Paid media is steered by marginal tranches, not by average ROAS.

SHORT ANSWER

Decision questionHow do you use “Paid media: allocating by incremental return” to reallocate the paid-media budget?

Paid media covers the inventory bought to create, capture or reactivate demand. Its allocation must not compare attributed ROAS measured on different scopes, but the incremental return of the next euro on a common contribution basis. Each channel has a response curve, an observed spend range, a role and dependencies. Search can capture demand created by video; retargeting can claim a sale that was already likely. Optimisation therefore protects the useful roles while moving the tranches whose marginal return falls below the threshold. Move the next euro to the tranche with the highest risk-adjusted incremental return, subject to reach, learning, capacity and coherence constraints.

SUPPORTED DECISIONMove the next euro to the tranche with the highest risk-adjusted incremental return, subject to reach, learning, capacity and coherence constraints.

Evidence status

What was reproduced, and what it does not prove.

The fixture and the Python run verify the saturating response curve, the robust marginal return and the constraints of the educational allocator used by the tier example. They verify neither incrementality, nor ROAS, nor the effectiveness of a real media inventory.

Technical specification of the budget allocator, specification in French (opens in a new tab)
v1, 31 August 2026 · CC BY 4.0 · specification in French · Method document written by marketing-mix.net. The figures are illustrative: they are not market measurements.

Key distinctions

Three conditions before calculating.

Increment rather than attribution

Platforms report the conversions associated with their interactions. Budgeting needs the sales that would not otherwise have happened. Pause tests, geo experiments and MMM calibration bring operational metrics closer to that estimand.

Curve and range

An average return at €200k does not predict the return on the next €50k. Estimate the local slope, publish the observed support and penalise extrapolation. An untested tranche can be funded as learning, not presented as a certain return.

Roles and synergies

Creating demand and capturing it are not mechanical substitutes. Document the interactions, the lags and the presence thresholds. Cutting a reach channel can weaken a search channel several weeks later.

Method

Allocating paid media by marginal contribution

01Normalise the economics
Convert effects into contribution after margin, discounts, returns and variable costs; choose a common horizon and separate immediate, lagged and brand effects.
02Estimate the curves
Combine history, experiments and comparable priors; compute the marginal return per tranche and its interval; flag collinearity, saturation and the extrapolation range.
03Formalise the constraints
Define minimum presence, maximum capacity, commitments, reach, diversification and learning budget; measure the opportunity cost of each constraint.
04Reallocate in steps
Rank the tranches by conservative return, fund up to the threshold, then reserve a share for learning; recompute after each window consistent with the effect lag.

ORIGINAL ASSET

The map of paid-media marginal tranches

The channels are broken down into fundable tranches, each with its own contribution, confidence level and role.

0THRESHOLD 20%50%

CRM 1

50%fund

Video 1

36%fund

CRM 2

32%fund

Search 1

26%maintain

Search 2

24%maintain

Social 1

8%reduce

Audience

-5%test

CSVDownload the paid-media tranche map↓

ILLUSTRATIVE EXAMPLE · SIMULATED DATA

Moving €150k creates €72k of contribution without increasing the budget

01 · SITUATION

Search, social, video and CRM receive €500k, €300k, €400k and €100k. The next €50k tranches have central incremental contributions of €22k, €15k, €31k and €34k, with conservative bounds of €12k, €4k, €18k and €25k.

02 · CALCULATION

Three social and search tranches have conservative returns of 0.08, 0.24 and 0.26. The CRM, video and a second CRM tranche are worth 0.50, 0.36 and 0.32. Reallocating €150k raises the conservative contribution from €29k to €59k, that is +€30k. In the central scenario, the gain reaches €72k. The video minimum of €350k protects reach.

03 · DECISION

The plan funds CRM then video, reduces social and one search tranche, and keeps €40k to test a new audience. Average ROAS was not used to set priorities.

Acceptance conditions

What must be true to act.

01
All channels are compared on a common incremental contribution.
02
Every tranche states its slope, interval and observed range.
03
Synergies and presence thresholds are treated as explicit constraints.
04
A share of the budget funds learning when uncertainty has value.

Limits

What this analysis does not prove.

Historical curves can change with creative, targeting and competition.
Lagged effects make overly frequent reallocations unstable.
Channel-level granularity sometimes hides strong differences between audiences and creatives.

References

Works cited.

Danaher, P. J. & van Heerde, H. J. (2018), “Delusion in Attribution”, Journal of Marketing Research.
Kohavi, R., Tang, D. & Xu, Y. (2020), Trustworthy Online Controlled Experiments.
Naik, P. A. & Raman, K. (2003), “Understanding the Impact of Synergy in Multimedia Communications”, Journal of Marketing Research.