When Attribution Replaces Accountability: The Problem No One Wants to Name

Twenty-two pages. Sans-serif typography, lavender gradients, slide numbers in every corner. The attribution dashboard propping up 380 million pesos of last quarter's investment has the format and authority of a court ruling. The only thing missing is a counterparty.

On the cover, the agency's logo. Inside, the shiny numbers that say what the marketing director needs to say in Monday's committee meeting. ROAS by channel, contribution by publisher, a proprietary calibrated multi-touch model, two years of data, adjusted for seasonality. Every figure precise to the second decimal.

An obvious question that's comfortable not to ask: who built this model? The same agency that manages 70% of the mix the model declares successful. A more uncomfortable question: when was this output last checked against a real incrementality test? Never. When the idea of a geo test came up, someone asked whether the agency would agree to run the experiment on the channels it manages itself, and the conversation ended there.

"Well, the dashboard already shows they're working." That phrase, said without irony in a meeting room in Las Condes a few weeks ago, is the problem no one wants to name. Platform attribution and agency attribution have become, across much of the industry, a way of avoiding accountability. An alibi with clean typography.

Three Reasons, None the Product's Fault

Why does an attribution model end up functioning as a political crutch instead of a measurement tool? There are three structural reasons worth naming directly.

1. The model is built, funded, or maintained by whoever benefits from its output.

This is the most uncomfortable piece. A considerable share of the attribution models circulating in the region were built by the same agency that executes the investment, by the same platform that sells the inventory, or by a consultant who has a partnership agreement with one of the two. There doesn't need to be bad intent. But the direction of the incentives is clear, and any statistician will tell you that when the person building the model is also the one who comes out favored in its results, the assumptions quietly drift toward the expected conclusion.

As a recent piece of research on the topic notes, "when attribution replaces accountability, the model stops being a testable hypothesis and becomes a prefabricated argument."

2. Attribution is retrospective, not predictive, and that protects whoever uses it.

An attribution dashboard tells you what happened. It ranks channels according to the logic of whatever model was chosen — last-click, first-click, time-decay, platform data-driven, whatever. What it doesn't tell you is what would have happened with a different mix, or what will happen next quarter if you shift the budget.

For someone who has to defend a number in a meeting, that asymmetry is a blessing. You can justify what you already spent without committing to a prediction that could fail. Attribution as a backward-looking audit tool is comfortable, because you always have data to support it. Attribution as a forward-looking decision tool would require taking a stance, and taking a stance is exactly what organizations avoid when there's no incentive to do so.

3. It's easier to defend a number than to explain an assumption.

A well-built MMM comes with confidence intervals, explicitly stated assumptions, cross-calibration against incrementality tests, and the honesty to say "this channel contributed between X% and Y% with this probability, conditional on these assumptions." An attribution dashboard comes with a fixed number: ROAS 4.2; channel contribution, 38%.

In an organization where the political cost of defending a range is higher than the political cost of defending a precise but false number, the dashboard always wins. Not because it's better, but because it's more comfortable. Spurious precision is easier to present than honest uncertainty, and budget committees tend to reward the former.

The LATAM Angle: The Same Problem With Two Extra Twists

In Chile, Mexico, Colombia, and Peru, this problem has additional layers worth highlighting.

First, agency market concentration. In most markets in the region, a handful of holding companies control the bulk of large accounts. That means the attribution tool, the planning team, the buying operation, and the measurement model often all live inside the same vendor. The separation between "who decides the mix" and "who measures the mix" disappears when both are the same invoice.

Second, measurement budgets. A brand investing 5 million dollars a year rarely has the extra budget to hire an independent audit. The result is that the brand accepts its agency's model not because it's optimal, but because it's the only accessible option.

Third, CMO turnover. With an average tenure of 18 to 24 months at consumer brands in LATAM, a CMO has no incentive to invest in measurement systems that take a year to show real value. The agency's inherited system is the low political-risk option.

What Makes Marketing Mix Modeling (MMM) Different

MMM doesn't magically solve this problem, but it approaches it from a different structural angle:

Real Independence: It separates the actor who builds the model from the one who executes the investment. An independent MMM, like the ones NeuroRadar implements on frameworks such as Robyn, has no incentive to favor a specific channel.
Holistic View: It models the full mix, including TV, radio, OOH, and organic — channels that platform dashboards ignore entirely.
Exposed Assumptions: It forces you to declare the shape of the adstock curve and the saturation ranges. When you defend the results, you also defend the assumptions.
Validation Against Reality: Accuracy is evaluated against real incrementality tests (geo-experiments, holdouts), not just clicks on a platform.

What MMM Does NOT Do

For the sake of intellectual honesty, it's vital to know what it doesn't solve:

It doesn't operate in real time: Serious models need historical data and are updated quarterly. It doesn't replace daily tactical optimization.
It doesn't pick the creative: It tells you which channel contributes more, not whether the 15-second video beats the carousel.
It's not free of assumptions: It's free of hidden assumptions. It makes the variables visible so human judgment can act on better data.

What to Do Now

If you suspect attribution has become an alibi in your organization, ask these three questions at your next meeting:

Who built this model, and who actually benefits from its result?
When was the last time we checked this data against a real incrementality test?
What budget decision would we have made differently if the model's result had been different?

If the answers are uncomfortable, it's time to separate the audit from the operation. You can see what an independent MMM dashboard looks like with real data in our open demo.

Frequently Asked Questions

Why can agency attribution be biased?
When the party building the model is also the party managing the investment, incentives are aligned to favor the channels that same actor controls. This turns measurement into a political alibi instead of an optimization tool.

What sets independent MMM apart from traditional attribution?
Independent MMM separates measurement from execution, uses aggregated data (no cookies), transparently exposes its assumptions, and is validated through real incrementality tests (geo-tests or holdouts).

Does MMM completely replace platform dashboards?
No. Platform metrics remain useful for daily tactical optimization. MMM is used for strategic budget planning and auditing the real profitability of each channel at the macro level.

How long does it take to see results with an MMM model?
For the model to be statistically sound, 18 to 24 months of historical data is recommended. Models are typically updated quarterly to adjust long-term investment strategy.