Context · Business model

B2C ecommerce

Short consideration cycles, high volume, purchase as the outcome. The easiest place to measure and the easiest place to fool yourself.

What changes here

The same four questions, and what each one runs into in this context.

Short windows make last-touch look accurate. Branded search and retargeting absorb credit for demand that already existed.

Geo holdouts are fast and cheap here. Watch for brand search contamination across market borders.

Marginal returns bend quickly at scale, and the promotional calendar confounds almost every naive comparison.

Repeat purchase drives value, so optimising to first purchase understates what a customer is worth.

Attribution Is Not Incrementality

Attribution assigns credit for conversions you observed. Incrementality estimates the ones that would not have happened otherwise. The gap between them is where budgets go wrong.
Planned

Reading Platform Discrepancies Without Losing Your Mind

Two systems report different numbers for the same campaign. A method for finding which of the four usual causes is responsible.
Planned

The Same Click, Different Funnels

One ad click means something different in a self-serve checkout than in a sales-qualified pipeline. Why the funnel grain you choose determines what you can measure.
Planned

What Tracking Loss Actually Costs You

Consent, identity resolution, and cross-device loss are not one problem. Separating random loss from systematic loss, and why only one of them threatens validity.

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.
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.
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.

Planned

Calibrating an MMM to Experiments

The practical bridge between slow causal evidence and fast allocation models: using experimental estimates as priors and as validation.
Planned

Marginal Return, Not Average Return

The channel with the best historical ROI is frequently the one closest to saturation. Why budget decisions need the slope, not the level.
Planned

Paid Search and SEO Are Not Independent

Branded paid search often captures demand that organic would have captured anyway. Designing the test that measures total branded conversions rather than paid ones.
Worked example Planned

An Experiment-Calibrated Marketing Mix Model

Baseline, adstock, saturation, and controls, validated against holdout estimates and used to produce a constrained budget allocation with stated uncertainty.