Attribution Is Broken. Here Is What We Use Instead.

Introduction

Privacy changes, blocked cookies, and buyers who research on five devices have made single-model attribution a comfortable fiction. You still have to decide where next quarter’s budget goes, so the answer is not a better model — it is several imperfect ones that disagree usefully.

1. Accept That No Model Sees Everything

Last-click overpays for capture. First-click overpays for awareness. Every model is a lens with a known bias. Knowing each one’s bias is more useful than believing any single number.

2. Run Holdouts On Your Biggest Channel

Switch a channel off in a region, or for a fortnight, and measure what happens to total revenue. It is the closest thing to truth available, and it costs less than a quarter of misallocated spend.

3. Ask People Where They Heard About You

A single open field on the signup form — “how did you hear about us?” — routinely surfaces channels no dashboard credits. It is qualitative, it is imperfect, and it catches podcasts, referrals, and communities that tracking never will.

4. Judge Channels On A Portfolio, Not In Isolation

Some channels create demand and some capture it. Grading a demand-creating channel on last-click conversions guarantees you will cut the thing feeding the rest of the funnel.

Final Thought

The goal is not a perfect number. It is being roughly right in the same direction across several independent methods.