84% of companies are stuck in a measurement "doom loop", according to Gartner's 2026 research — underfunded measurement leaves marketing's impact unclear, which feeds C-suite skepticism and tighter budgets. Companies caught in it are half as likely to exceed their growth targets. 

Inside most of the enterprises I work with, that failure has a specific shape: the CFO has a revenue forecast, the CMO has a marketing performance report, and the two don't agree. Each is built on its own model, and neither team can prove the other wrong.

That disagreement is the predictable output of how enterprise measurement is structured — built in, not accidental.

Here, I explain why measurement breaks at enterprise scale, why the usual fixes treat the wrong cause, and what a causal approach actually measures.

Measuring Marketing ROI at a Glance

• Marketing ROI measures the incremental profit caused by marketing spend, not the revenue that can be associated with it after the fact.
• At enterprise scale, every business function runs its own correlation-based model — and those models produce contradictory answers.
• Attribution, ROAS, and blended CPA assign credit; none of them isolate cause, so they reward spend that would have converted anyway.
• The root cause is model architecture, not data quality or pipeline integration.
• Causal incrementality measurement quantifies the counterfactual: what would have happened without a given investment.
• A single causal model across functions gives finance and marketing one version of the truth rather than two competing ones.

What is Marketing ROI?

Marketing ROI is the incremental profit generated by marketing investment, divided by the cost of that investment. It answers one question: how much growth did this spend actually cause, net of everything else happening at the same time. It differs from ROAS and attribution because those measure credit for outcomes, not the causal contribution behind them.

Why is Marketing ROI Hard to Measure at Enterprise Scale?

Marketing ROI is hard to measure at enterprise scale because measurement is fragmented across functions, and each fragment runs a different model. Finance models revenue. Marketing models channel performance. Sales models pipeline.

Every model is built on its own data, its own assumptions, and its own definition of success, so they return different answers about the same business. The problem compounds as channels multiply. A modern enterprise spends across search, social, retail media, connected TV, and out-of-home at once.

These channels move together, often in response to the same campaign calendar and the same seasonal demand. When everything rises and falls together, a correlation-based model cannot tell which channel did the work.

Walled gardens make this worse. Each platform measures its own performance and reports its own success, so the numbers that reach the CMO are self-graded and cannot be reconciled across the portfolio. The result is a stack of confident, mutually contradictory reports and no defensible answer to the only question the board cares about: what is driving growth.

What Standard Marketing ROI Metrics Measure

The metrics most enterprises rely on were each built for a narrower job than the one they are now asked to do. Let me take each in turn and show where they run into trouble.

ROAS and Blended CPA

Return on ad spend and blended cost per acquisition measure efficiency, not incrementality. ROAS divides attributed revenue by ad spend; blended CPA divides total spend by total conversions. Both treat every conversion as if marketing caused it.

In practice, a large share of those conversions would have happened anyway. Customers already intending to buy click an ad on their way to the purchase they had decided to make.

ROAS and blended CPA count that sale as marketing-driven, which is credit-claiming, not growth-causing. The metric looks strongest exactly where marketing did the least incremental work, because high-intent audiences convert cheaply regardless of the spend pointed at them.

At enterprise scale, this creates a systematic misread across the portfolio. A brand running $100M+ across channels can show strong ROAS on every line while total business growth flatlines — because the model confirms spend against existing intent, not demand that marketing created. Efficiency metrics and incremental contribution move in opposite directions.

Multi-touch Attribution

Multi-touch attribution distributes credit for a conversion across the touchpoints that preceded it. It answers which channels appeared on the path to purchase, then assigns each a fractional share of the outcome according to a weighting rule.

The weighting rule is the weakness. Whether credit is split evenly, front-loaded, or back-loaded is a modeling choice, not a measured fact, and different rules produce different winners from identical data.

Attribution also sees only what it can track. As third-party cookies disappear and journeys cross devices and walled gardens, the tracked path is a shrinking and biased sample of the real one. The model grows more confident about less.

The deeper problem, and what I tell every team working with attribution data: this tool is answering the wrong question. Distributing credit across a customer journey assumes those touchpoints caused the conversion. It does not test whether the sale would have happened without any of them — the only framing with a budget allocation decision attached to it.

Marketing Mix Modeling

Marketing Mix Modeling estimates the contribution of each channel by regressing historical sales against historical spend. It does not depend on user-level tracking, which is why many enterprises have returned to it as cookies fade.

The statistical engine underneath is the constraint. Traditional MMM is built on ordinary least squares regression, which estimates the effect of one variable while holding the others still. When channels move together — and at enterprise scale they almost always do — the model cannot separate their effects.

This is multicollinearity, the Gordian Knot of MMM: correlated inputs make the individual coefficients unstable, so small changes in the data swing the estimated channel contributions wildly. MMM has served marketers well for decades — I'm not dismissing it. But it was never built to hold every function and every correlated channel in one stable view.

The Model Architecture Problem Holding Enterprises Back

The diagnosis I hear most often is this: a data problem. The pipelines are messy, the channels are siloed, the platforms won't share.

Clean the data, unify the pipelines, and the numbers will reconcile. I've seen this play out many times. It's wrong — and acting on it is why so many measurement projects deliver tidier inputs and the same contradictions.

The real problem is model architecture. Every metric above is correlation-based: it observes what moved alongside sales and assigns credit accordingly.

