About
Where all of this comes from.
The ad platform says it drove the sale. The warehouse says something else. Someone has to decide which of those two numbers goes into the report, and that call has been my job for most of the last decade.
I am Pavlo Kostenko, Chief Data Officer at A-Players in Limassol, Cyprus. The team is about eleven people: analysts, data engineers, data science, BI. I have been on the island long enough to have started learning Greek and nowhere near long enough to be any good at it.
Here is the part I will argue about. An attribution model is not a measurement, it is an accounting convention, and every convention flatters somebody. Last click flatters whoever turned up at the end. Platform-reported conversions flatter the platform, which is grading its own homework and has yet to give itself a poor mark. None of that is fraud, it is self-report, and we treat self-report as weak evidence everywhere else in life. The question that survives all of it is whether the spend caused the revenue or just stood near it when the revenue arrived.
In practice that means BigQuery and dbt, Python where the statistics get serious, SQL for everything else, AppsFlyer and Adjust on mobile, GA4 and the Facebook Conversions API on the web, Tableau and Looker at the end where people actually look. Fifteen years in, the industries have changed far more than the work has: gaming, gambling, health and fitness, mobile utilities, now web subscriptions around AI products. Same pipes. They leak in the same places.
I came into this from a university department: PhD in information technologies, seven years teaching, and one habit from those years I never managed to drop. A number without a method attached is a rumour. Write a paper and you have to state what you measured, over which window, and what you excluded, or it comes back from review. Marketing reporting quietly dropped that requirement somewhere. I keep putting it back, which does not always make me pleasant company in a quarterly review.
How this site works
Every figure is labelled: measured by us, claimed by a platform, or estimated. The notes in the margin say which window, which filter, and what the number does not cover. Pedantic, yes. It is also the whole difference between a finding and a slide.
The sections are named after warehouse layers because that is honestly how the writing sorts itself. Marts is finished work, Raw is what landed recently and has not been cleaned up, Storage is everything by year. Most of it is first-hand, and a decent share of it is me being wrong about something in public.
Contact
hi@pavlokostenko.com. Consulting questions and speaking invitations land there. So does the occasional message telling me a number on this site is wrong, which is the most useful mail I get.