Skip to content
Was It Causal?

Incrementality · Discipline 2 of 4

Did marketing change the outcome?

What would have happened without the intervention? Counterfactuals, experiments, quasi-experiments, and the diagnostics that decide whether an estimate deserves to be believed. Causal ML sits here too, where the question moves from an average effect to whose effect differs.

Where this sits

Each rung depends on the one below it. This discipline answers the highlighted one.

  1. 01

    Observe

    What happened?

  2. 02

    Describe

    Where did conversions appear to come from?

  3. 03

    Estimate

    What changed because of marketing?

  4. 04

    Explain

    How did channels and outside factors contribute?

  5. 05

    Decide

    What should happen next?

  6. 06

    Learn

    What uncertainty should we reduce next?

Worked examples

Full analyses in this discipline, carried from the business decision through to a recommendation and an explicit account of what the result does not establish.

Applied Worked example

Designing a Credible Geo Holdout

A paid social incrementality test carried end to end: market pairing, power, pre-trend diagnostics, a difference-in-differences estimate with its interval, incremental CAC, and what the design cannot settle.

Simulated data Read  →

Planned

The shape of this discipline, published as a roadmap. These are titles and scope, not finished work. They are here so you can see where this is going, not to suggest it has arrived.

Foundation Planned

Counterfactuals, Estimands, and Saying What You Mean

Before choosing a method, state precisely what you are trying to estimate: which units, which treatment, which outcome, over what horizon.

Not yet written
Applied Planned

Choosing Between User, Geo, and Cluster Holdouts

The design decision is about what you can withhold cleanly, not about statistical efficiency. A decision procedure for picking the unit of randomization.

Not yet written
Technical Planned

When Randomization Is Impossible

Difference-in-differences, synthetic control, and interrupted time series: what each one assumes, and the diagnostic that would falsify it.

Not yet written
Applied Planned

The Diagnostics That Earn Belief

Balance, overlap, pre-trends, placebo tests, and sensitivity analysis. Which to run, in what order, and what each one can and cannot rule out.

Not yet written
Technical Planned

When the Experiment Leaks

Noncompliance, contamination, and spillovers. Why intent-to-treat is usually the right estimand even though it answers a slightly different question.

Not yet written
Applied Planned

Triangulating When Your Evidence Disagrees

An experiment, an MMM, and attribution give three different answers. A structured way to reconcile them without averaging away the information.

Not yet written