R / 002Research brief / Media economics

Tangentix Research

ROAS is not a capital-allocation system

Why average channel efficiency cannot answer where the next dollar should go—and what a decision-grade alternative requires.

Published: July 8, 2026
Reading time: 4 minutes
Author: Tangentix Research

Executive decision brief

The conclusion, its use, and its boundary.

Paper: R / 002
Evidence status: Illustrative framework / no observed outcome

Working conclusion

Average ROAS cannot decide where the next dollar goes. Move capital only when marginal response, uncertainty, downside, and operating constraints point to the same bounded action.

Decision to settle: Which channel deserves the next dollar?
Decision output: A balanced allocation proposal with a planning range, guardrails, owner, and realized-review condition.
Claim boundary: All visible allocation and contribution values are illustrative.

Figure 01 / Evidence companion

The method, made inspectable.

Illustrative framework / no observed outcome. The visual is paired with HTML interpretation so the argument does not depend on screenshot pixels.

Open evidence record Decision workspace / TX-WS-MED-01

Artifact: Decision workspace
Evidence status: Illustrative framework / no observed outcome
Record ID: TX-WS-MED-01
Record: Media Optimizer decision workspace / rolling 12 weeks / illustrative values
Decision question: Where should the next media dollar move, within what downside boundary?

Capital allocation / North America

Fund the next dollar against marginal return.

Rolling 12 weeks

Portfolio governed$4.8mannualized media

Proposed move$180kwithin portfolio

Modeled contribution+$220–390k80% planning range

Read every supporting value in the allocation record below.

ChannelCurrentMarginal indexSaturationProposal

Non-brand search28%1.42 58% +$105k

Paid social34%0.73 86% −$140k

Video prospecting16%1.18 49% +$75k

Affiliate22%0.91 76% −$40k

Capital state

A modeled move is not an authorization. An authorization is not a result.

Forecast, authorization, and realized review remain separate records so platform credit cannot quietly become a capital decision.

01
Modeled allocationProposal formed
Baseline, 80% planning interval, marginal-response range, downside boundary, and exposure cap are ready for the decision forum.
Proposed
02
Authorization gateAwaiting accountable sign-off
The move remains proposed until the decision owner and finance reviewer accept the guardrails.
Not approved
03
Realized reviewNo outcome observed
After the review window, compare eligible outcomes with the frozen baseline before the next allocation moves.
Not observed

Scenario desk

Compare the move before committing capital.

Ranges express model uncertainty. They are planning boundaries, not promised outcomes.

Use horizontal scrolling to compare every allocation scenario.

ScenarioCapital movedContribution rangePrimary riskState

Hold current mix$0+$40k–$120kSaturation compoundsReference

Guardrailed reallocation$180k+$220k–$390kLearning preservedRecommended

Aggressive reallocation$340k+$250k–$470kThreshold at riskBounded

Baseline: Approved pre-move response model
Uncertainty basis: 80% planning interval / marginal-response model
Review state: Pending realized outcome
Release condition: Lower bound remains positive after holdout or geographic adjustment

Decision workspace · sanitized reconstruction · illustrative valuesSignal → funding decision → measurement rule

Paper dossier

Decision question: Which channel deserves the next dollar?
Primary method: Marginal response and constrained allocation
Evidence status: Illustrative framework / no observed outcome

Media intelligenceMarginal returnCapital allocation

Editorial responsibility Tangentix Research

Review state: Published / Revision 1.1
Last reviewed: July 17, 2026
Primary sources checked: July 17, 2026

Tangentix Research owns the method framing, evidence-status label, stated limitations, decision rule, and linked-source check. No client result is inferred beyond the evidence disclosed here.

Cite this paper

Tangentix Research. “ROAS is not a capital-allocation system.” Tangentix Research, July 8, 2026. https://tangentix.io/research/roas-is-not-capital-allocation/

Return on ad spend is a reporting ratio. Capital allocation is a constrained decision under uncertainty. Treating the first as the second is how efficient-looking portfolios become economically stagnant.

