Causal inference. What it is and why we need to know it.

Geo Ceccarelli
Sep 17
3 min read

For years, marketing lived with a near-perfect alibi: correlation. "The user saw the ad, then bought" was enough evidence to justify budgets and strategies. But a good detective knows that a coincidence is not proof.

Today, with the disappearance of cookies, that alibi is crumbling and marketing has been called to trial. In my latest investigation for Cherry Picking , I examined the new advancement that is emerging: Causal Inference.

The Unreliable Alibi: Why the "Last Click" Is an Unreliable Witness

The problem with traditional attribution models is that they don't distinguish causation from coincidence. They record a sequence of events, but they can't answer the fundamental question: would the sale have happened anyway, even without our intervention? Relying on this data is like building a case based on an unreliable witness. It's time to look for concrete evidence.

Causal Inference: Measuring Real Impact

Causal inference abandons conjecture and adopts a scientific method. Just like in a medical trial, it compares a "test group" (exposed to a campaign) with a "control group" who is not shown the campaign. The difference in their behavior reveals the incremental impact: the value generated directly and uniquely by our action.

With advanced techniques like Uplift Modeling, we can go a step further: not only do we measure impact, but we identify "persuadable" customers—those who act simply because they're influenced by our message—allowing us to focus investments where they really matter.

The Verdict: Who's Already Using This Method?

Adoption isn't uniform. A vanguard of tech giants— Netflix, Amazon, Booking.com —have been using it for years, using causal inference for years. Why?

They have two things others don't: a huge amount of first-party data and a top-notch team of in-house data scientists. For them, measuring the real impact of a new feature, campaign, or discount isn't a luxury; it's the way they operate.

There are early adopters, advanced brands like Zalando and Novartis, who are implementing it to gain a competitive advantage.

And then there are the most advanced media centers and agencies, which are building teams and skills to offer this service, acting as the market's new "forensic experts."

The rest of the industry is in a phase of awareness. They know they need to change, but they don't yet have the skills.

The 3 Rules of the New Investigation:

Correlation is Not Causation: The first step is to accept that old models are dead. The goal is to measure impact, not just count clicks.
Measuring is Experimenting: Without a control group, there is no proof, only opinion. Testing culture becomes central.
The Advantage is in the Skills: The difference will not be made by the tools, but by the analysts capable of asking the right questions and interpreting the evidence.

Conclusion: The Question That Matters

Causal inference forces marketing to stop justifying costs and start proving value.

The question we must finally answer is no longer "what did we do with the money?", but "what did we make happen, that otherwise wouldn't have happened?" .

It's the only question that matters.

To follow the entire investigation and delve deeper into the case, listen to the Cherry Picking episode