What should attribution methodology documentation contain?

The input precisely, the comparison the estimate is made against, where the uncertainty comes from, and the method's stated limits. The last is what separates documentation from marketing, because every method has a boundary.

Why does plain English matter for a technical method?

Because the reader who most needs to understand it is the marketing manager answering a CFO or the founder allocating budget, not a statistician. A document written for peer review cannot be used in the meeting where the number is challenged.

What are the limits of an observational causal read?

It is not experimental, it cannot see creative fatigue or auction dynamics or anything outside channel-level data, and it cannot tell you what an individual customer did.

Plain-English methodology is part of the product

A number you cannot interrogate is a number you cannot defend in a budget meeting. The documentation is a feature, not a courtesy.

Three platforms, three answers, one budget to defend. Your GA4 export already holds the answer, and the first finding is free and never leaves your browser. Get your first finding free

By Joris van Huët, Founder & CEOUpdated 2026-10-063 min read

A number you cannot interrogate is a number you cannot defend, which makes the documentation part of the product rather than a courtesy attached to it. The test is whether a non-specialist can read it and come away able to argue with the output.

What the documentation has to contain

Element	Why it is load-bearing
The input, precisely	You need to know what the estimate is built from
The comparison being made	Without it, "lift" is a word rather than a measurement
Where uncertainty comes from	It explains why intervals widen and narrow
The stated limits	What the method cannot do, named by the vendor

The fourth is the one that separates documentation from marketing. Every method has a boundary. A document that names its own is written by people who understand it; a document that implies none is written by people selling it.

Why plain English specifically

Because the reader who most needs to understand the method is not a statistician. It is the marketing manager who has to answer a CFO, or the founder deciding where the next euro goes. A document written for peer review is a document that cannot be used in the meeting where the number gets challenged.

Plain English is not simplification. It is choosing the audience that has to act.

The three questions the document should let you answer unaided

What is this number compared against. How sure is the method, and what would make it surer. What is not measured, and why not.

If reading a vendor's documentation does not equip you for those three, the documentation is describing the product rather than explaining the method. The buyer-side framing is in why vendors should explain methodology openly.

What we publish

The input is a Google Analytics CSV export you produce.

First read €99, once, refundable within 30 days, no questions asked.

The limits we state

The read is observational, not experimental. It cannot see creative fatigue, auction dynamics or anything else outside channel-level data.

Naming those costs nothing and is the reason the rest is credible.

The one-line standard

If you cannot explain the method to a colleague in two sentences, you cannot defend the number in a meeting, and no amount of interface polish fixes that.

Frequently asked questions

What should attribution methodology documentation contain?
The input precisely, the comparison the estimate is made against, where the uncertainty comes from, and the method's stated limits. The last is what separates documentation from marketing, because every method has a boundary.
Why does plain English matter for a technical method?
Because the reader who most needs to understand it is the marketing manager answering a CFO or the founder allocating budget, not a statistician. A document written for peer review cannot be used in the meeting where the number is challenged.
What are the limits of an observational causal read?
It is not experimental, it cannot see creative fatigue or auction dynamics or anything outside channel-level data, and it cannot tell you what an individual customer did.

Go deeper: Incrementality testing, explained.

Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.

Have an audience of ecommerce operators? Earn 20% recurring commission by recommending the €99 read.

Keep reading

Terms in this article

Browse the full glossary