Average return answers the wrong question

ROAS divides credited revenue by spend across an observed period. It summarizes what happened on average. The funding decision is marginal: what contribution should the next unit of spend create from this point forward?

Two channels can report the same ROAS and deserve opposite decisions. One may still be in a high-elasticity range. The other may be saturated, harvesting demand created elsewhere, or benefiting from a favorable attribution rule.

Four corrections before reallocating

1. Separate reported and incremental contribution

Platform credit is an input, not the verdict. Reconcile it against experiments, baseline models, and observable demand substitution.

2. Estimate the response curve

The relationship between spend and return is not linear. The model needs to identify the high-elasticity range, the inflection point, and the diminishing-return zone.

3. Make constraints explicit

A mathematically optimal move may violate delivery volume, brand coverage, geographic commitments, learning thresholds, or contractual floors. Those constraints belong inside the recommendation.

4. Carry uncertainty into the decision

A point estimate creates false confidence. The allocation proposal should show the expected range, confidence, downside boundary, and the observation window required to learn whether the move worked.

From dashboard to weekly proposal

A decision-grade media system produces a proposal, not another scorecard. Each proposed move should specify:

capital source and destination;
marginal-return evidence;
saturation or concentration risk;
expected incremental contribution range;
constraints preserved;
owner and approval state;
measurement rule and review date.

The sanitized workspace shown with this paper uses an internally balanced portfolio: every increase is funded by a corresponding decrease. Its values are illustrative; the operating logic is the point.

The decision rule

Do not scale because the modeled mean is positive. Scale when the credible range clears the business hurdle and the portfolio can observe the realized effect.

That rule changes the weekly conversation. Teams stop asking which channel has the highest historical ratio and start asking which governed move has the strongest evidence-adjusted economic case.

ROAS remains useful as a diagnostic. It simply should not be granted authority it was never designed to carry.

Research record

Evidence, limitations, and reproducibility.

Revision: 1.1
Published: July 8, 2026
Last reviewed: July 17, 2026

Evidence record

Use horizontal scrolling to inspect every source and its role in the argument.

SourceStatusRole in the argument

Media Optimizer scenario deskIllustrative framework / no observed outcomeCompares hold, guardrailed, and aggressive allocation scenarios

Marginal-response frameworkTangentix method frameworkSeparates average return, marginal response, saturation, and uncertainty

Realized outcome reviewPendingNo scale decision is represented as complete before the review gate

Known limitations

01
All visible allocation and contribution values are illustrative.
02
Scenario ranges depend on model specification, data quality, and stated operating constraints.
03
The workspace does not include a realized post-allocation effect.

Selected references

Primary sources behind the method.

These sources provide methodological context. Their inclusion does not imply that an external study validates Tangentix client outcomes or the illustrative evidence shown here.

01 Jin, Yuxue, Yueqing Wang, Yunting Sun, David Chan, and Jim Koehler. Bayesian Methods for Media Mix Modeling with Carryover and Shape Effects. Google Research, 2017.
Primary reference for modeling advertising carryover, diminishing returns, posterior mROAS, and the uncertainty surrounding an optimized media mix.
02 Zhang, Yingxiang, et al. Media Mix Model Calibration With Bayesian Priors. Google Research, 2024.
Primary reference for calibrating media-mix estimates with experimental evidence and carrying prior sensitivity into ROAS inference.

Reproducibility note

What another analyst would need to preserve.

Reconcile channel spend and contribution, document model form and constraints, preserve the pre-move baseline, balance every source and destination, and compare the forecast range with the realized review window.

Applied method / Media Intelligence

Turn response evidence into a bounded capital move.

A paper establishes the method and its limits. The applied system puts the method beside the accountable decision, working evidence, and release condition.

Decision it informs: Which channel deserves the next dollar?
Working output: Capital-allocation diagnostic record
First evidence to inspect: Spend, channel outcomes, finance baseline, prior tests, and operating constraints.
Application boundary: The method informs a decision only after the available evidence, counterfactual, and accountable owner have been checked.