spend. The curve rises with spending but flattens as it approaches its assumed upper level. All amounts use the same currency and measurement horizon.

At $50,000 of spend, modeled revenue is about $333,333 and average ROAS is about 6.67. At $100,000, revenue is $500,000 and average ROAS is 5.00. At $150,000, revenue is $600,000 and average ROAS is 4.00.

Each average can look attractive on its own. But increasing spend from $100,000 to $150,000 adds only $100,000 of revenue for $50,000 of spending, an incremental revenue multiple of 2.00. That is well below the 5.00 average at the starting budget.

Assume the modeled revenue has a constant 40% contribution margin before advertising spend. At $50,000 of spend, contribution after advertising is about $83,333. At $100,000, it is $100,000. At $150,000, it falls to $90,000.

The extra $50,000 above the $100,000 budget creates $40,000 of contribution from its $100,000 incremental revenue. It therefore reduces the modeled result by $10,000. Revenue still grows and the average revenue multiple remains four, yet the additional spending worsens profit under these assumptions.

The required incremental revenue multiple is 2.5 because 40% of 2.5 dollars equals one dollar of spend. Adjust that threshold when the incremental product mix, returns, payment costs or fulfillment costs differ. The company's average gross margin is not automatically the right margin for the additional customers.

For the illustrative curve, the local marginal revenue per spending dollar is $1,000,000 × $100,000 divided by ($100,000 + spend) squared. At $100,000 of spend it equals 2.5. Multiplying by the assumed 40% contribution margin gives one dollar of contribution for a very small additional dollar of advertising.

The average incremental return over a large change is different because the curve keeps flattening along that change. That is why the full move from $100,000 to $150,000 earns a revenue multiple of 2.0 even though the local rate at the starting point is 2.5.

Use the finite difference for the actual proposed budget: predicted outcome at the new spend less predicted outcome at the old spend, divided by the spending difference. A derivative is useful for understanding the neighborhood and optimization, but it should not be applied unchanged to a large increase.

Include any discrete commitment needed to make the increase possible. An additional creative production fee or minimum placement package belongs in the proposed alternative even if it is absent from the smooth media-response curve. Evaluate the feasible purchase as a whole before treating a theoretical optimum as an available buying option.

The shape is an assumption or an estimate, not a property guaranteed by the channel name. Review the observed spending range, changes in targeting and offer, seasonality, supply constraints and evidence supporting the response. A curve fitted mostly near $50,000 may say little about a $300,000 budget.

Show uncertainty in both the response level and the curvature. Different plausible parameter values can imply different useful budgets. The apparent precision of an optimizer's recommendation does not exceed the precision of the evidence used to estimate that curve.

Use experiments or other credible evidence to check the response where a material increase is being considered. A small bounded expansion can provide information, but its design should identify the spending effect rather than merely observe another high-demand period. Preserve the original forecast to evaluate what the added spend actually taught the team.

Define whether revenue includes the full modeled response window or only the current period. A channel with delayed effects can be disadvantaged by a short comparison window, while an unsupported long tail can overstate its value. Use compatible horizons and explicit carryover assumptions when comparing alternatives.

If funding the increase requires cutting another channel, include the contribution expected to be lost there. The decision is then a reallocation, not a free addition to the budget. Compare the relevant marginal changes on both sides rather than rank channels only by their average ROAS.

Shared audiences and channel interactions can also matter. Independently adding two channel curves may count some response twice if the models do not represent their relationship. Ask how the combined forecast handles those effects before using separate optima as a company-wide budget.

Document the current budget, proposed increment, predicted revenue difference, contribution assumption, response horizon and uncertainty. Show the modeled contribution after spend at feasible alternatives. Management can then see both the potential gain and how dependent it is on the curve's shape.

After the change, review incremental spending and supported incremental outcomes over the agreed window. Preserve the distinction between a strong average on the established budget and a good return on the next commitment. That distinction is what makes the analysis useful for deciding where additional cash should go.