Correlation cannot answer a causal question, no matter how clean the data feeding it. Running ten correlation-based models across ten functions does not produce one truth; it produces ten well-formatted disagreements.

This is the diagnostic reversal enterprises miss. The contradiction between the CFO's numbers and the CMO's numbers is not a sign that someone's data is wrong.

It is the expected behavior of an architecture that was never designed to produce a single answer. Until the architecture changes, better data only sharpens the disagreement.

What Replaces Attribution: Causal Incrementality Measurement

The fix is not a better attribution model. It is a different question.

Attribution asks which touchpoints to credit for a sale; causal incrementality asks what would have happened without the investment at all. The gap between those two outcomes is the incremental effect, and it is the only figure that confirms whether spend caused growth.

This is the shift from correlation to causation. It rests on causal inference, the mathematical framework formalized in Judea Pearl's Causality: Models, Reasoning, and Inference, which provides the methods for estimating cause and effect from observational data rather than inferring it from correlation. Causal inference builds a reasoning step between the data and the decision: it models the counterfactual explicitly rather than assuming that whatever correlated with the outcome must have produced it.

Measured causally, the picture often inverts. Channels that looked efficient under ROAS turn out to have harvested demand that already existed, while channels that looked weak turn out to have created it. Marketing ROI becomes a statement about cause, which is the only version finance and marketing can both stand behind.

How a Large Causal Model Unifies Enterprise Marketing ROI

A Large Causal Model resolves the architecture problem by replacing many function-specific models with one. Instead of finance, marketing, and sales each running a separate correlation-based model, a single causal model holds the whole business and decomposes performance into the specific drivers that caused it. One model produces one answer, so the CFO's forecast and the CMO's plan are built on the same foundation.

POEM365 is the Large Causal Foundation Model we built at DATA POEM on the FOUNT architecture. Pre-trained on 250+ billion transactions and $5 trillion in spend data, it adapts to each enterprise's own data to produce a client-specific Enterprise Growth Model that reaches 90%+ forecast accuracy at go-live.

What that produces is a unified growth decomposition — causal incrementality measurement across every channel and function — rather than a stack of channel-by-channel reports.

The architectural payoff is that multicollinearity stops being fatal. We built FOUNT to estimate causal effects across correlated channels rather than fight the regression instability that breaks ordinary least squares. That is why a single model can do what 200+ specialized models could not in the M5 forecasting competition, the world's most rigorous time-series benchmark — see the FOUNT architecture for the full benchmark results.

Managed deployment runs 8–12 weeks to production.

Marketing ROI Metrics Worth Tracking

No single number captures marketing's contribution, but a focused set, read causally, beats a sprawling dashboard read correlationally. These are the metrics I'd tell any enterprise CMO to track:

• Incremental ROI: profit caused by the investment, net of what would have happened anyway. The headline measure; everything else is diagnostic.
• Causal contribution by channel: each channel's share of caused growth, not its share of attributed conversions.
• Counterfactual baseline: the modeled outcome with no marketing investment, against which incrementality is measured.
• Portfolio and halo effects: how spend on one product or channel lifts or cannibalizes the others, which channel-level metrics miss entirely.
• Customer lifetime value: the long-run value of acquired customers, so short-term efficiency metrics don't crowd out durable growth.

Read together and from one model, these tell a board what the spend caused. Read separately from competing models, they reproduce the contradiction the enterprise started with.

Frequently Asked Questions About Measuring Marketing ROI

What is a Good Marketing ROI?

A good marketing ROI is one measured causally and proven incremental, not a fixed ratio. Benchmarks like 5:1 are widely cited but mislead at enterprise scale, because they count attributed revenue that includes sales and marketing did not cause. The defensible target is positive incremental profit against a modeled counterfactual.

Why Does Marketing ROI Differ from the CFO's Revenue Numbers?

Marketing ROI differs from the CFO's revenue numbers because each is produced by a separate correlation-based model with its own assumptions: marketing credits touchpoints, finance models revenue. Neither isolates cause, so the two figures describe the same business and disagree. One unified causal model removes the discrepancy.

What is the Difference Between ROAS and Marketing ROI?

ROAS divides attributed revenue by ad spend and measures efficiency. Marketing ROI measures incremental profit caused by the spend, net of conversions that would have happened anyway. ROAS can look strong while incremental ROI is near zero, because high-intent audiences convert regardless of the advertising pointed at them.

How Long Does it Take to See Accurate Marketing ROI Measurement?

Accurate causal measurement depends on deployment, not on accumulating more dashboards. A managed POEM365 deployment reaches production in 6 weeks and delivers 90%+ forecast accuracy at go-live, because the model is pre-trained and then adapted to the enterprise's own data rather than built from scratch.

Move from Measurement to Decisions

Measuring marketing ROI accurately is the precondition for the decision every leader is actually trying to make: where to put the next dollar of growth investment. As long as finance and marketing argue from separate correlation-based models, that decision stays a negotiation between contradictory reports rather than a reading of one shared truth.

A causal model changes the terms. When the CFO's forecast and the CMO's plan come from the same decomposition of what drives the business, marketing ROI stops being a number to defend and turns into a basis for the next decision. That's what we built POEM365 to deliver.

See how DATA POEM turns fragmented measurement into one reliable causal answer across the enterprise.