Which data source deserves the most trust?
Usually, the revenue attributed to your marketing campaigns on Meta Ads, Google Ads and in your CRM, added up, bears little resemblance to the data in your web analytics tool or the reports from your e-commerce platform.
No wonder: none of these tools uses the same attribution model.
Beyond that issue, you have every right to question the very principle of marketing attribution.
The higher the average order value and the longer the purchase cycle, the more this effect is structurally amplified.
Attribution Is a Myth (and Psychology Confirms It)
Before we return to tools and models, let's take a detour through cognitive psychology.
There is a well-documented bias called the "fundamental attribution error."
Social psychologists Lee Ross and Edward E. Jones identified and studied this bias in depth in the 1970s.
In essence: when we try to explain an event, we naturally overestimate the weight of internal factors (individual actions) and underestimate the weight of external factors (context, environment).
Applied to marketing, this bias is everywhere.
When we try to explain an event, we naturally overestimate the weight of internal factors (individual actions) and underestimate the weight of external factors (context, environment)
A customer buys a €2,000 necklace on your website.
Google Ads claims a share of that conversion because she clicked on a brand protection ad 10 days earlier.
Meta Ads claims 100% of the same conversion because she saw an ad on Instagram the day of the sale.
But what no one measures is that she actually discovered the brand in a store three months ago.
A friend told her about it on WhatsApp, she read an article in Vogue, she has followed the brand's Instagram account for weeks, she talked through the price with her partner...
That whole journey (the context) is invisible.
The platforms credit the sale to their own last touchpoint (the internal factor). It's a perfect example of the fundamental attribution error applied at the scale of an advertising ecosystem.
When the Algorithm Optimizes Against Incrementality
Ad algorithms optimize the delivery of conversion campaigns to maximize volume or value.
So far, nothing worrying. Quite the opposite.
In practice, your campaigns focus on users who are actively searching for your products or similar ones.
The problem is that these in-market audiences represent only a tiny share of your total addressable market.
Beyond this significant limitation, even when your customer databases and engaged visitors are rigorously excluded from your ad targeting, doubt remains.
To what extent did simply seeing or clicking an ad really drive the purchase?
The platforms spend your budget to reach users who, in many cases, would have bought without seeing the ad their purchase was "attributed" to.
To address this, Meta Ads has rolled out an incremental attribution model that remains a black box, just as opaque as Google Ads' data-driven attribution models.
The reason is simple (and nobody really wants to admit it): we can't establish with certainty a causal link between an ad interaction and a purchase.
In the Maze of the "Messy Middle"
Attribution biases matter even more when average order values are high.
Impulse purchases are rarer and purchase cycles are longer.
The trigger is emotional, but it happens within a climate of trust that built up over time or under the influence of personal or authority figures.
To understand it better, Google's research on the Messy Middle offers a useful framework.
The purchase journey has never been a funnel. It's a complex loop in which consumers move between two states:
- Exploration : They discover collections, read the press, soak up the brand world.
- Evaluation: They compare, talk with friends and family, visit a store, check prices online.
In luxury and premium brands, a customer can move between exploration and evaluation for weeks, even months.
Reducing this cognitive maze to the last click of a retargeting campaign is not only simplistic, but potentially dangerous when it drives strategic decisions.
The Mirage of Geo-Holdout Tests
As classic attribution has become obsolete, the industry has turned to incrementality measurement, often pitched as the miracle fix through geo tests (geo-holdout tests).
On paper, the idea is appealing: turn off ads in region A, keep them running in region B, and compare.
In reality, this method runs into several limitations:
Lack of volume
Reaching statistical significance to validate a geo test takes time and budget that few brands can afford to freeze.
The illusion of "twin" regions
A test is only methodologically valid if it compares two areas with strictly identical buying behavior.
Yet finding a control region that is a perfect socio-economic mirror of your main market (to compare Paris with another French city, for example) is often a tall order.
A poor fit with the long term
A standard incrementality test rarely runs longer than 4 to 6 weeks, mainly to limit the opportunity cost of pausing media.
That's far too short to capture a premium purchase cycle that can stretch over several months.
The test ends well before it can measure the true long-term impact, which inevitably leads to underestimating it.
The "spillover effect"
Customers of premium or luxury brands are often on the move.
An ad might be seen on a phone in London, a site browsed through a VPN from a hotel in Milan, and a purchase completed in a Paris store.
Geo tests struggle badly to account for this reality.
Zero-Party Data, Aptly Named
Another way to get a clearer view of your marketing attribution is to survey customers directly to understand what really drove their purchase.
At first glance, this self-reported attribution looks like a concrete, solid alternative to the black boxes of platform data-driven attribution.
Unfortunately, this alternative has its limits too:
Statistical significance
Are the number and variety of respondents representative enough of all your customers?
The fundamental attribution error
The very one we devoted a section to, and one that undermines the reliability of survey answers.
So statistical significance hardly matters, because respondents overestimate the weight of their own actions compared with the external factors that really influenced their decision.
What If Triangulation Is the Answer?
For several years now, new approaches have emerged to try to answer the questions most C-suite members ask.
Among the most recent, often billed as the future of measurement: "Triangulation," or the "Suite of Truth" (you don't say).
The idea is to layer and reconcile three complex approaches:
Marketing Mix Modeling (MMM) for the big picture
MMM analyzes large budget and historical datasets to understand the overall impact of your investments on sales (an analysis that goes from the general to the specific).
Multi-Touch Attribution (MTA) for the individual journey
It tries to reconstruct each buyer's exact path, interaction by interaction, even though the gradual disappearance of tracking technologies (the infamous cookies) makes this exercise more and more illusory.
Conversion Lift Studies (or incrementality tests) to measure real gains
These real-world experiments act as the final arbiter, isolating what a campaign truly delivered and correcting the interpretation errors of the first two models.
The problem is that stacking three complex statistical models, all structurally flawed, doesn't produce an absolute source of truth. More often it produces an expensive, hard-to-read tangle.
However sophisticated, these triangulation models still feed on truncated source data.
The equation is unforgiving: applying cutting-edge math to incomplete data will always give an uncertain result.
By nature, these algorithms stay blind to factors that are decisive in purchase journeys: peer recommendations, the influence of authority figures, private conversations, and the intimate, unfathomable and often irrational emotions that shape the desirability of a product or brand.
The Farmer and the Financier
Perfect marketing attribution is a pipe dream.
The problem is dizzyingly complex, and the solution may lie in accepting a degree of uncertainty.
Rather than wearing yourself out searching for absolute truth in ever more opaque statistical models, a lasting approach would combine two complementary mindsets: the pragmatism of a farmer's common sense and the rigor of a financial view.
A farmer's common sense in marketing is intuition built on intimate knowledge of your market and your product.
It means accepting that a key share of marketing decisions must rest on qualitative, empirical convictions. It means understanding your audience's codes, knowing where to capture its attention, organically or through paid media, and admitting that desirability and emotion can't be put into an equation.
The impact of this brand-building work will always partly escape traditional tracking tools.
That is precisely why the financial approach steps in as the essential safeguard for that intuition.
Forget vanity engagement metrics and the biased ROAS from ad platforms, and focus on what matters: the P&L.
The question is no longer exactly how many sales a given channel claims, but what the macroeconomic impact of your trade-offs is.
- What actually happens when you significantly change your ad spend or your media strategy?
- Does incremental revenue from new customers grow?
- Does overall contribution margin follow?
In high-value sectors, marketing isn't steered down to the pixel but by balance.
Navigating calmly under this thick cloud of data takes a fine understanding of your market combined with the cold reality of financial results.
Absolute truth isn't in the sum of your ad dashboards. You see it in the healthy, lasting growth of your profitability.
